Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * postgres_fdw.c
4 : * Foreign-data wrapper for remote PostgreSQL servers
5 : *
6 : * Portions Copyright (c) 2012-2025, PostgreSQL Global Development Group
7 : *
8 : * IDENTIFICATION
9 : * contrib/postgres_fdw/postgres_fdw.c
10 : *
11 : *-------------------------------------------------------------------------
12 : */
13 : #include "postgres.h"
14 :
15 : #include <limits.h>
16 :
17 : #include "access/htup_details.h"
18 : #include "access/sysattr.h"
19 : #include "access/table.h"
20 : #include "catalog/pg_opfamily.h"
21 : #include "commands/defrem.h"
22 : #include "commands/explain_format.h"
23 : #include "commands/explain_state.h"
24 : #include "executor/execAsync.h"
25 : #include "foreign/fdwapi.h"
26 : #include "funcapi.h"
27 : #include "miscadmin.h"
28 : #include "nodes/makefuncs.h"
29 : #include "nodes/nodeFuncs.h"
30 : #include "optimizer/appendinfo.h"
31 : #include "optimizer/cost.h"
32 : #include "optimizer/inherit.h"
33 : #include "optimizer/optimizer.h"
34 : #include "optimizer/pathnode.h"
35 : #include "optimizer/paths.h"
36 : #include "optimizer/planmain.h"
37 : #include "optimizer/prep.h"
38 : #include "optimizer/restrictinfo.h"
39 : #include "optimizer/tlist.h"
40 : #include "parser/parsetree.h"
41 : #include "postgres_fdw.h"
42 : #include "storage/latch.h"
43 : #include "utils/builtins.h"
44 : #include "utils/float.h"
45 : #include "utils/guc.h"
46 : #include "utils/lsyscache.h"
47 : #include "utils/memutils.h"
48 : #include "utils/rel.h"
49 : #include "utils/sampling.h"
50 : #include "utils/selfuncs.h"
51 :
52 30 : PG_MODULE_MAGIC_EXT(
53 : .name = "postgres_fdw",
54 : .version = PG_VERSION
55 : );
56 :
57 : /* Default CPU cost to start up a foreign query. */
58 : #define DEFAULT_FDW_STARTUP_COST 100.0
59 :
60 : /* Default CPU cost to process 1 row (above and beyond cpu_tuple_cost). */
61 : #define DEFAULT_FDW_TUPLE_COST 0.2
62 :
63 : /* If no remote estimates, assume a sort costs 20% extra */
64 : #define DEFAULT_FDW_SORT_MULTIPLIER 1.2
65 :
66 : /*
67 : * Indexes of FDW-private information stored in fdw_private lists.
68 : *
69 : * These items are indexed with the enum FdwScanPrivateIndex, so an item
70 : * can be fetched with list_nth(). For example, to get the SELECT statement:
71 : * sql = strVal(list_nth(fdw_private, FdwScanPrivateSelectSql));
72 : */
73 : enum FdwScanPrivateIndex
74 : {
75 : /* SQL statement to execute remotely (as a String node) */
76 : FdwScanPrivateSelectSql,
77 : /* Integer list of attribute numbers retrieved by the SELECT */
78 : FdwScanPrivateRetrievedAttrs,
79 : /* Integer representing the desired fetch_size */
80 : FdwScanPrivateFetchSize,
81 :
82 : /*
83 : * String describing join i.e. names of relations being joined and types
84 : * of join, added when the scan is join
85 : */
86 : FdwScanPrivateRelations,
87 : };
88 :
89 : /*
90 : * Similarly, this enum describes what's kept in the fdw_private list for
91 : * a ModifyTable node referencing a postgres_fdw foreign table. We store:
92 : *
93 : * 1) INSERT/UPDATE/DELETE statement text to be sent to the remote server
94 : * 2) Integer list of target attribute numbers for INSERT/UPDATE
95 : * (NIL for a DELETE)
96 : * 3) Length till the end of VALUES clause for INSERT
97 : * (-1 for a DELETE/UPDATE)
98 : * 4) Boolean flag showing if the remote query has a RETURNING clause
99 : * 5) Integer list of attribute numbers retrieved by RETURNING, if any
100 : */
101 : enum FdwModifyPrivateIndex
102 : {
103 : /* SQL statement to execute remotely (as a String node) */
104 : FdwModifyPrivateUpdateSql,
105 : /* Integer list of target attribute numbers for INSERT/UPDATE */
106 : FdwModifyPrivateTargetAttnums,
107 : /* Length till the end of VALUES clause (as an Integer node) */
108 : FdwModifyPrivateLen,
109 : /* has-returning flag (as a Boolean node) */
110 : FdwModifyPrivateHasReturning,
111 : /* Integer list of attribute numbers retrieved by RETURNING */
112 : FdwModifyPrivateRetrievedAttrs,
113 : };
114 :
115 : /*
116 : * Similarly, this enum describes what's kept in the fdw_private list for
117 : * a ForeignScan node that modifies a foreign table directly. We store:
118 : *
119 : * 1) UPDATE/DELETE statement text to be sent to the remote server
120 : * 2) Boolean flag showing if the remote query has a RETURNING clause
121 : * 3) Integer list of attribute numbers retrieved by RETURNING, if any
122 : * 4) Boolean flag showing if we set the command es_processed
123 : */
124 : enum FdwDirectModifyPrivateIndex
125 : {
126 : /* SQL statement to execute remotely (as a String node) */
127 : FdwDirectModifyPrivateUpdateSql,
128 : /* has-returning flag (as a Boolean node) */
129 : FdwDirectModifyPrivateHasReturning,
130 : /* Integer list of attribute numbers retrieved by RETURNING */
131 : FdwDirectModifyPrivateRetrievedAttrs,
132 : /* set-processed flag (as a Boolean node) */
133 : FdwDirectModifyPrivateSetProcessed,
134 : };
135 :
136 : /*
137 : * Execution state of a foreign scan using postgres_fdw.
138 : */
139 : typedef struct PgFdwScanState
140 : {
141 : Relation rel; /* relcache entry for the foreign table. NULL
142 : * for a foreign join scan. */
143 : TupleDesc tupdesc; /* tuple descriptor of scan */
144 : AttInMetadata *attinmeta; /* attribute datatype conversion metadata */
145 :
146 : /* extracted fdw_private data */
147 : char *query; /* text of SELECT command */
148 : List *retrieved_attrs; /* list of retrieved attribute numbers */
149 :
150 : /* for remote query execution */
151 : PGconn *conn; /* connection for the scan */
152 : PgFdwConnState *conn_state; /* extra per-connection state */
153 : unsigned int cursor_number; /* quasi-unique ID for my cursor */
154 : bool cursor_exists; /* have we created the cursor? */
155 : int numParams; /* number of parameters passed to query */
156 : FmgrInfo *param_flinfo; /* output conversion functions for them */
157 : List *param_exprs; /* executable expressions for param values */
158 : const char **param_values; /* textual values of query parameters */
159 :
160 : /* for storing result tuples */
161 : HeapTuple *tuples; /* array of currently-retrieved tuples */
162 : int num_tuples; /* # of tuples in array */
163 : int next_tuple; /* index of next one to return */
164 :
165 : /* batch-level state, for optimizing rewinds and avoiding useless fetch */
166 : int fetch_ct_2; /* Min(# of fetches done, 2) */
167 : bool eof_reached; /* true if last fetch reached EOF */
168 :
169 : /* for asynchronous execution */
170 : bool async_capable; /* engage asynchronous-capable logic? */
171 :
172 : /* working memory contexts */
173 : MemoryContext batch_cxt; /* context holding current batch of tuples */
174 : MemoryContext temp_cxt; /* context for per-tuple temporary data */
175 :
176 : int fetch_size; /* number of tuples per fetch */
177 : } PgFdwScanState;
178 :
179 : /*
180 : * Execution state of a foreign insert/update/delete operation.
181 : */
182 : typedef struct PgFdwModifyState
183 : {
184 : Relation rel; /* relcache entry for the foreign table */
185 : AttInMetadata *attinmeta; /* attribute datatype conversion metadata */
186 :
187 : /* for remote query execution */
188 : PGconn *conn; /* connection for the scan */
189 : PgFdwConnState *conn_state; /* extra per-connection state */
190 : char *p_name; /* name of prepared statement, if created */
191 :
192 : /* extracted fdw_private data */
193 : char *query; /* text of INSERT/UPDATE/DELETE command */
194 : char *orig_query; /* original text of INSERT command */
195 : List *target_attrs; /* list of target attribute numbers */
196 : int values_end; /* length up to the end of VALUES */
197 : int batch_size; /* value of FDW option "batch_size" */
198 : bool has_returning; /* is there a RETURNING clause? */
199 : List *retrieved_attrs; /* attr numbers retrieved by RETURNING */
200 :
201 : /* info about parameters for prepared statement */
202 : AttrNumber ctidAttno; /* attnum of input resjunk ctid column */
203 : int p_nums; /* number of parameters to transmit */
204 : FmgrInfo *p_flinfo; /* output conversion functions for them */
205 :
206 : /* batch operation stuff */
207 : int num_slots; /* number of slots to insert */
208 :
209 : /* working memory context */
210 : MemoryContext temp_cxt; /* context for per-tuple temporary data */
211 :
212 : /* for update row movement if subplan result rel */
213 : struct PgFdwModifyState *aux_fmstate; /* foreign-insert state, if
214 : * created */
215 : } PgFdwModifyState;
216 :
217 : /*
218 : * Execution state of a foreign scan that modifies a foreign table directly.
219 : */
220 : typedef struct PgFdwDirectModifyState
221 : {
222 : Relation rel; /* relcache entry for the foreign table */
223 : AttInMetadata *attinmeta; /* attribute datatype conversion metadata */
224 :
225 : /* extracted fdw_private data */
226 : char *query; /* text of UPDATE/DELETE command */
227 : bool has_returning; /* is there a RETURNING clause? */
228 : List *retrieved_attrs; /* attr numbers retrieved by RETURNING */
229 : bool set_processed; /* do we set the command es_processed? */
230 :
231 : /* for remote query execution */
232 : PGconn *conn; /* connection for the update */
233 : PgFdwConnState *conn_state; /* extra per-connection state */
234 : int numParams; /* number of parameters passed to query */
235 : FmgrInfo *param_flinfo; /* output conversion functions for them */
236 : List *param_exprs; /* executable expressions for param values */
237 : const char **param_values; /* textual values of query parameters */
238 :
239 : /* for storing result tuples */
240 : PGresult *result; /* result for query */
241 : int num_tuples; /* # of result tuples */
242 : int next_tuple; /* index of next one to return */
243 : Relation resultRel; /* relcache entry for the target relation */
244 : AttrNumber *attnoMap; /* array of attnums of input user columns */
245 : AttrNumber ctidAttno; /* attnum of input ctid column */
246 : AttrNumber oidAttno; /* attnum of input oid column */
247 : bool hasSystemCols; /* are there system columns of resultRel? */
248 :
249 : /* working memory context */
250 : MemoryContext temp_cxt; /* context for per-tuple temporary data */
251 : } PgFdwDirectModifyState;
252 :
253 : /*
254 : * Workspace for analyzing a foreign table.
255 : */
256 : typedef struct PgFdwAnalyzeState
257 : {
258 : Relation rel; /* relcache entry for the foreign table */
259 : AttInMetadata *attinmeta; /* attribute datatype conversion metadata */
260 : List *retrieved_attrs; /* attr numbers retrieved by query */
261 :
262 : /* collected sample rows */
263 : HeapTuple *rows; /* array of size targrows */
264 : int targrows; /* target # of sample rows */
265 : int numrows; /* # of sample rows collected */
266 :
267 : /* for random sampling */
268 : double samplerows; /* # of rows fetched */
269 : double rowstoskip; /* # of rows to skip before next sample */
270 : ReservoirStateData rstate; /* state for reservoir sampling */
271 :
272 : /* working memory contexts */
273 : MemoryContext anl_cxt; /* context for per-analyze lifespan data */
274 : MemoryContext temp_cxt; /* context for per-tuple temporary data */
275 : } PgFdwAnalyzeState;
276 :
277 : /*
278 : * This enum describes what's kept in the fdw_private list for a ForeignPath.
279 : * We store:
280 : *
281 : * 1) Boolean flag showing if the remote query has the final sort
282 : * 2) Boolean flag showing if the remote query has the LIMIT clause
283 : */
284 : enum FdwPathPrivateIndex
285 : {
286 : /* has-final-sort flag (as a Boolean node) */
287 : FdwPathPrivateHasFinalSort,
288 : /* has-limit flag (as a Boolean node) */
289 : FdwPathPrivateHasLimit,
290 : };
291 :
292 : /* Struct for extra information passed to estimate_path_cost_size() */
293 : typedef struct
294 : {
295 : PathTarget *target;
296 : bool has_final_sort;
297 : bool has_limit;
298 : double limit_tuples;
299 : int64 count_est;
300 : int64 offset_est;
301 : } PgFdwPathExtraData;
302 :
303 : /*
304 : * Identify the attribute where data conversion fails.
305 : */
306 : typedef struct ConversionLocation
307 : {
308 : AttrNumber cur_attno; /* attribute number being processed, or 0 */
309 : Relation rel; /* foreign table being processed, or NULL */
310 : ForeignScanState *fsstate; /* plan node being processed, or NULL */
311 : } ConversionLocation;
312 :
313 : /* Callback argument for ec_member_matches_foreign */
314 : typedef struct
315 : {
316 : Expr *current; /* current expr, or NULL if not yet found */
317 : List *already_used; /* expressions already dealt with */
318 : } ec_member_foreign_arg;
319 :
320 : /*
321 : * SQL functions
322 : */
323 24 : PG_FUNCTION_INFO_V1(postgres_fdw_handler);
324 :
325 : /*
326 : * FDW callback routines
327 : */
328 : static void postgresGetForeignRelSize(PlannerInfo *root,
329 : RelOptInfo *baserel,
330 : Oid foreigntableid);
331 : static void postgresGetForeignPaths(PlannerInfo *root,
332 : RelOptInfo *baserel,
333 : Oid foreigntableid);
334 : static ForeignScan *postgresGetForeignPlan(PlannerInfo *root,
335 : RelOptInfo *foreignrel,
336 : Oid foreigntableid,
337 : ForeignPath *best_path,
338 : List *tlist,
339 : List *scan_clauses,
340 : Plan *outer_plan);
341 : static void postgresBeginForeignScan(ForeignScanState *node, int eflags);
342 : static TupleTableSlot *postgresIterateForeignScan(ForeignScanState *node);
343 : static void postgresReScanForeignScan(ForeignScanState *node);
344 : static void postgresEndForeignScan(ForeignScanState *node);
345 : static void postgresAddForeignUpdateTargets(PlannerInfo *root,
346 : Index rtindex,
347 : RangeTblEntry *target_rte,
348 : Relation target_relation);
349 : static List *postgresPlanForeignModify(PlannerInfo *root,
350 : ModifyTable *plan,
351 : Index resultRelation,
352 : int subplan_index);
353 : static void postgresBeginForeignModify(ModifyTableState *mtstate,
354 : ResultRelInfo *resultRelInfo,
355 : List *fdw_private,
356 : int subplan_index,
357 : int eflags);
358 : static TupleTableSlot *postgresExecForeignInsert(EState *estate,
359 : ResultRelInfo *resultRelInfo,
360 : TupleTableSlot *slot,
361 : TupleTableSlot *planSlot);
362 : static TupleTableSlot **postgresExecForeignBatchInsert(EState *estate,
363 : ResultRelInfo *resultRelInfo,
364 : TupleTableSlot **slots,
365 : TupleTableSlot **planSlots,
366 : int *numSlots);
367 : static int postgresGetForeignModifyBatchSize(ResultRelInfo *resultRelInfo);
368 : static TupleTableSlot *postgresExecForeignUpdate(EState *estate,
369 : ResultRelInfo *resultRelInfo,
370 : TupleTableSlot *slot,
371 : TupleTableSlot *planSlot);
372 : static TupleTableSlot *postgresExecForeignDelete(EState *estate,
373 : ResultRelInfo *resultRelInfo,
374 : TupleTableSlot *slot,
375 : TupleTableSlot *planSlot);
376 : static void postgresEndForeignModify(EState *estate,
377 : ResultRelInfo *resultRelInfo);
378 : static void postgresBeginForeignInsert(ModifyTableState *mtstate,
379 : ResultRelInfo *resultRelInfo);
380 : static void postgresEndForeignInsert(EState *estate,
381 : ResultRelInfo *resultRelInfo);
382 : static int postgresIsForeignRelUpdatable(Relation rel);
383 : static bool postgresPlanDirectModify(PlannerInfo *root,
384 : ModifyTable *plan,
385 : Index resultRelation,
386 : int subplan_index);
387 : static void postgresBeginDirectModify(ForeignScanState *node, int eflags);
388 : static TupleTableSlot *postgresIterateDirectModify(ForeignScanState *node);
389 : static void postgresEndDirectModify(ForeignScanState *node);
390 : static void postgresExplainForeignScan(ForeignScanState *node,
391 : ExplainState *es);
392 : static void postgresExplainForeignModify(ModifyTableState *mtstate,
393 : ResultRelInfo *rinfo,
394 : List *fdw_private,
395 : int subplan_index,
396 : ExplainState *es);
397 : static void postgresExplainDirectModify(ForeignScanState *node,
398 : ExplainState *es);
399 : static void postgresExecForeignTruncate(List *rels,
400 : DropBehavior behavior,
401 : bool restart_seqs);
402 : static bool postgresAnalyzeForeignTable(Relation relation,
403 : AcquireSampleRowsFunc *func,
404 : BlockNumber *totalpages);
405 : static List *postgresImportForeignSchema(ImportForeignSchemaStmt *stmt,
406 : Oid serverOid);
407 : static void postgresGetForeignJoinPaths(PlannerInfo *root,
408 : RelOptInfo *joinrel,
409 : RelOptInfo *outerrel,
410 : RelOptInfo *innerrel,
411 : JoinType jointype,
412 : JoinPathExtraData *extra);
413 : static bool postgresRecheckForeignScan(ForeignScanState *node,
414 : TupleTableSlot *slot);
415 : static void postgresGetForeignUpperPaths(PlannerInfo *root,
416 : UpperRelationKind stage,
417 : RelOptInfo *input_rel,
418 : RelOptInfo *output_rel,
419 : void *extra);
420 : static bool postgresIsForeignPathAsyncCapable(ForeignPath *path);
421 : static void postgresForeignAsyncRequest(AsyncRequest *areq);
422 : static void postgresForeignAsyncConfigureWait(AsyncRequest *areq);
423 : static void postgresForeignAsyncNotify(AsyncRequest *areq);
424 :
425 : /*
426 : * Helper functions
427 : */
428 : static void estimate_path_cost_size(PlannerInfo *root,
429 : RelOptInfo *foreignrel,
430 : List *param_join_conds,
431 : List *pathkeys,
432 : PgFdwPathExtraData *fpextra,
433 : double *p_rows, int *p_width,
434 : int *p_disabled_nodes,
435 : Cost *p_startup_cost, Cost *p_total_cost);
436 : static void get_remote_estimate(const char *sql,
437 : PGconn *conn,
438 : double *rows,
439 : int *width,
440 : Cost *startup_cost,
441 : Cost *total_cost);
442 : static void adjust_foreign_grouping_path_cost(PlannerInfo *root,
443 : List *pathkeys,
444 : double retrieved_rows,
445 : double width,
446 : double limit_tuples,
447 : int *p_disabled_nodes,
448 : Cost *p_startup_cost,
449 : Cost *p_run_cost);
450 : static bool ec_member_matches_foreign(PlannerInfo *root, RelOptInfo *rel,
451 : EquivalenceClass *ec, EquivalenceMember *em,
452 : void *arg);
453 : static void create_cursor(ForeignScanState *node);
454 : static void fetch_more_data(ForeignScanState *node);
455 : static void close_cursor(PGconn *conn, unsigned int cursor_number,
456 : PgFdwConnState *conn_state);
457 : static PgFdwModifyState *create_foreign_modify(EState *estate,
458 : RangeTblEntry *rte,
459 : ResultRelInfo *resultRelInfo,
460 : CmdType operation,
461 : Plan *subplan,
462 : char *query,
463 : List *target_attrs,
464 : int values_end,
465 : bool has_returning,
466 : List *retrieved_attrs);
467 : static TupleTableSlot **execute_foreign_modify(EState *estate,
468 : ResultRelInfo *resultRelInfo,
469 : CmdType operation,
470 : TupleTableSlot **slots,
471 : TupleTableSlot **planSlots,
472 : int *numSlots);
473 : static void prepare_foreign_modify(PgFdwModifyState *fmstate);
474 : static const char **convert_prep_stmt_params(PgFdwModifyState *fmstate,
475 : ItemPointer tupleid,
476 : TupleTableSlot **slots,
477 : int numSlots);
478 : static void store_returning_result(PgFdwModifyState *fmstate,
479 : TupleTableSlot *slot, PGresult *res);
480 : static void finish_foreign_modify(PgFdwModifyState *fmstate);
481 : static void deallocate_query(PgFdwModifyState *fmstate);
482 : static List *build_remote_returning(Index rtindex, Relation rel,
483 : List *returningList);
484 : static void rebuild_fdw_scan_tlist(ForeignScan *fscan, List *tlist);
485 : static void execute_dml_stmt(ForeignScanState *node);
486 : static TupleTableSlot *get_returning_data(ForeignScanState *node);
487 : static void init_returning_filter(PgFdwDirectModifyState *dmstate,
488 : List *fdw_scan_tlist,
489 : Index rtindex);
490 : static TupleTableSlot *apply_returning_filter(PgFdwDirectModifyState *dmstate,
491 : ResultRelInfo *resultRelInfo,
492 : TupleTableSlot *slot,
493 : EState *estate);
494 : static void prepare_query_params(PlanState *node,
495 : List *fdw_exprs,
496 : int numParams,
497 : FmgrInfo **param_flinfo,
498 : List **param_exprs,
499 : const char ***param_values);
500 : static void process_query_params(ExprContext *econtext,
501 : FmgrInfo *param_flinfo,
502 : List *param_exprs,
503 : const char **param_values);
504 : static int postgresAcquireSampleRowsFunc(Relation relation, int elevel,
505 : HeapTuple *rows, int targrows,
506 : double *totalrows,
507 : double *totaldeadrows);
508 : static void analyze_row_processor(PGresult *res, int row,
509 : PgFdwAnalyzeState *astate);
510 : static void produce_tuple_asynchronously(AsyncRequest *areq, bool fetch);
511 : static void fetch_more_data_begin(AsyncRequest *areq);
512 : static void complete_pending_request(AsyncRequest *areq);
513 : static HeapTuple make_tuple_from_result_row(PGresult *res,
514 : int row,
515 : Relation rel,
516 : AttInMetadata *attinmeta,
517 : List *retrieved_attrs,
518 : ForeignScanState *fsstate,
519 : MemoryContext temp_context);
520 : static void conversion_error_callback(void *arg);
521 : static bool foreign_join_ok(PlannerInfo *root, RelOptInfo *joinrel,
522 : JoinType jointype, RelOptInfo *outerrel, RelOptInfo *innerrel,
523 : JoinPathExtraData *extra);
524 : static bool foreign_grouping_ok(PlannerInfo *root, RelOptInfo *grouped_rel,
525 : Node *havingQual);
526 : static List *get_useful_pathkeys_for_relation(PlannerInfo *root,
527 : RelOptInfo *rel);
528 : static List *get_useful_ecs_for_relation(PlannerInfo *root, RelOptInfo *rel);
529 : static void add_paths_with_pathkeys_for_rel(PlannerInfo *root, RelOptInfo *rel,
530 : Path *epq_path, List *restrictlist);
531 : static void add_foreign_grouping_paths(PlannerInfo *root,
532 : RelOptInfo *input_rel,
533 : RelOptInfo *grouped_rel,
534 : GroupPathExtraData *extra);
535 : static void add_foreign_ordered_paths(PlannerInfo *root,
536 : RelOptInfo *input_rel,
537 : RelOptInfo *ordered_rel);
538 : static void add_foreign_final_paths(PlannerInfo *root,
539 : RelOptInfo *input_rel,
540 : RelOptInfo *final_rel,
541 : FinalPathExtraData *extra);
542 : static void apply_server_options(PgFdwRelationInfo *fpinfo);
543 : static void apply_table_options(PgFdwRelationInfo *fpinfo);
544 : static void merge_fdw_options(PgFdwRelationInfo *fpinfo,
545 : const PgFdwRelationInfo *fpinfo_o,
546 : const PgFdwRelationInfo *fpinfo_i);
547 : static int get_batch_size_option(Relation rel);
548 :
549 :
550 : /*
551 : * Foreign-data wrapper handler function: return a struct with pointers
552 : * to my callback routines.
553 : */
554 : Datum
555 1352 : postgres_fdw_handler(PG_FUNCTION_ARGS)
556 : {
557 1352 : FdwRoutine *routine = makeNode(FdwRoutine);
558 :
559 : /* Functions for scanning foreign tables */
560 1352 : routine->GetForeignRelSize = postgresGetForeignRelSize;
561 1352 : routine->GetForeignPaths = postgresGetForeignPaths;
562 1352 : routine->GetForeignPlan = postgresGetForeignPlan;
563 1352 : routine->BeginForeignScan = postgresBeginForeignScan;
564 1352 : routine->IterateForeignScan = postgresIterateForeignScan;
565 1352 : routine->ReScanForeignScan = postgresReScanForeignScan;
566 1352 : routine->EndForeignScan = postgresEndForeignScan;
567 :
568 : /* Functions for updating foreign tables */
569 1352 : routine->AddForeignUpdateTargets = postgresAddForeignUpdateTargets;
570 1352 : routine->PlanForeignModify = postgresPlanForeignModify;
571 1352 : routine->BeginForeignModify = postgresBeginForeignModify;
572 1352 : routine->ExecForeignInsert = postgresExecForeignInsert;
573 1352 : routine->ExecForeignBatchInsert = postgresExecForeignBatchInsert;
574 1352 : routine->GetForeignModifyBatchSize = postgresGetForeignModifyBatchSize;
575 1352 : routine->ExecForeignUpdate = postgresExecForeignUpdate;
576 1352 : routine->ExecForeignDelete = postgresExecForeignDelete;
577 1352 : routine->EndForeignModify = postgresEndForeignModify;
578 1352 : routine->BeginForeignInsert = postgresBeginForeignInsert;
579 1352 : routine->EndForeignInsert = postgresEndForeignInsert;
580 1352 : routine->IsForeignRelUpdatable = postgresIsForeignRelUpdatable;
581 1352 : routine->PlanDirectModify = postgresPlanDirectModify;
582 1352 : routine->BeginDirectModify = postgresBeginDirectModify;
583 1352 : routine->IterateDirectModify = postgresIterateDirectModify;
584 1352 : routine->EndDirectModify = postgresEndDirectModify;
585 :
586 : /* Function for EvalPlanQual rechecks */
587 1352 : routine->RecheckForeignScan = postgresRecheckForeignScan;
588 : /* Support functions for EXPLAIN */
589 1352 : routine->ExplainForeignScan = postgresExplainForeignScan;
590 1352 : routine->ExplainForeignModify = postgresExplainForeignModify;
591 1352 : routine->ExplainDirectModify = postgresExplainDirectModify;
592 :
593 : /* Support function for TRUNCATE */
594 1352 : routine->ExecForeignTruncate = postgresExecForeignTruncate;
595 :
596 : /* Support functions for ANALYZE */
597 1352 : routine->AnalyzeForeignTable = postgresAnalyzeForeignTable;
598 :
599 : /* Support functions for IMPORT FOREIGN SCHEMA */
600 1352 : routine->ImportForeignSchema = postgresImportForeignSchema;
601 :
602 : /* Support functions for join push-down */
603 1352 : routine->GetForeignJoinPaths = postgresGetForeignJoinPaths;
604 :
605 : /* Support functions for upper relation push-down */
606 1352 : routine->GetForeignUpperPaths = postgresGetForeignUpperPaths;
607 :
608 : /* Support functions for asynchronous execution */
609 1352 : routine->IsForeignPathAsyncCapable = postgresIsForeignPathAsyncCapable;
610 1352 : routine->ForeignAsyncRequest = postgresForeignAsyncRequest;
611 1352 : routine->ForeignAsyncConfigureWait = postgresForeignAsyncConfigureWait;
612 1352 : routine->ForeignAsyncNotify = postgresForeignAsyncNotify;
613 :
614 1352 : PG_RETURN_POINTER(routine);
615 : }
616 :
617 : /*
618 : * postgresGetForeignRelSize
619 : * Estimate # of rows and width of the result of the scan
620 : *
621 : * We should consider the effect of all baserestrictinfo clauses here, but
622 : * not any join clauses.
623 : */
624 : static void
625 2360 : postgresGetForeignRelSize(PlannerInfo *root,
626 : RelOptInfo *baserel,
627 : Oid foreigntableid)
628 : {
629 : PgFdwRelationInfo *fpinfo;
630 : ListCell *lc;
631 :
632 : /*
633 : * We use PgFdwRelationInfo to pass various information to subsequent
634 : * functions.
635 : */
636 2360 : fpinfo = (PgFdwRelationInfo *) palloc0(sizeof(PgFdwRelationInfo));
637 2360 : baserel->fdw_private = fpinfo;
638 :
639 : /* Base foreign tables need to be pushed down always. */
640 2360 : fpinfo->pushdown_safe = true;
641 :
642 : /* Look up foreign-table catalog info. */
643 2360 : fpinfo->table = GetForeignTable(foreigntableid);
644 2360 : fpinfo->server = GetForeignServer(fpinfo->table->serverid);
645 :
646 : /*
647 : * Extract user-settable option values. Note that per-table settings of
648 : * use_remote_estimate, fetch_size and async_capable override per-server
649 : * settings of them, respectively.
650 : */
651 2360 : fpinfo->use_remote_estimate = false;
652 2360 : fpinfo->fdw_startup_cost = DEFAULT_FDW_STARTUP_COST;
653 2360 : fpinfo->fdw_tuple_cost = DEFAULT_FDW_TUPLE_COST;
654 2360 : fpinfo->shippable_extensions = NIL;
655 2360 : fpinfo->fetch_size = 100;
656 2360 : fpinfo->async_capable = false;
657 :
658 2360 : apply_server_options(fpinfo);
659 2360 : apply_table_options(fpinfo);
660 :
661 : /*
662 : * If the table or the server is configured to use remote estimates,
663 : * identify which user to do remote access as during planning. This
664 : * should match what ExecCheckPermissions() does. If we fail due to lack
665 : * of permissions, the query would have failed at runtime anyway.
666 : */
667 2360 : if (fpinfo->use_remote_estimate)
668 : {
669 : Oid userid;
670 :
671 606 : userid = OidIsValid(baserel->userid) ? baserel->userid : GetUserId();
672 606 : fpinfo->user = GetUserMapping(userid, fpinfo->server->serverid);
673 : }
674 : else
675 1754 : fpinfo->user = NULL;
676 :
677 : /*
678 : * Identify which baserestrictinfo clauses can be sent to the remote
679 : * server and which can't.
680 : */
681 2356 : classifyConditions(root, baserel, baserel->baserestrictinfo,
682 : &fpinfo->remote_conds, &fpinfo->local_conds);
683 :
684 : /*
685 : * Identify which attributes will need to be retrieved from the remote
686 : * server. These include all attrs needed for joins or final output, plus
687 : * all attrs used in the local_conds. (Note: if we end up using a
688 : * parameterized scan, it's possible that some of the join clauses will be
689 : * sent to the remote and thus we wouldn't really need to retrieve the
690 : * columns used in them. Doesn't seem worth detecting that case though.)
691 : */
692 2356 : fpinfo->attrs_used = NULL;
693 2356 : pull_varattnos((Node *) baserel->reltarget->exprs, baserel->relid,
694 : &fpinfo->attrs_used);
695 2506 : foreach(lc, fpinfo->local_conds)
696 : {
697 150 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc);
698 :
699 150 : pull_varattnos((Node *) rinfo->clause, baserel->relid,
700 : &fpinfo->attrs_used);
701 : }
702 :
703 : /*
704 : * Compute the selectivity and cost of the local_conds, so we don't have
705 : * to do it over again for each path. The best we can do for these
706 : * conditions is to estimate selectivity on the basis of local statistics.
707 : */
708 4712 : fpinfo->local_conds_sel = clauselist_selectivity(root,
709 : fpinfo->local_conds,
710 2356 : baserel->relid,
711 : JOIN_INNER,
712 : NULL);
713 :
714 2356 : cost_qual_eval(&fpinfo->local_conds_cost, fpinfo->local_conds, root);
715 :
716 : /*
717 : * Set # of retrieved rows and cached relation costs to some negative
718 : * value, so that we can detect when they are set to some sensible values,
719 : * during one (usually the first) of the calls to estimate_path_cost_size.
720 : */
721 2356 : fpinfo->retrieved_rows = -1;
722 2356 : fpinfo->rel_startup_cost = -1;
723 2356 : fpinfo->rel_total_cost = -1;
724 :
725 : /*
726 : * If the table or the server is configured to use remote estimates,
727 : * connect to the foreign server and execute EXPLAIN to estimate the
728 : * number of rows selected by the restriction clauses, as well as the
729 : * average row width. Otherwise, estimate using whatever statistics we
730 : * have locally, in a way similar to ordinary tables.
731 : */
732 2356 : if (fpinfo->use_remote_estimate)
733 : {
734 : /*
735 : * Get cost/size estimates with help of remote server. Save the
736 : * values in fpinfo so we don't need to do it again to generate the
737 : * basic foreign path.
738 : */
739 602 : estimate_path_cost_size(root, baserel, NIL, NIL, NULL,
740 : &fpinfo->rows, &fpinfo->width,
741 : &fpinfo->disabled_nodes,
742 : &fpinfo->startup_cost, &fpinfo->total_cost);
743 :
744 : /* Report estimated baserel size to planner. */
745 602 : baserel->rows = fpinfo->rows;
746 602 : baserel->reltarget->width = fpinfo->width;
747 : }
748 : else
749 : {
750 : /*
751 : * If the foreign table has never been ANALYZEd, it will have
752 : * reltuples < 0, meaning "unknown". We can't do much if we're not
753 : * allowed to consult the remote server, but we can use a hack similar
754 : * to plancat.c's treatment of empty relations: use a minimum size
755 : * estimate of 10 pages, and divide by the column-datatype-based width
756 : * estimate to get the corresponding number of tuples.
757 : */
758 1754 : if (baserel->tuples < 0)
759 : {
760 562 : baserel->pages = 10;
761 562 : baserel->tuples =
762 562 : (10 * BLCKSZ) / (baserel->reltarget->width +
763 : MAXALIGN(SizeofHeapTupleHeader));
764 : }
765 :
766 : /* Estimate baserel size as best we can with local statistics. */
767 1754 : set_baserel_size_estimates(root, baserel);
768 :
769 : /* Fill in basically-bogus cost estimates for use later. */
770 1754 : estimate_path_cost_size(root, baserel, NIL, NIL, NULL,
771 : &fpinfo->rows, &fpinfo->width,
772 : &fpinfo->disabled_nodes,
773 : &fpinfo->startup_cost, &fpinfo->total_cost);
774 : }
775 :
776 : /*
777 : * fpinfo->relation_name gets the numeric rangetable index of the foreign
778 : * table RTE. (If this query gets EXPLAIN'd, we'll convert that to a
779 : * human-readable string at that time.)
780 : */
781 2356 : fpinfo->relation_name = psprintf("%u", baserel->relid);
782 :
783 : /* No outer and inner relations. */
784 2356 : fpinfo->make_outerrel_subquery = false;
785 2356 : fpinfo->make_innerrel_subquery = false;
786 2356 : fpinfo->lower_subquery_rels = NULL;
787 2356 : fpinfo->hidden_subquery_rels = NULL;
788 : /* Set the relation index. */
789 2356 : fpinfo->relation_index = baserel->relid;
790 2356 : }
791 :
792 : /*
793 : * get_useful_ecs_for_relation
794 : * Determine which EquivalenceClasses might be involved in useful
795 : * orderings of this relation.
796 : *
797 : * This function is in some respects a mirror image of the core function
798 : * pathkeys_useful_for_merging: for a regular table, we know what indexes
799 : * we have and want to test whether any of them are useful. For a foreign
800 : * table, we don't know what indexes are present on the remote side but
801 : * want to speculate about which ones we'd like to use if they existed.
802 : *
803 : * This function returns a list of potentially-useful equivalence classes,
804 : * but it does not guarantee that an EquivalenceMember exists which contains
805 : * Vars only from the given relation. For example, given ft1 JOIN t1 ON
806 : * ft1.x + t1.x = 0, this function will say that the equivalence class
807 : * containing ft1.x + t1.x is potentially useful. Supposing ft1 is remote and
808 : * t1 is local (or on a different server), it will turn out that no useful
809 : * ORDER BY clause can be generated. It's not our job to figure that out
810 : * here; we're only interested in identifying relevant ECs.
811 : */
812 : static List *
813 1044 : get_useful_ecs_for_relation(PlannerInfo *root, RelOptInfo *rel)
814 : {
815 1044 : List *useful_eclass_list = NIL;
816 : ListCell *lc;
817 : Relids relids;
818 :
819 : /*
820 : * First, consider whether any active EC is potentially useful for a merge
821 : * join against this relation.
822 : */
823 1044 : if (rel->has_eclass_joins)
824 : {
825 1340 : foreach(lc, root->eq_classes)
826 : {
827 926 : EquivalenceClass *cur_ec = (EquivalenceClass *) lfirst(lc);
828 :
829 926 : if (eclass_useful_for_merging(root, cur_ec, rel))
830 494 : useful_eclass_list = lappend(useful_eclass_list, cur_ec);
831 : }
832 : }
833 :
834 : /*
835 : * Next, consider whether there are any non-EC derivable join clauses that
836 : * are merge-joinable. If the joininfo list is empty, we can exit
837 : * quickly.
838 : */
839 1044 : if (rel->joininfo == NIL)
840 760 : return useful_eclass_list;
841 :
842 : /* If this is a child rel, we must use the topmost parent rel to search. */
843 284 : if (IS_OTHER_REL(rel))
844 : {
845 : Assert(!bms_is_empty(rel->top_parent_relids));
846 40 : relids = rel->top_parent_relids;
847 : }
848 : else
849 244 : relids = rel->relids;
850 :
851 : /* Check each join clause in turn. */
852 690 : foreach(lc, rel->joininfo)
853 : {
854 406 : RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(lc);
855 :
856 : /* Consider only mergejoinable clauses */
857 406 : if (restrictinfo->mergeopfamilies == NIL)
858 28 : continue;
859 :
860 : /* Make sure we've got canonical ECs. */
861 378 : update_mergeclause_eclasses(root, restrictinfo);
862 :
863 : /*
864 : * restrictinfo->mergeopfamilies != NIL is sufficient to guarantee
865 : * that left_ec and right_ec will be initialized, per comments in
866 : * distribute_qual_to_rels.
867 : *
868 : * We want to identify which side of this merge-joinable clause
869 : * contains columns from the relation produced by this RelOptInfo. We
870 : * test for overlap, not containment, because there could be extra
871 : * relations on either side. For example, suppose we've got something
872 : * like ((A JOIN B ON A.x = B.x) JOIN C ON A.y = C.y) LEFT JOIN D ON
873 : * A.y = D.y. The input rel might be the joinrel between A and B, and
874 : * we'll consider the join clause A.y = D.y. relids contains a
875 : * relation not involved in the join class (B) and the equivalence
876 : * class for the left-hand side of the clause contains a relation not
877 : * involved in the input rel (C). Despite the fact that we have only
878 : * overlap and not containment in either direction, A.y is potentially
879 : * useful as a sort column.
880 : *
881 : * Note that it's even possible that relids overlaps neither side of
882 : * the join clause. For example, consider A LEFT JOIN B ON A.x = B.x
883 : * AND A.x = 1. The clause A.x = 1 will appear in B's joininfo list,
884 : * but overlaps neither side of B. In that case, we just skip this
885 : * join clause, since it doesn't suggest a useful sort order for this
886 : * relation.
887 : */
888 378 : if (bms_overlap(relids, restrictinfo->right_ec->ec_relids))
889 172 : useful_eclass_list = list_append_unique_ptr(useful_eclass_list,
890 172 : restrictinfo->right_ec);
891 206 : else if (bms_overlap(relids, restrictinfo->left_ec->ec_relids))
892 188 : useful_eclass_list = list_append_unique_ptr(useful_eclass_list,
893 188 : restrictinfo->left_ec);
894 : }
895 :
896 284 : return useful_eclass_list;
897 : }
898 :
899 : /*
900 : * get_useful_pathkeys_for_relation
901 : * Determine which orderings of a relation might be useful.
902 : *
903 : * Getting data in sorted order can be useful either because the requested
904 : * order matches the final output ordering for the overall query we're
905 : * planning, or because it enables an efficient merge join. Here, we try
906 : * to figure out which pathkeys to consider.
907 : */
908 : static List *
909 3022 : get_useful_pathkeys_for_relation(PlannerInfo *root, RelOptInfo *rel)
910 : {
911 3022 : List *useful_pathkeys_list = NIL;
912 : List *useful_eclass_list;
913 3022 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) rel->fdw_private;
914 3022 : EquivalenceClass *query_ec = NULL;
915 : ListCell *lc;
916 :
917 : /*
918 : * Pushing the query_pathkeys to the remote server is always worth
919 : * considering, because it might let us avoid a local sort.
920 : */
921 3022 : fpinfo->qp_is_pushdown_safe = false;
922 3022 : if (root->query_pathkeys)
923 : {
924 1212 : bool query_pathkeys_ok = true;
925 :
926 2288 : foreach(lc, root->query_pathkeys)
927 : {
928 1550 : PathKey *pathkey = (PathKey *) lfirst(lc);
929 :
930 : /*
931 : * The planner and executor don't have any clever strategy for
932 : * taking data sorted by a prefix of the query's pathkeys and
933 : * getting it to be sorted by all of those pathkeys. We'll just
934 : * end up resorting the entire data set. So, unless we can push
935 : * down all of the query pathkeys, forget it.
936 : */
937 1550 : if (!is_foreign_pathkey(root, rel, pathkey))
938 : {
939 474 : query_pathkeys_ok = false;
940 474 : break;
941 : }
942 : }
943 :
944 1212 : if (query_pathkeys_ok)
945 : {
946 738 : useful_pathkeys_list = list_make1(list_copy(root->query_pathkeys));
947 738 : fpinfo->qp_is_pushdown_safe = true;
948 : }
949 : }
950 :
951 : /*
952 : * Even if we're not using remote estimates, having the remote side do the
953 : * sort generally won't be any worse than doing it locally, and it might
954 : * be much better if the remote side can generate data in the right order
955 : * without needing a sort at all. However, what we're going to do next is
956 : * try to generate pathkeys that seem promising for possible merge joins,
957 : * and that's more speculative. A wrong choice might hurt quite a bit, so
958 : * bail out if we can't use remote estimates.
959 : */
960 3022 : if (!fpinfo->use_remote_estimate)
961 1978 : return useful_pathkeys_list;
962 :
963 : /* Get the list of interesting EquivalenceClasses. */
964 1044 : useful_eclass_list = get_useful_ecs_for_relation(root, rel);
965 :
966 : /* Extract unique EC for query, if any, so we don't consider it again. */
967 1044 : if (list_length(root->query_pathkeys) == 1)
968 : {
969 354 : PathKey *query_pathkey = linitial(root->query_pathkeys);
970 :
971 354 : query_ec = query_pathkey->pk_eclass;
972 : }
973 :
974 : /*
975 : * As a heuristic, the only pathkeys we consider here are those of length
976 : * one. It's surely possible to consider more, but since each one we
977 : * choose to consider will generate a round-trip to the remote side, we
978 : * need to be a bit cautious here. It would sure be nice to have a local
979 : * cache of information about remote index definitions...
980 : */
981 1844 : foreach(lc, useful_eclass_list)
982 : {
983 800 : EquivalenceClass *cur_ec = lfirst(lc);
984 : PathKey *pathkey;
985 :
986 : /* If redundant with what we did above, skip it. */
987 800 : if (cur_ec == query_ec)
988 162 : continue;
989 :
990 : /* Can't push down the sort if the EC's opfamily is not shippable. */
991 738 : if (!is_shippable(linitial_oid(cur_ec->ec_opfamilies),
992 : OperatorFamilyRelationId, fpinfo))
993 0 : continue;
994 :
995 : /* If no pushable expression for this rel, skip it. */
996 738 : if (find_em_for_rel(root, cur_ec, rel) == NULL)
997 100 : continue;
998 :
999 : /* Looks like we can generate a pathkey, so let's do it. */
1000 638 : pathkey = make_canonical_pathkey(root, cur_ec,
1001 638 : linitial_oid(cur_ec->ec_opfamilies),
1002 : COMPARE_LT,
1003 : false);
1004 638 : useful_pathkeys_list = lappend(useful_pathkeys_list,
1005 638 : list_make1(pathkey));
1006 : }
1007 :
1008 1044 : return useful_pathkeys_list;
1009 : }
1010 :
1011 : /*
1012 : * postgresGetForeignPaths
1013 : * Create possible scan paths for a scan on the foreign table
1014 : */
1015 : static void
1016 2356 : postgresGetForeignPaths(PlannerInfo *root,
1017 : RelOptInfo *baserel,
1018 : Oid foreigntableid)
1019 : {
1020 2356 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) baserel->fdw_private;
1021 : ForeignPath *path;
1022 : List *ppi_list;
1023 : ListCell *lc;
1024 :
1025 : /*
1026 : * Create simplest ForeignScan path node and add it to baserel. This path
1027 : * corresponds to SeqScan path of regular tables (though depending on what
1028 : * baserestrict conditions we were able to send to remote, there might
1029 : * actually be an indexscan happening there). We already did all the work
1030 : * to estimate cost and size of this path.
1031 : *
1032 : * Although this path uses no join clauses, it could still have required
1033 : * parameterization due to LATERAL refs in its tlist.
1034 : */
1035 2356 : path = create_foreignscan_path(root, baserel,
1036 : NULL, /* default pathtarget */
1037 : fpinfo->rows,
1038 : fpinfo->disabled_nodes,
1039 : fpinfo->startup_cost,
1040 : fpinfo->total_cost,
1041 : NIL, /* no pathkeys */
1042 : baserel->lateral_relids,
1043 : NULL, /* no extra plan */
1044 : NIL, /* no fdw_restrictinfo list */
1045 : NIL); /* no fdw_private list */
1046 2356 : add_path(baserel, (Path *) path);
1047 :
1048 : /* Add paths with pathkeys */
1049 2356 : add_paths_with_pathkeys_for_rel(root, baserel, NULL, NIL);
1050 :
1051 : /*
1052 : * If we're not using remote estimates, stop here. We have no way to
1053 : * estimate whether any join clauses would be worth sending across, so
1054 : * don't bother building parameterized paths.
1055 : */
1056 2356 : if (!fpinfo->use_remote_estimate)
1057 1754 : return;
1058 :
1059 : /*
1060 : * Thumb through all join clauses for the rel to identify which outer
1061 : * relations could supply one or more safe-to-send-to-remote join clauses.
1062 : * We'll build a parameterized path for each such outer relation.
1063 : *
1064 : * It's convenient to manage this by representing each candidate outer
1065 : * relation by the ParamPathInfo node for it. We can then use the
1066 : * ppi_clauses list in the ParamPathInfo node directly as a list of the
1067 : * interesting join clauses for that rel. This takes care of the
1068 : * possibility that there are multiple safe join clauses for such a rel,
1069 : * and also ensures that we account for unsafe join clauses that we'll
1070 : * still have to enforce locally (since the parameterized-path machinery
1071 : * insists that we handle all movable clauses).
1072 : */
1073 602 : ppi_list = NIL;
1074 884 : foreach(lc, baserel->joininfo)
1075 : {
1076 282 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
1077 : Relids required_outer;
1078 : ParamPathInfo *param_info;
1079 :
1080 : /* Check if clause can be moved to this rel */
1081 282 : if (!join_clause_is_movable_to(rinfo, baserel))
1082 192 : continue;
1083 :
1084 : /* See if it is safe to send to remote */
1085 90 : if (!is_foreign_expr(root, baserel, rinfo->clause))
1086 14 : continue;
1087 :
1088 : /* Calculate required outer rels for the resulting path */
1089 76 : required_outer = bms_union(rinfo->clause_relids,
1090 76 : baserel->lateral_relids);
1091 : /* We do not want the foreign rel itself listed in required_outer */
1092 76 : required_outer = bms_del_member(required_outer, baserel->relid);
1093 :
1094 : /*
1095 : * required_outer probably can't be empty here, but if it were, we
1096 : * couldn't make a parameterized path.
1097 : */
1098 76 : if (bms_is_empty(required_outer))
1099 0 : continue;
1100 :
1101 : /* Get the ParamPathInfo */
1102 76 : param_info = get_baserel_parampathinfo(root, baserel,
1103 : required_outer);
1104 : Assert(param_info != NULL);
1105 :
1106 : /*
1107 : * Add it to list unless we already have it. Testing pointer equality
1108 : * is OK since get_baserel_parampathinfo won't make duplicates.
1109 : */
1110 76 : ppi_list = list_append_unique_ptr(ppi_list, param_info);
1111 : }
1112 :
1113 : /*
1114 : * The above scan examined only "generic" join clauses, not those that
1115 : * were absorbed into EquivalenceClauses. See if we can make anything out
1116 : * of EquivalenceClauses.
1117 : */
1118 602 : if (baserel->has_eclass_joins)
1119 : {
1120 : /*
1121 : * We repeatedly scan the eclass list looking for column references
1122 : * (or expressions) belonging to the foreign rel. Each time we find
1123 : * one, we generate a list of equivalence joinclauses for it, and then
1124 : * see if any are safe to send to the remote. Repeat till there are
1125 : * no more candidate EC members.
1126 : */
1127 : ec_member_foreign_arg arg;
1128 :
1129 266 : arg.already_used = NIL;
1130 : for (;;)
1131 282 : {
1132 : List *clauses;
1133 :
1134 : /* Make clauses, skipping any that join to lateral_referencers */
1135 548 : arg.current = NULL;
1136 548 : clauses = generate_implied_equalities_for_column(root,
1137 : baserel,
1138 : ec_member_matches_foreign,
1139 : &arg,
1140 : baserel->lateral_referencers);
1141 :
1142 : /* Done if there are no more expressions in the foreign rel */
1143 548 : if (arg.current == NULL)
1144 : {
1145 : Assert(clauses == NIL);
1146 266 : break;
1147 : }
1148 :
1149 : /* Scan the extracted join clauses */
1150 628 : foreach(lc, clauses)
1151 : {
1152 346 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
1153 : Relids required_outer;
1154 : ParamPathInfo *param_info;
1155 :
1156 : /* Check if clause can be moved to this rel */
1157 346 : if (!join_clause_is_movable_to(rinfo, baserel))
1158 0 : continue;
1159 :
1160 : /* See if it is safe to send to remote */
1161 346 : if (!is_foreign_expr(root, baserel, rinfo->clause))
1162 14 : continue;
1163 :
1164 : /* Calculate required outer rels for the resulting path */
1165 332 : required_outer = bms_union(rinfo->clause_relids,
1166 332 : baserel->lateral_relids);
1167 332 : required_outer = bms_del_member(required_outer, baserel->relid);
1168 332 : if (bms_is_empty(required_outer))
1169 0 : continue;
1170 :
1171 : /* Get the ParamPathInfo */
1172 332 : param_info = get_baserel_parampathinfo(root, baserel,
1173 : required_outer);
1174 : Assert(param_info != NULL);
1175 :
1176 : /* Add it to list unless we already have it */
1177 332 : ppi_list = list_append_unique_ptr(ppi_list, param_info);
1178 : }
1179 :
1180 : /* Try again, now ignoring the expression we found this time */
1181 282 : arg.already_used = lappend(arg.already_used, arg.current);
1182 : }
1183 : }
1184 :
1185 : /*
1186 : * Now build a path for each useful outer relation.
1187 : */
1188 990 : foreach(lc, ppi_list)
1189 : {
1190 388 : ParamPathInfo *param_info = (ParamPathInfo *) lfirst(lc);
1191 : double rows;
1192 : int width;
1193 : int disabled_nodes;
1194 : Cost startup_cost;
1195 : Cost total_cost;
1196 :
1197 : /* Get a cost estimate from the remote */
1198 388 : estimate_path_cost_size(root, baserel,
1199 : param_info->ppi_clauses, NIL, NULL,
1200 : &rows, &width, &disabled_nodes,
1201 : &startup_cost, &total_cost);
1202 :
1203 : /*
1204 : * ppi_rows currently won't get looked at by anything, but still we
1205 : * may as well ensure that it matches our idea of the rowcount.
1206 : */
1207 388 : param_info->ppi_rows = rows;
1208 :
1209 : /* Make the path */
1210 388 : path = create_foreignscan_path(root, baserel,
1211 : NULL, /* default pathtarget */
1212 : rows,
1213 : disabled_nodes,
1214 : startup_cost,
1215 : total_cost,
1216 : NIL, /* no pathkeys */
1217 : param_info->ppi_req_outer,
1218 : NULL,
1219 : NIL, /* no fdw_restrictinfo list */
1220 : NIL); /* no fdw_private list */
1221 388 : add_path(baserel, (Path *) path);
1222 : }
1223 : }
1224 :
1225 : /*
1226 : * postgresGetForeignPlan
1227 : * Create ForeignScan plan node which implements selected best path
1228 : */
1229 : static ForeignScan *
1230 1984 : postgresGetForeignPlan(PlannerInfo *root,
1231 : RelOptInfo *foreignrel,
1232 : Oid foreigntableid,
1233 : ForeignPath *best_path,
1234 : List *tlist,
1235 : List *scan_clauses,
1236 : Plan *outer_plan)
1237 : {
1238 1984 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) foreignrel->fdw_private;
1239 : Index scan_relid;
1240 : List *fdw_private;
1241 1984 : List *remote_exprs = NIL;
1242 1984 : List *local_exprs = NIL;
1243 1984 : List *params_list = NIL;
1244 1984 : List *fdw_scan_tlist = NIL;
1245 1984 : List *fdw_recheck_quals = NIL;
1246 : List *retrieved_attrs;
1247 : StringInfoData sql;
1248 1984 : bool has_final_sort = false;
1249 1984 : bool has_limit = false;
1250 : ListCell *lc;
1251 :
1252 : /*
1253 : * Get FDW private data created by postgresGetForeignUpperPaths(), if any.
1254 : */
1255 1984 : if (best_path->fdw_private)
1256 : {
1257 302 : has_final_sort = boolVal(list_nth(best_path->fdw_private,
1258 : FdwPathPrivateHasFinalSort));
1259 302 : has_limit = boolVal(list_nth(best_path->fdw_private,
1260 : FdwPathPrivateHasLimit));
1261 : }
1262 :
1263 1984 : if (IS_SIMPLE_REL(foreignrel))
1264 : {
1265 : /*
1266 : * For base relations, set scan_relid as the relid of the relation.
1267 : */
1268 1420 : scan_relid = foreignrel->relid;
1269 :
1270 : /*
1271 : * In a base-relation scan, we must apply the given scan_clauses.
1272 : *
1273 : * Separate the scan_clauses into those that can be executed remotely
1274 : * and those that can't. baserestrictinfo clauses that were
1275 : * previously determined to be safe or unsafe by classifyConditions
1276 : * are found in fpinfo->remote_conds and fpinfo->local_conds. Anything
1277 : * else in the scan_clauses list will be a join clause, which we have
1278 : * to check for remote-safety.
1279 : *
1280 : * Note: the join clauses we see here should be the exact same ones
1281 : * previously examined by postgresGetForeignPaths. Possibly it'd be
1282 : * worth passing forward the classification work done then, rather
1283 : * than repeating it here.
1284 : *
1285 : * This code must match "extract_actual_clauses(scan_clauses, false)"
1286 : * except for the additional decision about remote versus local
1287 : * execution.
1288 : */
1289 2134 : foreach(lc, scan_clauses)
1290 : {
1291 714 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc);
1292 :
1293 : /* Ignore any pseudoconstants, they're dealt with elsewhere */
1294 714 : if (rinfo->pseudoconstant)
1295 8 : continue;
1296 :
1297 706 : if (list_member_ptr(fpinfo->remote_conds, rinfo))
1298 538 : remote_exprs = lappend(remote_exprs, rinfo->clause);
1299 168 : else if (list_member_ptr(fpinfo->local_conds, rinfo))
1300 142 : local_exprs = lappend(local_exprs, rinfo->clause);
1301 26 : else if (is_foreign_expr(root, foreignrel, rinfo->clause))
1302 22 : remote_exprs = lappend(remote_exprs, rinfo->clause);
1303 : else
1304 4 : local_exprs = lappend(local_exprs, rinfo->clause);
1305 : }
1306 :
1307 : /*
1308 : * For a base-relation scan, we have to support EPQ recheck, which
1309 : * should recheck all the remote quals.
1310 : */
1311 1420 : fdw_recheck_quals = remote_exprs;
1312 : }
1313 : else
1314 : {
1315 : /*
1316 : * Join relation or upper relation - set scan_relid to 0.
1317 : */
1318 564 : scan_relid = 0;
1319 :
1320 : /*
1321 : * For a join rel, baserestrictinfo is NIL and we are not considering
1322 : * parameterization right now, so there should be no scan_clauses for
1323 : * a joinrel or an upper rel either.
1324 : */
1325 : Assert(!scan_clauses);
1326 :
1327 : /*
1328 : * Instead we get the conditions to apply from the fdw_private
1329 : * structure.
1330 : */
1331 564 : remote_exprs = extract_actual_clauses(fpinfo->remote_conds, false);
1332 564 : local_exprs = extract_actual_clauses(fpinfo->local_conds, false);
1333 :
1334 : /*
1335 : * We leave fdw_recheck_quals empty in this case, since we never need
1336 : * to apply EPQ recheck clauses. In the case of a joinrel, EPQ
1337 : * recheck is handled elsewhere --- see postgresGetForeignJoinPaths().
1338 : * If we're planning an upperrel (ie, remote grouping or aggregation)
1339 : * then there's no EPQ to do because SELECT FOR UPDATE wouldn't be
1340 : * allowed, and indeed we *can't* put the remote clauses into
1341 : * fdw_recheck_quals because the unaggregated Vars won't be available
1342 : * locally.
1343 : */
1344 :
1345 : /* Build the list of columns to be fetched from the foreign server. */
1346 564 : fdw_scan_tlist = build_tlist_to_deparse(foreignrel);
1347 :
1348 : /*
1349 : * Ensure that the outer plan produces a tuple whose descriptor
1350 : * matches our scan tuple slot. Also, remove the local conditions
1351 : * from outer plan's quals, lest they be evaluated twice, once by the
1352 : * local plan and once by the scan.
1353 : */
1354 564 : if (outer_plan)
1355 : {
1356 : /*
1357 : * Right now, we only consider grouping and aggregation beyond
1358 : * joins. Queries involving aggregates or grouping do not require
1359 : * EPQ mechanism, hence should not have an outer plan here.
1360 : */
1361 : Assert(!IS_UPPER_REL(foreignrel));
1362 :
1363 : /*
1364 : * First, update the plan's qual list if possible. In some cases
1365 : * the quals might be enforced below the topmost plan level, in
1366 : * which case we'll fail to remove them; it's not worth working
1367 : * harder than this.
1368 : */
1369 54 : foreach(lc, local_exprs)
1370 : {
1371 6 : Node *qual = lfirst(lc);
1372 :
1373 6 : outer_plan->qual = list_delete(outer_plan->qual, qual);
1374 :
1375 : /*
1376 : * For an inner join the local conditions of foreign scan plan
1377 : * can be part of the joinquals as well. (They might also be
1378 : * in the mergequals or hashquals, but we can't touch those
1379 : * without breaking the plan.)
1380 : */
1381 6 : if (IsA(outer_plan, NestLoop) ||
1382 2 : IsA(outer_plan, MergeJoin) ||
1383 2 : IsA(outer_plan, HashJoin))
1384 : {
1385 4 : Join *join_plan = (Join *) outer_plan;
1386 :
1387 4 : if (join_plan->jointype == JOIN_INNER)
1388 4 : join_plan->joinqual = list_delete(join_plan->joinqual,
1389 : qual);
1390 : }
1391 : }
1392 :
1393 : /*
1394 : * Now fix the subplan's tlist --- this might result in inserting
1395 : * a Result node atop the plan tree.
1396 : */
1397 48 : outer_plan = change_plan_targetlist(outer_plan, fdw_scan_tlist,
1398 48 : best_path->path.parallel_safe);
1399 : }
1400 : }
1401 :
1402 : /*
1403 : * Build the query string to be sent for execution, and identify
1404 : * expressions to be sent as parameters.
1405 : */
1406 1984 : initStringInfo(&sql);
1407 1984 : deparseSelectStmtForRel(&sql, root, foreignrel, fdw_scan_tlist,
1408 : remote_exprs, best_path->path.pathkeys,
1409 : has_final_sort, has_limit, false,
1410 : &retrieved_attrs, ¶ms_list);
1411 :
1412 : /* Remember remote_exprs for possible use by postgresPlanDirectModify */
1413 1984 : fpinfo->final_remote_exprs = remote_exprs;
1414 :
1415 : /*
1416 : * Build the fdw_private list that will be available to the executor.
1417 : * Items in the list must match order in enum FdwScanPrivateIndex.
1418 : */
1419 1984 : fdw_private = list_make3(makeString(sql.data),
1420 : retrieved_attrs,
1421 : makeInteger(fpinfo->fetch_size));
1422 1984 : if (IS_JOIN_REL(foreignrel) || IS_UPPER_REL(foreignrel))
1423 564 : fdw_private = lappend(fdw_private,
1424 564 : makeString(fpinfo->relation_name));
1425 :
1426 : /*
1427 : * Create the ForeignScan node for the given relation.
1428 : *
1429 : * Note that the remote parameter expressions are stored in the fdw_exprs
1430 : * field of the finished plan node; we can't keep them in private state
1431 : * because then they wouldn't be subject to later planner processing.
1432 : */
1433 1984 : return make_foreignscan(tlist,
1434 : local_exprs,
1435 : scan_relid,
1436 : params_list,
1437 : fdw_private,
1438 : fdw_scan_tlist,
1439 : fdw_recheck_quals,
1440 : outer_plan);
1441 : }
1442 :
1443 : /*
1444 : * Construct a tuple descriptor for the scan tuples handled by a foreign join.
1445 : */
1446 : static TupleDesc
1447 320 : get_tupdesc_for_join_scan_tuples(ForeignScanState *node)
1448 : {
1449 320 : ForeignScan *fsplan = (ForeignScan *) node->ss.ps.plan;
1450 320 : EState *estate = node->ss.ps.state;
1451 : TupleDesc tupdesc;
1452 :
1453 : /*
1454 : * The core code has already set up a scan tuple slot based on
1455 : * fsplan->fdw_scan_tlist, and this slot's tupdesc is mostly good enough,
1456 : * but there's one case where it isn't. If we have any whole-row row
1457 : * identifier Vars, they may have vartype RECORD, and we need to replace
1458 : * that with the associated table's actual composite type. This ensures
1459 : * that when we read those ROW() expression values from the remote server,
1460 : * we can convert them to a composite type the local server knows.
1461 : */
1462 320 : tupdesc = CreateTupleDescCopy(node->ss.ss_ScanTupleSlot->tts_tupleDescriptor);
1463 1346 : for (int i = 0; i < tupdesc->natts; i++)
1464 : {
1465 1026 : Form_pg_attribute att = TupleDescAttr(tupdesc, i);
1466 : Var *var;
1467 : RangeTblEntry *rte;
1468 : Oid reltype;
1469 :
1470 : /* Nothing to do if it's not a generic RECORD attribute */
1471 1026 : if (att->atttypid != RECORDOID || att->atttypmod >= 0)
1472 1020 : continue;
1473 :
1474 : /*
1475 : * If we can't identify the referenced table, do nothing. This'll
1476 : * likely lead to failure later, but perhaps we can muddle through.
1477 : */
1478 6 : var = (Var *) list_nth_node(TargetEntry, fsplan->fdw_scan_tlist,
1479 : i)->expr;
1480 6 : if (!IsA(var, Var) || var->varattno != 0)
1481 0 : continue;
1482 6 : rte = list_nth(estate->es_range_table, var->varno - 1);
1483 6 : if (rte->rtekind != RTE_RELATION)
1484 0 : continue;
1485 6 : reltype = get_rel_type_id(rte->relid);
1486 6 : if (!OidIsValid(reltype))
1487 0 : continue;
1488 6 : att->atttypid = reltype;
1489 : /* shouldn't need to change anything else */
1490 : }
1491 320 : return tupdesc;
1492 : }
1493 :
1494 : /*
1495 : * postgresBeginForeignScan
1496 : * Initiate an executor scan of a foreign PostgreSQL table.
1497 : */
1498 : static void
1499 1760 : postgresBeginForeignScan(ForeignScanState *node, int eflags)
1500 : {
1501 1760 : ForeignScan *fsplan = (ForeignScan *) node->ss.ps.plan;
1502 1760 : EState *estate = node->ss.ps.state;
1503 : PgFdwScanState *fsstate;
1504 : RangeTblEntry *rte;
1505 : Oid userid;
1506 : ForeignTable *table;
1507 : UserMapping *user;
1508 : int rtindex;
1509 : int numParams;
1510 :
1511 : /*
1512 : * Do nothing in EXPLAIN (no ANALYZE) case. node->fdw_state stays NULL.
1513 : */
1514 1760 : if (eflags & EXEC_FLAG_EXPLAIN_ONLY)
1515 760 : return;
1516 :
1517 : /*
1518 : * We'll save private state in node->fdw_state.
1519 : */
1520 1000 : fsstate = (PgFdwScanState *) palloc0(sizeof(PgFdwScanState));
1521 1000 : node->fdw_state = fsstate;
1522 :
1523 : /*
1524 : * Identify which user to do the remote access as. This should match what
1525 : * ExecCheckPermissions() does.
1526 : */
1527 1000 : userid = OidIsValid(fsplan->checkAsUser) ? fsplan->checkAsUser : GetUserId();
1528 1000 : if (fsplan->scan.scanrelid > 0)
1529 680 : rtindex = fsplan->scan.scanrelid;
1530 : else
1531 320 : rtindex = bms_next_member(fsplan->fs_base_relids, -1);
1532 1000 : rte = exec_rt_fetch(rtindex, estate);
1533 :
1534 : /* Get info about foreign table. */
1535 1000 : table = GetForeignTable(rte->relid);
1536 1000 : user = GetUserMapping(userid, table->serverid);
1537 :
1538 : /*
1539 : * Get connection to the foreign server. Connection manager will
1540 : * establish new connection if necessary.
1541 : */
1542 1000 : fsstate->conn = GetConnection(user, false, &fsstate->conn_state);
1543 :
1544 : /* Assign a unique ID for my cursor */
1545 984 : fsstate->cursor_number = GetCursorNumber(fsstate->conn);
1546 984 : fsstate->cursor_exists = false;
1547 :
1548 : /* Get private info created by planner functions. */
1549 984 : fsstate->query = strVal(list_nth(fsplan->fdw_private,
1550 : FdwScanPrivateSelectSql));
1551 984 : fsstate->retrieved_attrs = (List *) list_nth(fsplan->fdw_private,
1552 : FdwScanPrivateRetrievedAttrs);
1553 984 : fsstate->fetch_size = intVal(list_nth(fsplan->fdw_private,
1554 : FdwScanPrivateFetchSize));
1555 :
1556 : /* Create contexts for batches of tuples and per-tuple temp workspace. */
1557 984 : fsstate->batch_cxt = AllocSetContextCreate(estate->es_query_cxt,
1558 : "postgres_fdw tuple data",
1559 : ALLOCSET_DEFAULT_SIZES);
1560 984 : fsstate->temp_cxt = AllocSetContextCreate(estate->es_query_cxt,
1561 : "postgres_fdw temporary data",
1562 : ALLOCSET_SMALL_SIZES);
1563 :
1564 : /*
1565 : * Get info we'll need for converting data fetched from the foreign server
1566 : * into local representation and error reporting during that process.
1567 : */
1568 984 : if (fsplan->scan.scanrelid > 0)
1569 : {
1570 666 : fsstate->rel = node->ss.ss_currentRelation;
1571 666 : fsstate->tupdesc = RelationGetDescr(fsstate->rel);
1572 : }
1573 : else
1574 : {
1575 318 : fsstate->rel = NULL;
1576 318 : fsstate->tupdesc = get_tupdesc_for_join_scan_tuples(node);
1577 : }
1578 :
1579 984 : fsstate->attinmeta = TupleDescGetAttInMetadata(fsstate->tupdesc);
1580 :
1581 : /*
1582 : * Prepare for processing of parameters used in remote query, if any.
1583 : */
1584 984 : numParams = list_length(fsplan->fdw_exprs);
1585 984 : fsstate->numParams = numParams;
1586 984 : if (numParams > 0)
1587 38 : prepare_query_params((PlanState *) node,
1588 : fsplan->fdw_exprs,
1589 : numParams,
1590 : &fsstate->param_flinfo,
1591 : &fsstate->param_exprs,
1592 : &fsstate->param_values);
1593 :
1594 : /* Set the async-capable flag */
1595 984 : fsstate->async_capable = node->ss.ps.async_capable;
1596 : }
1597 :
1598 : /*
1599 : * postgresIterateForeignScan
1600 : * Retrieve next row from the result set, or clear tuple slot to indicate
1601 : * EOF.
1602 : */
1603 : static TupleTableSlot *
1604 141764 : postgresIterateForeignScan(ForeignScanState *node)
1605 : {
1606 141764 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
1607 141764 : TupleTableSlot *slot = node->ss.ss_ScanTupleSlot;
1608 :
1609 : /*
1610 : * In sync mode, if this is the first call after Begin or ReScan, we need
1611 : * to create the cursor on the remote side. In async mode, we would have
1612 : * already created the cursor before we get here, even if this is the
1613 : * first call after Begin or ReScan.
1614 : */
1615 141764 : if (!fsstate->cursor_exists)
1616 1518 : create_cursor(node);
1617 :
1618 : /*
1619 : * Get some more tuples, if we've run out.
1620 : */
1621 141758 : if (fsstate->next_tuple >= fsstate->num_tuples)
1622 : {
1623 : /* In async mode, just clear tuple slot. */
1624 4074 : if (fsstate->async_capable)
1625 64 : return ExecClearTuple(slot);
1626 : /* No point in another fetch if we already detected EOF, though. */
1627 4010 : if (!fsstate->eof_reached)
1628 2672 : fetch_more_data(node);
1629 : /* If we didn't get any tuples, must be end of data. */
1630 4000 : if (fsstate->next_tuple >= fsstate->num_tuples)
1631 1478 : return ExecClearTuple(slot);
1632 : }
1633 :
1634 : /*
1635 : * Return the next tuple.
1636 : */
1637 140206 : ExecStoreHeapTuple(fsstate->tuples[fsstate->next_tuple++],
1638 : slot,
1639 : false);
1640 :
1641 140206 : return slot;
1642 : }
1643 :
1644 : /*
1645 : * postgresReScanForeignScan
1646 : * Restart the scan.
1647 : */
1648 : static void
1649 802 : postgresReScanForeignScan(ForeignScanState *node)
1650 : {
1651 802 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
1652 : char sql[64];
1653 : PGresult *res;
1654 :
1655 : /* If we haven't created the cursor yet, nothing to do. */
1656 802 : if (!fsstate->cursor_exists)
1657 88 : return;
1658 :
1659 : /*
1660 : * If the node is async-capable, and an asynchronous fetch for it has
1661 : * begun, the asynchronous fetch might not have yet completed. Check if
1662 : * the node is async-capable, and an asynchronous fetch for it is still in
1663 : * progress; if so, complete the asynchronous fetch before restarting the
1664 : * scan.
1665 : */
1666 738 : if (fsstate->async_capable &&
1667 42 : fsstate->conn_state->pendingAreq &&
1668 4 : fsstate->conn_state->pendingAreq->requestee == (PlanState *) node)
1669 2 : fetch_more_data(node);
1670 :
1671 : /*
1672 : * If any internal parameters affecting this node have changed, we'd
1673 : * better destroy and recreate the cursor. Otherwise, if the remote
1674 : * server is v14 or older, rewinding it should be good enough; if not,
1675 : * rewind is only allowed for scrollable cursors, but we don't have a way
1676 : * to check the scrollability of it, so destroy and recreate it in any
1677 : * case. If we've only fetched zero or one batch, we needn't even rewind
1678 : * the cursor, just rescan what we have.
1679 : */
1680 738 : if (node->ss.ps.chgParam != NULL)
1681 : {
1682 676 : fsstate->cursor_exists = false;
1683 676 : snprintf(sql, sizeof(sql), "CLOSE c%u",
1684 : fsstate->cursor_number);
1685 : }
1686 62 : else if (fsstate->fetch_ct_2 > 1)
1687 : {
1688 38 : if (PQserverVersion(fsstate->conn) < 150000)
1689 0 : snprintf(sql, sizeof(sql), "MOVE BACKWARD ALL IN c%u",
1690 : fsstate->cursor_number);
1691 : else
1692 : {
1693 38 : fsstate->cursor_exists = false;
1694 38 : snprintf(sql, sizeof(sql), "CLOSE c%u",
1695 : fsstate->cursor_number);
1696 : }
1697 : }
1698 : else
1699 : {
1700 : /* Easy: just rescan what we already have in memory, if anything */
1701 24 : fsstate->next_tuple = 0;
1702 24 : return;
1703 : }
1704 :
1705 714 : res = pgfdw_exec_query(fsstate->conn, sql, fsstate->conn_state);
1706 714 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1707 0 : pgfdw_report_error(res, fsstate->conn, sql);
1708 714 : PQclear(res);
1709 :
1710 : /* Now force a fresh FETCH. */
1711 714 : fsstate->tuples = NULL;
1712 714 : fsstate->num_tuples = 0;
1713 714 : fsstate->next_tuple = 0;
1714 714 : fsstate->fetch_ct_2 = 0;
1715 714 : fsstate->eof_reached = false;
1716 : }
1717 :
1718 : /*
1719 : * postgresEndForeignScan
1720 : * Finish scanning foreign table and dispose objects used for this scan
1721 : */
1722 : static void
1723 1702 : postgresEndForeignScan(ForeignScanState *node)
1724 : {
1725 1702 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
1726 :
1727 : /* if fsstate is NULL, we are in EXPLAIN; nothing to do */
1728 1702 : if (fsstate == NULL)
1729 760 : return;
1730 :
1731 : /* Close the cursor if open, to prevent accumulation of cursors */
1732 942 : if (fsstate->cursor_exists)
1733 900 : close_cursor(fsstate->conn, fsstate->cursor_number,
1734 : fsstate->conn_state);
1735 :
1736 : /* Release remote connection */
1737 942 : ReleaseConnection(fsstate->conn);
1738 942 : fsstate->conn = NULL;
1739 :
1740 : /* MemoryContexts will be deleted automatically. */
1741 : }
1742 :
1743 : /*
1744 : * postgresAddForeignUpdateTargets
1745 : * Add resjunk column(s) needed for update/delete on a foreign table
1746 : */
1747 : static void
1748 378 : postgresAddForeignUpdateTargets(PlannerInfo *root,
1749 : Index rtindex,
1750 : RangeTblEntry *target_rte,
1751 : Relation target_relation)
1752 : {
1753 : Var *var;
1754 :
1755 : /*
1756 : * In postgres_fdw, what we need is the ctid, same as for a regular table.
1757 : */
1758 :
1759 : /* Make a Var representing the desired value */
1760 378 : var = makeVar(rtindex,
1761 : SelfItemPointerAttributeNumber,
1762 : TIDOID,
1763 : -1,
1764 : InvalidOid,
1765 : 0);
1766 :
1767 : /* Register it as a row-identity column needed by this target rel */
1768 378 : add_row_identity_var(root, var, rtindex, "ctid");
1769 378 : }
1770 :
1771 : /*
1772 : * postgresPlanForeignModify
1773 : * Plan an insert/update/delete operation on a foreign table
1774 : */
1775 : static List *
1776 340 : postgresPlanForeignModify(PlannerInfo *root,
1777 : ModifyTable *plan,
1778 : Index resultRelation,
1779 : int subplan_index)
1780 : {
1781 340 : CmdType operation = plan->operation;
1782 340 : RangeTblEntry *rte = planner_rt_fetch(resultRelation, root);
1783 : Relation rel;
1784 : StringInfoData sql;
1785 340 : List *targetAttrs = NIL;
1786 340 : List *withCheckOptionList = NIL;
1787 340 : List *returningList = NIL;
1788 340 : List *retrieved_attrs = NIL;
1789 340 : bool doNothing = false;
1790 340 : int values_end_len = -1;
1791 :
1792 340 : initStringInfo(&sql);
1793 :
1794 : /*
1795 : * Core code already has some lock on each rel being planned, so we can
1796 : * use NoLock here.
1797 : */
1798 340 : rel = table_open(rte->relid, NoLock);
1799 :
1800 : /*
1801 : * In an INSERT, we transmit all columns that are defined in the foreign
1802 : * table. In an UPDATE, if there are BEFORE ROW UPDATE triggers on the
1803 : * foreign table, we transmit all columns like INSERT; else we transmit
1804 : * only columns that were explicitly targets of the UPDATE, so as to avoid
1805 : * unnecessary data transmission. (We can't do that for INSERT since we
1806 : * would miss sending default values for columns not listed in the source
1807 : * statement, and for UPDATE if there are BEFORE ROW UPDATE triggers since
1808 : * those triggers might change values for non-target columns, in which
1809 : * case we would miss sending changed values for those columns.)
1810 : */
1811 340 : if (operation == CMD_INSERT ||
1812 120 : (operation == CMD_UPDATE &&
1813 120 : rel->trigdesc &&
1814 36 : rel->trigdesc->trig_update_before_row))
1815 206 : {
1816 206 : TupleDesc tupdesc = RelationGetDescr(rel);
1817 : int attnum;
1818 :
1819 868 : for (attnum = 1; attnum <= tupdesc->natts; attnum++)
1820 : {
1821 662 : CompactAttribute *attr = TupleDescCompactAttr(tupdesc, attnum - 1);
1822 :
1823 662 : if (!attr->attisdropped)
1824 628 : targetAttrs = lappend_int(targetAttrs, attnum);
1825 : }
1826 : }
1827 134 : else if (operation == CMD_UPDATE)
1828 : {
1829 : int col;
1830 90 : RelOptInfo *rel = find_base_rel(root, resultRelation);
1831 90 : Bitmapset *allUpdatedCols = get_rel_all_updated_cols(root, rel);
1832 :
1833 90 : col = -1;
1834 200 : while ((col = bms_next_member(allUpdatedCols, col)) >= 0)
1835 : {
1836 : /* bit numbers are offset by FirstLowInvalidHeapAttributeNumber */
1837 110 : AttrNumber attno = col + FirstLowInvalidHeapAttributeNumber;
1838 :
1839 110 : if (attno <= InvalidAttrNumber) /* shouldn't happen */
1840 0 : elog(ERROR, "system-column update is not supported");
1841 110 : targetAttrs = lappend_int(targetAttrs, attno);
1842 : }
1843 : }
1844 :
1845 : /*
1846 : * Extract the relevant WITH CHECK OPTION list if any.
1847 : */
1848 340 : if (plan->withCheckOptionLists)
1849 32 : withCheckOptionList = (List *) list_nth(plan->withCheckOptionLists,
1850 : subplan_index);
1851 :
1852 : /*
1853 : * Extract the relevant RETURNING list if any.
1854 : */
1855 340 : if (plan->returningLists)
1856 64 : returningList = (List *) list_nth(plan->returningLists, subplan_index);
1857 :
1858 : /*
1859 : * ON CONFLICT DO UPDATE and DO NOTHING case with inference specification
1860 : * should have already been rejected in the optimizer, as presently there
1861 : * is no way to recognize an arbiter index on a foreign table. Only DO
1862 : * NOTHING is supported without an inference specification.
1863 : */
1864 340 : if (plan->onConflictAction == ONCONFLICT_NOTHING)
1865 2 : doNothing = true;
1866 338 : else if (plan->onConflictAction != ONCONFLICT_NONE)
1867 0 : elog(ERROR, "unexpected ON CONFLICT specification: %d",
1868 : (int) plan->onConflictAction);
1869 :
1870 : /*
1871 : * Construct the SQL command string.
1872 : */
1873 340 : switch (operation)
1874 : {
1875 176 : case CMD_INSERT:
1876 176 : deparseInsertSql(&sql, rte, resultRelation, rel,
1877 : targetAttrs, doNothing,
1878 : withCheckOptionList, returningList,
1879 : &retrieved_attrs, &values_end_len);
1880 176 : break;
1881 120 : case CMD_UPDATE:
1882 120 : deparseUpdateSql(&sql, rte, resultRelation, rel,
1883 : targetAttrs,
1884 : withCheckOptionList, returningList,
1885 : &retrieved_attrs);
1886 120 : break;
1887 44 : case CMD_DELETE:
1888 44 : deparseDeleteSql(&sql, rte, resultRelation, rel,
1889 : returningList,
1890 : &retrieved_attrs);
1891 44 : break;
1892 0 : default:
1893 0 : elog(ERROR, "unexpected operation: %d", (int) operation);
1894 : break;
1895 : }
1896 :
1897 340 : table_close(rel, NoLock);
1898 :
1899 : /*
1900 : * Build the fdw_private list that will be available to the executor.
1901 : * Items in the list must match enum FdwModifyPrivateIndex, above.
1902 : */
1903 340 : return list_make5(makeString(sql.data),
1904 : targetAttrs,
1905 : makeInteger(values_end_len),
1906 : makeBoolean((retrieved_attrs != NIL)),
1907 : retrieved_attrs);
1908 : }
1909 :
1910 : /*
1911 : * postgresBeginForeignModify
1912 : * Begin an insert/update/delete operation on a foreign table
1913 : */
1914 : static void
1915 340 : postgresBeginForeignModify(ModifyTableState *mtstate,
1916 : ResultRelInfo *resultRelInfo,
1917 : List *fdw_private,
1918 : int subplan_index,
1919 : int eflags)
1920 : {
1921 : PgFdwModifyState *fmstate;
1922 : char *query;
1923 : List *target_attrs;
1924 : bool has_returning;
1925 : int values_end_len;
1926 : List *retrieved_attrs;
1927 : RangeTblEntry *rte;
1928 :
1929 : /*
1930 : * Do nothing in EXPLAIN (no ANALYZE) case. resultRelInfo->ri_FdwState
1931 : * stays NULL.
1932 : */
1933 340 : if (eflags & EXEC_FLAG_EXPLAIN_ONLY)
1934 92 : return;
1935 :
1936 : /* Deconstruct fdw_private data. */
1937 248 : query = strVal(list_nth(fdw_private,
1938 : FdwModifyPrivateUpdateSql));
1939 248 : target_attrs = (List *) list_nth(fdw_private,
1940 : FdwModifyPrivateTargetAttnums);
1941 248 : values_end_len = intVal(list_nth(fdw_private,
1942 : FdwModifyPrivateLen));
1943 248 : has_returning = boolVal(list_nth(fdw_private,
1944 : FdwModifyPrivateHasReturning));
1945 248 : retrieved_attrs = (List *) list_nth(fdw_private,
1946 : FdwModifyPrivateRetrievedAttrs);
1947 :
1948 : /* Find RTE. */
1949 248 : rte = exec_rt_fetch(resultRelInfo->ri_RangeTableIndex,
1950 : mtstate->ps.state);
1951 :
1952 : /* Construct an execution state. */
1953 248 : fmstate = create_foreign_modify(mtstate->ps.state,
1954 : rte,
1955 : resultRelInfo,
1956 : mtstate->operation,
1957 248 : outerPlanState(mtstate)->plan,
1958 : query,
1959 : target_attrs,
1960 : values_end_len,
1961 : has_returning,
1962 : retrieved_attrs);
1963 :
1964 248 : resultRelInfo->ri_FdwState = fmstate;
1965 : }
1966 :
1967 : /*
1968 : * postgresExecForeignInsert
1969 : * Insert one row into a foreign table
1970 : */
1971 : static TupleTableSlot *
1972 1784 : postgresExecForeignInsert(EState *estate,
1973 : ResultRelInfo *resultRelInfo,
1974 : TupleTableSlot *slot,
1975 : TupleTableSlot *planSlot)
1976 : {
1977 1784 : PgFdwModifyState *fmstate = (PgFdwModifyState *) resultRelInfo->ri_FdwState;
1978 : TupleTableSlot **rslot;
1979 1784 : int numSlots = 1;
1980 :
1981 : /*
1982 : * If the fmstate has aux_fmstate set, use the aux_fmstate (see
1983 : * postgresBeginForeignInsert())
1984 : */
1985 1784 : if (fmstate->aux_fmstate)
1986 0 : resultRelInfo->ri_FdwState = fmstate->aux_fmstate;
1987 1784 : rslot = execute_foreign_modify(estate, resultRelInfo, CMD_INSERT,
1988 : &slot, &planSlot, &numSlots);
1989 : /* Revert that change */
1990 1776 : if (fmstate->aux_fmstate)
1991 0 : resultRelInfo->ri_FdwState = fmstate;
1992 :
1993 1776 : return rslot ? *rslot : NULL;
1994 : }
1995 :
1996 : /*
1997 : * postgresExecForeignBatchInsert
1998 : * Insert multiple rows into a foreign table
1999 : */
2000 : static TupleTableSlot **
2001 84 : postgresExecForeignBatchInsert(EState *estate,
2002 : ResultRelInfo *resultRelInfo,
2003 : TupleTableSlot **slots,
2004 : TupleTableSlot **planSlots,
2005 : int *numSlots)
2006 : {
2007 84 : PgFdwModifyState *fmstate = (PgFdwModifyState *) resultRelInfo->ri_FdwState;
2008 : TupleTableSlot **rslot;
2009 :
2010 : /*
2011 : * If the fmstate has aux_fmstate set, use the aux_fmstate (see
2012 : * postgresBeginForeignInsert())
2013 : */
2014 84 : if (fmstate->aux_fmstate)
2015 0 : resultRelInfo->ri_FdwState = fmstate->aux_fmstate;
2016 84 : rslot = execute_foreign_modify(estate, resultRelInfo, CMD_INSERT,
2017 : slots, planSlots, numSlots);
2018 : /* Revert that change */
2019 82 : if (fmstate->aux_fmstate)
2020 0 : resultRelInfo->ri_FdwState = fmstate;
2021 :
2022 82 : return rslot;
2023 : }
2024 :
2025 : /*
2026 : * postgresGetForeignModifyBatchSize
2027 : * Determine the maximum number of tuples that can be inserted in bulk
2028 : *
2029 : * Returns the batch size specified for server or table. When batching is not
2030 : * allowed (e.g. for tables with BEFORE/AFTER ROW triggers or with RETURNING
2031 : * clause), returns 1.
2032 : */
2033 : static int
2034 292 : postgresGetForeignModifyBatchSize(ResultRelInfo *resultRelInfo)
2035 : {
2036 : int batch_size;
2037 292 : PgFdwModifyState *fmstate = (PgFdwModifyState *) resultRelInfo->ri_FdwState;
2038 :
2039 : /* should be called only once */
2040 : Assert(resultRelInfo->ri_BatchSize == 0);
2041 :
2042 : /*
2043 : * Should never get called when the insert is being performed on a table
2044 : * that is also among the target relations of an UPDATE operation, because
2045 : * postgresBeginForeignInsert() currently rejects such insert attempts.
2046 : */
2047 : Assert(fmstate == NULL || fmstate->aux_fmstate == NULL);
2048 :
2049 : /*
2050 : * In EXPLAIN without ANALYZE, ri_FdwState is NULL, so we have to lookup
2051 : * the option directly in server/table options. Otherwise just use the
2052 : * value we determined earlier.
2053 : */
2054 292 : if (fmstate)
2055 266 : batch_size = fmstate->batch_size;
2056 : else
2057 26 : batch_size = get_batch_size_option(resultRelInfo->ri_RelationDesc);
2058 :
2059 : /*
2060 : * Disable batching when we have to use RETURNING, there are any
2061 : * BEFORE/AFTER ROW INSERT triggers on the foreign table, or there are any
2062 : * WITH CHECK OPTION constraints from parent views.
2063 : *
2064 : * When there are any BEFORE ROW INSERT triggers on the table, we can't
2065 : * support it, because such triggers might query the table we're inserting
2066 : * into and act differently if the tuples that have already been processed
2067 : * and prepared for insertion are not there.
2068 : */
2069 292 : if (resultRelInfo->ri_projectReturning != NULL ||
2070 250 : resultRelInfo->ri_WithCheckOptions != NIL ||
2071 232 : (resultRelInfo->ri_TrigDesc &&
2072 28 : (resultRelInfo->ri_TrigDesc->trig_insert_before_row ||
2073 2 : resultRelInfo->ri_TrigDesc->trig_insert_after_row)))
2074 88 : return 1;
2075 :
2076 : /*
2077 : * If the foreign table has no columns, disable batching as the INSERT
2078 : * syntax doesn't allow batching multiple empty rows into a zero-column
2079 : * table in a single statement. This is needed for COPY FROM, in which
2080 : * case fmstate must be non-NULL.
2081 : */
2082 204 : if (fmstate && list_length(fmstate->target_attrs) == 0)
2083 2 : return 1;
2084 :
2085 : /*
2086 : * Otherwise use the batch size specified for server/table. The number of
2087 : * parameters in a batch is limited to 65535 (uint16), so make sure we
2088 : * don't exceed this limit by using the maximum batch_size possible.
2089 : */
2090 202 : if (fmstate && fmstate->p_nums > 0)
2091 186 : batch_size = Min(batch_size, PQ_QUERY_PARAM_MAX_LIMIT / fmstate->p_nums);
2092 :
2093 202 : return batch_size;
2094 : }
2095 :
2096 : /*
2097 : * postgresExecForeignUpdate
2098 : * Update one row in a foreign table
2099 : */
2100 : static TupleTableSlot *
2101 190 : postgresExecForeignUpdate(EState *estate,
2102 : ResultRelInfo *resultRelInfo,
2103 : TupleTableSlot *slot,
2104 : TupleTableSlot *planSlot)
2105 : {
2106 : TupleTableSlot **rslot;
2107 190 : int numSlots = 1;
2108 :
2109 190 : rslot = execute_foreign_modify(estate, resultRelInfo, CMD_UPDATE,
2110 : &slot, &planSlot, &numSlots);
2111 :
2112 190 : return rslot ? rslot[0] : NULL;
2113 : }
2114 :
2115 : /*
2116 : * postgresExecForeignDelete
2117 : * Delete one row from a foreign table
2118 : */
2119 : static TupleTableSlot *
2120 46 : postgresExecForeignDelete(EState *estate,
2121 : ResultRelInfo *resultRelInfo,
2122 : TupleTableSlot *slot,
2123 : TupleTableSlot *planSlot)
2124 : {
2125 : TupleTableSlot **rslot;
2126 46 : int numSlots = 1;
2127 :
2128 46 : rslot = execute_foreign_modify(estate, resultRelInfo, CMD_DELETE,
2129 : &slot, &planSlot, &numSlots);
2130 :
2131 46 : return rslot ? rslot[0] : NULL;
2132 : }
2133 :
2134 : /*
2135 : * postgresEndForeignModify
2136 : * Finish an insert/update/delete operation on a foreign table
2137 : */
2138 : static void
2139 312 : postgresEndForeignModify(EState *estate,
2140 : ResultRelInfo *resultRelInfo)
2141 : {
2142 312 : PgFdwModifyState *fmstate = (PgFdwModifyState *) resultRelInfo->ri_FdwState;
2143 :
2144 : /* If fmstate is NULL, we are in EXPLAIN; nothing to do */
2145 312 : if (fmstate == NULL)
2146 92 : return;
2147 :
2148 : /* Destroy the execution state */
2149 220 : finish_foreign_modify(fmstate);
2150 : }
2151 :
2152 : /*
2153 : * postgresBeginForeignInsert
2154 : * Begin an insert operation on a foreign table
2155 : */
2156 : static void
2157 128 : postgresBeginForeignInsert(ModifyTableState *mtstate,
2158 : ResultRelInfo *resultRelInfo)
2159 : {
2160 : PgFdwModifyState *fmstate;
2161 128 : ModifyTable *plan = castNode(ModifyTable, mtstate->ps.plan);
2162 128 : EState *estate = mtstate->ps.state;
2163 : Index resultRelation;
2164 128 : Relation rel = resultRelInfo->ri_RelationDesc;
2165 : RangeTblEntry *rte;
2166 128 : TupleDesc tupdesc = RelationGetDescr(rel);
2167 : int attnum;
2168 : int values_end_len;
2169 : StringInfoData sql;
2170 128 : List *targetAttrs = NIL;
2171 128 : List *retrieved_attrs = NIL;
2172 128 : bool doNothing = false;
2173 :
2174 : /*
2175 : * If the foreign table we are about to insert routed rows into is also an
2176 : * UPDATE subplan result rel that will be updated later, proceeding with
2177 : * the INSERT will result in the later UPDATE incorrectly modifying those
2178 : * routed rows, so prevent the INSERT --- it would be nice if we could
2179 : * handle this case; but for now, throw an error for safety.
2180 : */
2181 128 : if (plan && plan->operation == CMD_UPDATE &&
2182 18 : (resultRelInfo->ri_usesFdwDirectModify ||
2183 10 : resultRelInfo->ri_FdwState))
2184 12 : ereport(ERROR,
2185 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2186 : errmsg("cannot route tuples into foreign table to be updated \"%s\"",
2187 : RelationGetRelationName(rel))));
2188 :
2189 116 : initStringInfo(&sql);
2190 :
2191 : /* We transmit all columns that are defined in the foreign table. */
2192 346 : for (attnum = 1; attnum <= tupdesc->natts; attnum++)
2193 : {
2194 230 : CompactAttribute *attr = TupleDescCompactAttr(tupdesc, attnum - 1);
2195 :
2196 230 : if (!attr->attisdropped)
2197 226 : targetAttrs = lappend_int(targetAttrs, attnum);
2198 : }
2199 :
2200 : /* Check if we add the ON CONFLICT clause to the remote query. */
2201 116 : if (plan)
2202 : {
2203 68 : OnConflictAction onConflictAction = plan->onConflictAction;
2204 :
2205 : /* We only support DO NOTHING without an inference specification. */
2206 68 : if (onConflictAction == ONCONFLICT_NOTHING)
2207 4 : doNothing = true;
2208 64 : else if (onConflictAction != ONCONFLICT_NONE)
2209 0 : elog(ERROR, "unexpected ON CONFLICT specification: %d",
2210 : (int) onConflictAction);
2211 : }
2212 :
2213 : /*
2214 : * If the foreign table is a partition that doesn't have a corresponding
2215 : * RTE entry, we need to create a new RTE describing the foreign table for
2216 : * use by deparseInsertSql and create_foreign_modify() below, after first
2217 : * copying the parent's RTE and modifying some fields to describe the
2218 : * foreign partition to work on. However, if this is invoked by UPDATE,
2219 : * the existing RTE may already correspond to this partition if it is one
2220 : * of the UPDATE subplan target rels; in that case, we can just use the
2221 : * existing RTE as-is.
2222 : */
2223 116 : if (resultRelInfo->ri_RangeTableIndex == 0)
2224 : {
2225 80 : ResultRelInfo *rootResultRelInfo = resultRelInfo->ri_RootResultRelInfo;
2226 :
2227 80 : rte = exec_rt_fetch(rootResultRelInfo->ri_RangeTableIndex, estate);
2228 80 : rte = copyObject(rte);
2229 80 : rte->relid = RelationGetRelid(rel);
2230 80 : rte->relkind = RELKIND_FOREIGN_TABLE;
2231 :
2232 : /*
2233 : * For UPDATE, we must use the RT index of the first subplan target
2234 : * rel's RTE, because the core code would have built expressions for
2235 : * the partition, such as RETURNING, using that RT index as varno of
2236 : * Vars contained in those expressions.
2237 : */
2238 80 : if (plan && plan->operation == CMD_UPDATE &&
2239 6 : rootResultRelInfo->ri_RangeTableIndex == plan->rootRelation)
2240 6 : resultRelation = mtstate->resultRelInfo[0].ri_RangeTableIndex;
2241 : else
2242 74 : resultRelation = rootResultRelInfo->ri_RangeTableIndex;
2243 : }
2244 : else
2245 : {
2246 36 : resultRelation = resultRelInfo->ri_RangeTableIndex;
2247 36 : rte = exec_rt_fetch(resultRelation, estate);
2248 : }
2249 :
2250 : /* Construct the SQL command string. */
2251 116 : deparseInsertSql(&sql, rte, resultRelation, rel, targetAttrs, doNothing,
2252 : resultRelInfo->ri_WithCheckOptions,
2253 : resultRelInfo->ri_returningList,
2254 : &retrieved_attrs, &values_end_len);
2255 :
2256 : /* Construct an execution state. */
2257 116 : fmstate = create_foreign_modify(mtstate->ps.state,
2258 : rte,
2259 : resultRelInfo,
2260 : CMD_INSERT,
2261 : NULL,
2262 : sql.data,
2263 : targetAttrs,
2264 : values_end_len,
2265 : retrieved_attrs != NIL,
2266 : retrieved_attrs);
2267 :
2268 : /*
2269 : * If the given resultRelInfo already has PgFdwModifyState set, it means
2270 : * the foreign table is an UPDATE subplan result rel; in which case, store
2271 : * the resulting state into the aux_fmstate of the PgFdwModifyState.
2272 : */
2273 116 : if (resultRelInfo->ri_FdwState)
2274 : {
2275 : Assert(plan && plan->operation == CMD_UPDATE);
2276 : Assert(resultRelInfo->ri_usesFdwDirectModify == false);
2277 0 : ((PgFdwModifyState *) resultRelInfo->ri_FdwState)->aux_fmstate = fmstate;
2278 : }
2279 : else
2280 116 : resultRelInfo->ri_FdwState = fmstate;
2281 116 : }
2282 :
2283 : /*
2284 : * postgresEndForeignInsert
2285 : * Finish an insert operation on a foreign table
2286 : */
2287 : static void
2288 100 : postgresEndForeignInsert(EState *estate,
2289 : ResultRelInfo *resultRelInfo)
2290 : {
2291 100 : PgFdwModifyState *fmstate = (PgFdwModifyState *) resultRelInfo->ri_FdwState;
2292 :
2293 : Assert(fmstate != NULL);
2294 :
2295 : /*
2296 : * If the fmstate has aux_fmstate set, get the aux_fmstate (see
2297 : * postgresBeginForeignInsert())
2298 : */
2299 100 : if (fmstate->aux_fmstate)
2300 0 : fmstate = fmstate->aux_fmstate;
2301 :
2302 : /* Destroy the execution state */
2303 100 : finish_foreign_modify(fmstate);
2304 100 : }
2305 :
2306 : /*
2307 : * postgresIsForeignRelUpdatable
2308 : * Determine whether a foreign table supports INSERT, UPDATE and/or
2309 : * DELETE.
2310 : */
2311 : static int
2312 676 : postgresIsForeignRelUpdatable(Relation rel)
2313 : {
2314 : bool updatable;
2315 : ForeignTable *table;
2316 : ForeignServer *server;
2317 : ListCell *lc;
2318 :
2319 : /*
2320 : * By default, all postgres_fdw foreign tables are assumed updatable. This
2321 : * can be overridden by a per-server setting, which in turn can be
2322 : * overridden by a per-table setting.
2323 : */
2324 676 : updatable = true;
2325 :
2326 676 : table = GetForeignTable(RelationGetRelid(rel));
2327 676 : server = GetForeignServer(table->serverid);
2328 :
2329 3022 : foreach(lc, server->options)
2330 : {
2331 2346 : DefElem *def = (DefElem *) lfirst(lc);
2332 :
2333 2346 : if (strcmp(def->defname, "updatable") == 0)
2334 0 : updatable = defGetBoolean(def);
2335 : }
2336 1628 : foreach(lc, table->options)
2337 : {
2338 952 : DefElem *def = (DefElem *) lfirst(lc);
2339 :
2340 952 : if (strcmp(def->defname, "updatable") == 0)
2341 0 : updatable = defGetBoolean(def);
2342 : }
2343 :
2344 : /*
2345 : * Currently "updatable" means support for INSERT, UPDATE and DELETE.
2346 : */
2347 : return updatable ?
2348 676 : (1 << CMD_INSERT) | (1 << CMD_UPDATE) | (1 << CMD_DELETE) : 0;
2349 : }
2350 :
2351 : /*
2352 : * postgresRecheckForeignScan
2353 : * Execute a local join execution plan for a foreign join
2354 : */
2355 : static bool
2356 0 : postgresRecheckForeignScan(ForeignScanState *node, TupleTableSlot *slot)
2357 : {
2358 0 : Index scanrelid = ((Scan *) node->ss.ps.plan)->scanrelid;
2359 0 : PlanState *outerPlan = outerPlanState(node);
2360 : TupleTableSlot *result;
2361 :
2362 : /* For base foreign relations, it suffices to set fdw_recheck_quals */
2363 0 : if (scanrelid > 0)
2364 0 : return true;
2365 :
2366 : Assert(outerPlan != NULL);
2367 :
2368 : /* Execute a local join execution plan */
2369 0 : result = ExecProcNode(outerPlan);
2370 0 : if (TupIsNull(result))
2371 0 : return false;
2372 :
2373 : /* Store result in the given slot */
2374 0 : ExecCopySlot(slot, result);
2375 :
2376 0 : return true;
2377 : }
2378 :
2379 : /*
2380 : * find_modifytable_subplan
2381 : * Helper routine for postgresPlanDirectModify to find the
2382 : * ModifyTable subplan node that scans the specified RTI.
2383 : *
2384 : * Returns NULL if the subplan couldn't be identified. That's not a fatal
2385 : * error condition, we just abandon trying to do the update directly.
2386 : */
2387 : static ForeignScan *
2388 262 : find_modifytable_subplan(PlannerInfo *root,
2389 : ModifyTable *plan,
2390 : Index rtindex,
2391 : int subplan_index)
2392 : {
2393 262 : Plan *subplan = outerPlan(plan);
2394 :
2395 : /*
2396 : * The cases we support are (1) the desired ForeignScan is the immediate
2397 : * child of ModifyTable, or (2) it is the subplan_index'th child of an
2398 : * Append node that is the immediate child of ModifyTable. There is no
2399 : * point in looking further down, as that would mean that local joins are
2400 : * involved, so we can't do the update directly.
2401 : *
2402 : * There could be a Result atop the Append too, acting to compute the
2403 : * UPDATE targetlist values. We ignore that here; the tlist will be
2404 : * checked by our caller.
2405 : *
2406 : * In principle we could examine all the children of the Append, but it's
2407 : * currently unlikely that the core planner would generate such a plan
2408 : * with the children out-of-order. Moreover, such a search risks costing
2409 : * O(N^2) time when there are a lot of children.
2410 : */
2411 262 : if (IsA(subplan, Append))
2412 : {
2413 66 : Append *appendplan = (Append *) subplan;
2414 :
2415 66 : if (subplan_index < list_length(appendplan->appendplans))
2416 66 : subplan = (Plan *) list_nth(appendplan->appendplans, subplan_index);
2417 : }
2418 196 : else if (IsA(subplan, Result) &&
2419 12 : outerPlan(subplan) != NULL &&
2420 10 : IsA(outerPlan(subplan), Append))
2421 : {
2422 10 : Append *appendplan = (Append *) outerPlan(subplan);
2423 :
2424 10 : if (subplan_index < list_length(appendplan->appendplans))
2425 10 : subplan = (Plan *) list_nth(appendplan->appendplans, subplan_index);
2426 : }
2427 :
2428 : /* Now, have we got a ForeignScan on the desired rel? */
2429 262 : if (IsA(subplan, ForeignScan))
2430 : {
2431 228 : ForeignScan *fscan = (ForeignScan *) subplan;
2432 :
2433 228 : if (bms_is_member(rtindex, fscan->fs_base_relids))
2434 228 : return fscan;
2435 : }
2436 :
2437 34 : return NULL;
2438 : }
2439 :
2440 : /*
2441 : * postgresPlanDirectModify
2442 : * Consider a direct foreign table modification
2443 : *
2444 : * Decide whether it is safe to modify a foreign table directly, and if so,
2445 : * rewrite subplan accordingly.
2446 : */
2447 : static bool
2448 390 : postgresPlanDirectModify(PlannerInfo *root,
2449 : ModifyTable *plan,
2450 : Index resultRelation,
2451 : int subplan_index)
2452 : {
2453 390 : CmdType operation = plan->operation;
2454 : RelOptInfo *foreignrel;
2455 : RangeTblEntry *rte;
2456 : PgFdwRelationInfo *fpinfo;
2457 : Relation rel;
2458 : StringInfoData sql;
2459 : ForeignScan *fscan;
2460 390 : List *processed_tlist = NIL;
2461 390 : List *targetAttrs = NIL;
2462 : List *remote_exprs;
2463 390 : List *params_list = NIL;
2464 390 : List *returningList = NIL;
2465 390 : List *retrieved_attrs = NIL;
2466 :
2467 : /*
2468 : * Decide whether it is safe to modify a foreign table directly.
2469 : */
2470 :
2471 : /*
2472 : * The table modification must be an UPDATE or DELETE.
2473 : */
2474 390 : if (operation != CMD_UPDATE && operation != CMD_DELETE)
2475 128 : return false;
2476 :
2477 : /*
2478 : * Try to locate the ForeignScan subplan that's scanning resultRelation.
2479 : */
2480 262 : fscan = find_modifytable_subplan(root, plan, resultRelation, subplan_index);
2481 262 : if (!fscan)
2482 34 : return false;
2483 :
2484 : /*
2485 : * It's unsafe to modify a foreign table directly if there are any quals
2486 : * that should be evaluated locally.
2487 : */
2488 228 : if (fscan->scan.plan.qual != NIL)
2489 10 : return false;
2490 :
2491 : /* Safe to fetch data about the target foreign rel */
2492 218 : if (fscan->scan.scanrelid == 0)
2493 : {
2494 20 : foreignrel = find_join_rel(root, fscan->fs_relids);
2495 : /* We should have a rel for this foreign join. */
2496 : Assert(foreignrel);
2497 : }
2498 : else
2499 198 : foreignrel = root->simple_rel_array[resultRelation];
2500 218 : rte = root->simple_rte_array[resultRelation];
2501 218 : fpinfo = (PgFdwRelationInfo *) foreignrel->fdw_private;
2502 :
2503 : /*
2504 : * It's unsafe to update a foreign table directly, if any expressions to
2505 : * assign to the target columns are unsafe to evaluate remotely.
2506 : */
2507 218 : if (operation == CMD_UPDATE)
2508 : {
2509 : ListCell *lc,
2510 : *lc2;
2511 :
2512 : /*
2513 : * The expressions of concern are the first N columns of the processed
2514 : * targetlist, where N is the length of the rel's update_colnos.
2515 : */
2516 100 : get_translated_update_targetlist(root, resultRelation,
2517 : &processed_tlist, &targetAttrs);
2518 206 : forboth(lc, processed_tlist, lc2, targetAttrs)
2519 : {
2520 116 : TargetEntry *tle = lfirst_node(TargetEntry, lc);
2521 116 : AttrNumber attno = lfirst_int(lc2);
2522 :
2523 : /* update's new-value expressions shouldn't be resjunk */
2524 : Assert(!tle->resjunk);
2525 :
2526 116 : if (attno <= InvalidAttrNumber) /* shouldn't happen */
2527 0 : elog(ERROR, "system-column update is not supported");
2528 :
2529 116 : if (!is_foreign_expr(root, foreignrel, (Expr *) tle->expr))
2530 10 : return false;
2531 : }
2532 : }
2533 :
2534 : /*
2535 : * Ok, rewrite subplan so as to modify the foreign table directly.
2536 : */
2537 208 : initStringInfo(&sql);
2538 :
2539 : /*
2540 : * Core code already has some lock on each rel being planned, so we can
2541 : * use NoLock here.
2542 : */
2543 208 : rel = table_open(rte->relid, NoLock);
2544 :
2545 : /*
2546 : * Recall the qual clauses that must be evaluated remotely. (These are
2547 : * bare clauses not RestrictInfos, but deparse.c's appendConditions()
2548 : * doesn't care.)
2549 : */
2550 208 : remote_exprs = fpinfo->final_remote_exprs;
2551 :
2552 : /*
2553 : * Extract the relevant RETURNING list if any.
2554 : */
2555 208 : if (plan->returningLists)
2556 : {
2557 70 : returningList = (List *) list_nth(plan->returningLists, subplan_index);
2558 :
2559 : /*
2560 : * When performing an UPDATE/DELETE .. RETURNING on a join directly,
2561 : * we fetch from the foreign server any Vars specified in RETURNING
2562 : * that refer not only to the target relation but to non-target
2563 : * relations. So we'll deparse them into the RETURNING clause of the
2564 : * remote query; use a targetlist consisting of them instead, which
2565 : * will be adjusted to be new fdw_scan_tlist of the foreign-scan plan
2566 : * node below.
2567 : */
2568 70 : if (fscan->scan.scanrelid == 0)
2569 8 : returningList = build_remote_returning(resultRelation, rel,
2570 : returningList);
2571 : }
2572 :
2573 : /*
2574 : * Construct the SQL command string.
2575 : */
2576 208 : switch (operation)
2577 : {
2578 90 : case CMD_UPDATE:
2579 90 : deparseDirectUpdateSql(&sql, root, resultRelation, rel,
2580 : foreignrel,
2581 : processed_tlist,
2582 : targetAttrs,
2583 : remote_exprs, ¶ms_list,
2584 : returningList, &retrieved_attrs);
2585 90 : break;
2586 118 : case CMD_DELETE:
2587 118 : deparseDirectDeleteSql(&sql, root, resultRelation, rel,
2588 : foreignrel,
2589 : remote_exprs, ¶ms_list,
2590 : returningList, &retrieved_attrs);
2591 118 : break;
2592 0 : default:
2593 0 : elog(ERROR, "unexpected operation: %d", (int) operation);
2594 : break;
2595 : }
2596 :
2597 : /*
2598 : * Update the operation and target relation info.
2599 : */
2600 208 : fscan->operation = operation;
2601 208 : fscan->resultRelation = resultRelation;
2602 :
2603 : /*
2604 : * Update the fdw_exprs list that will be available to the executor.
2605 : */
2606 208 : fscan->fdw_exprs = params_list;
2607 :
2608 : /*
2609 : * Update the fdw_private list that will be available to the executor.
2610 : * Items in the list must match enum FdwDirectModifyPrivateIndex, above.
2611 : */
2612 208 : fscan->fdw_private = list_make4(makeString(sql.data),
2613 : makeBoolean((retrieved_attrs != NIL)),
2614 : retrieved_attrs,
2615 : makeBoolean(plan->canSetTag));
2616 :
2617 : /*
2618 : * Update the foreign-join-related fields.
2619 : */
2620 208 : if (fscan->scan.scanrelid == 0)
2621 : {
2622 : /* No need for the outer subplan. */
2623 16 : fscan->scan.plan.lefttree = NULL;
2624 :
2625 : /* Build new fdw_scan_tlist if UPDATE/DELETE .. RETURNING. */
2626 16 : if (returningList)
2627 4 : rebuild_fdw_scan_tlist(fscan, returningList);
2628 : }
2629 :
2630 : /*
2631 : * Finally, unset the async-capable flag if it is set, as we currently
2632 : * don't support asynchronous execution of direct modifications.
2633 : */
2634 208 : if (fscan->scan.plan.async_capable)
2635 16 : fscan->scan.plan.async_capable = false;
2636 :
2637 208 : table_close(rel, NoLock);
2638 208 : return true;
2639 : }
2640 :
2641 : /*
2642 : * postgresBeginDirectModify
2643 : * Prepare a direct foreign table modification
2644 : */
2645 : static void
2646 208 : postgresBeginDirectModify(ForeignScanState *node, int eflags)
2647 : {
2648 208 : ForeignScan *fsplan = (ForeignScan *) node->ss.ps.plan;
2649 208 : EState *estate = node->ss.ps.state;
2650 : PgFdwDirectModifyState *dmstate;
2651 : Index rtindex;
2652 : Oid userid;
2653 : ForeignTable *table;
2654 : UserMapping *user;
2655 : int numParams;
2656 :
2657 : /*
2658 : * Do nothing in EXPLAIN (no ANALYZE) case. node->fdw_state stays NULL.
2659 : */
2660 208 : if (eflags & EXEC_FLAG_EXPLAIN_ONLY)
2661 64 : return;
2662 :
2663 : /*
2664 : * We'll save private state in node->fdw_state.
2665 : */
2666 144 : dmstate = (PgFdwDirectModifyState *) palloc0(sizeof(PgFdwDirectModifyState));
2667 144 : node->fdw_state = dmstate;
2668 :
2669 : /*
2670 : * Identify which user to do the remote access as. This should match what
2671 : * ExecCheckPermissions() does.
2672 : */
2673 144 : userid = OidIsValid(fsplan->checkAsUser) ? fsplan->checkAsUser : GetUserId();
2674 :
2675 : /* Get info about foreign table. */
2676 144 : rtindex = node->resultRelInfo->ri_RangeTableIndex;
2677 144 : if (fsplan->scan.scanrelid == 0)
2678 8 : dmstate->rel = ExecOpenScanRelation(estate, rtindex, eflags);
2679 : else
2680 136 : dmstate->rel = node->ss.ss_currentRelation;
2681 144 : table = GetForeignTable(RelationGetRelid(dmstate->rel));
2682 144 : user = GetUserMapping(userid, table->serverid);
2683 :
2684 : /*
2685 : * Get connection to the foreign server. Connection manager will
2686 : * establish new connection if necessary.
2687 : */
2688 144 : dmstate->conn = GetConnection(user, false, &dmstate->conn_state);
2689 :
2690 : /* Update the foreign-join-related fields. */
2691 144 : if (fsplan->scan.scanrelid == 0)
2692 : {
2693 : /* Save info about foreign table. */
2694 8 : dmstate->resultRel = dmstate->rel;
2695 :
2696 : /*
2697 : * Set dmstate->rel to NULL to teach get_returning_data() and
2698 : * make_tuple_from_result_row() that columns fetched from the remote
2699 : * server are described by fdw_scan_tlist of the foreign-scan plan
2700 : * node, not the tuple descriptor for the target relation.
2701 : */
2702 8 : dmstate->rel = NULL;
2703 : }
2704 :
2705 : /* Initialize state variable */
2706 144 : dmstate->num_tuples = -1; /* -1 means not set yet */
2707 :
2708 : /* Get private info created by planner functions. */
2709 144 : dmstate->query = strVal(list_nth(fsplan->fdw_private,
2710 : FdwDirectModifyPrivateUpdateSql));
2711 144 : dmstate->has_returning = boolVal(list_nth(fsplan->fdw_private,
2712 : FdwDirectModifyPrivateHasReturning));
2713 144 : dmstate->retrieved_attrs = (List *) list_nth(fsplan->fdw_private,
2714 : FdwDirectModifyPrivateRetrievedAttrs);
2715 144 : dmstate->set_processed = boolVal(list_nth(fsplan->fdw_private,
2716 : FdwDirectModifyPrivateSetProcessed));
2717 :
2718 : /* Create context for per-tuple temp workspace. */
2719 144 : dmstate->temp_cxt = AllocSetContextCreate(estate->es_query_cxt,
2720 : "postgres_fdw temporary data",
2721 : ALLOCSET_SMALL_SIZES);
2722 :
2723 : /* Prepare for input conversion of RETURNING results. */
2724 144 : if (dmstate->has_returning)
2725 : {
2726 : TupleDesc tupdesc;
2727 :
2728 32 : if (fsplan->scan.scanrelid == 0)
2729 2 : tupdesc = get_tupdesc_for_join_scan_tuples(node);
2730 : else
2731 30 : tupdesc = RelationGetDescr(dmstate->rel);
2732 :
2733 32 : dmstate->attinmeta = TupleDescGetAttInMetadata(tupdesc);
2734 :
2735 : /*
2736 : * When performing an UPDATE/DELETE .. RETURNING on a join directly,
2737 : * initialize a filter to extract an updated/deleted tuple from a scan
2738 : * tuple.
2739 : */
2740 32 : if (fsplan->scan.scanrelid == 0)
2741 2 : init_returning_filter(dmstate, fsplan->fdw_scan_tlist, rtindex);
2742 : }
2743 :
2744 : /*
2745 : * Prepare for processing of parameters used in remote query, if any.
2746 : */
2747 144 : numParams = list_length(fsplan->fdw_exprs);
2748 144 : dmstate->numParams = numParams;
2749 144 : if (numParams > 0)
2750 0 : prepare_query_params((PlanState *) node,
2751 : fsplan->fdw_exprs,
2752 : numParams,
2753 : &dmstate->param_flinfo,
2754 : &dmstate->param_exprs,
2755 : &dmstate->param_values);
2756 : }
2757 :
2758 : /*
2759 : * postgresIterateDirectModify
2760 : * Execute a direct foreign table modification
2761 : */
2762 : static TupleTableSlot *
2763 836 : postgresIterateDirectModify(ForeignScanState *node)
2764 : {
2765 836 : PgFdwDirectModifyState *dmstate = (PgFdwDirectModifyState *) node->fdw_state;
2766 836 : EState *estate = node->ss.ps.state;
2767 836 : ResultRelInfo *resultRelInfo = node->resultRelInfo;
2768 :
2769 : /*
2770 : * If this is the first call after Begin, execute the statement.
2771 : */
2772 836 : if (dmstate->num_tuples == -1)
2773 142 : execute_dml_stmt(node);
2774 :
2775 : /*
2776 : * If the local query doesn't specify RETURNING, just clear tuple slot.
2777 : */
2778 828 : if (!resultRelInfo->ri_projectReturning)
2779 : {
2780 100 : TupleTableSlot *slot = node->ss.ss_ScanTupleSlot;
2781 100 : Instrumentation *instr = node->ss.ps.instrument;
2782 :
2783 : Assert(!dmstate->has_returning);
2784 :
2785 : /* Increment the command es_processed count if necessary. */
2786 100 : if (dmstate->set_processed)
2787 100 : estate->es_processed += dmstate->num_tuples;
2788 :
2789 : /* Increment the tuple count for EXPLAIN ANALYZE if necessary. */
2790 100 : if (instr)
2791 0 : instr->tuplecount += dmstate->num_tuples;
2792 :
2793 100 : return ExecClearTuple(slot);
2794 : }
2795 :
2796 : /*
2797 : * Get the next RETURNING tuple.
2798 : */
2799 728 : return get_returning_data(node);
2800 : }
2801 :
2802 : /*
2803 : * postgresEndDirectModify
2804 : * Finish a direct foreign table modification
2805 : */
2806 : static void
2807 192 : postgresEndDirectModify(ForeignScanState *node)
2808 : {
2809 192 : PgFdwDirectModifyState *dmstate = (PgFdwDirectModifyState *) node->fdw_state;
2810 :
2811 : /* if dmstate is NULL, we are in EXPLAIN; nothing to do */
2812 192 : if (dmstate == NULL)
2813 64 : return;
2814 :
2815 : /* Release PGresult */
2816 128 : PQclear(dmstate->result);
2817 :
2818 : /* Release remote connection */
2819 128 : ReleaseConnection(dmstate->conn);
2820 128 : dmstate->conn = NULL;
2821 :
2822 : /* MemoryContext will be deleted automatically. */
2823 : }
2824 :
2825 : /*
2826 : * postgresExplainForeignScan
2827 : * Produce extra output for EXPLAIN of a ForeignScan on a foreign table
2828 : */
2829 : static void
2830 780 : postgresExplainForeignScan(ForeignScanState *node, ExplainState *es)
2831 : {
2832 780 : ForeignScan *plan = castNode(ForeignScan, node->ss.ps.plan);
2833 780 : List *fdw_private = plan->fdw_private;
2834 :
2835 : /*
2836 : * Identify foreign scans that are really joins or upper relations. The
2837 : * input looks something like "(1) LEFT JOIN (2)", and we must replace the
2838 : * digit string(s), which are RT indexes, with the correct relation names.
2839 : * We do that here, not when the plan is created, because we can't know
2840 : * what aliases ruleutils.c will assign at plan creation time.
2841 : */
2842 780 : if (list_length(fdw_private) > FdwScanPrivateRelations)
2843 : {
2844 : StringInfo relations;
2845 : char *rawrelations;
2846 : char *ptr;
2847 : int minrti,
2848 : rtoffset;
2849 :
2850 242 : rawrelations = strVal(list_nth(fdw_private, FdwScanPrivateRelations));
2851 :
2852 : /*
2853 : * A difficulty with using a string representation of RT indexes is
2854 : * that setrefs.c won't update the string when flattening the
2855 : * rangetable. To find out what rtoffset was applied, identify the
2856 : * minimum RT index appearing in the string and compare it to the
2857 : * minimum member of plan->fs_base_relids. (We expect all the relids
2858 : * in the join will have been offset by the same amount; the Asserts
2859 : * below should catch it if that ever changes.)
2860 : */
2861 242 : minrti = INT_MAX;
2862 242 : ptr = rawrelations;
2863 5754 : while (*ptr)
2864 : {
2865 5512 : if (isdigit((unsigned char) *ptr))
2866 : {
2867 478 : int rti = strtol(ptr, &ptr, 10);
2868 :
2869 478 : if (rti < minrti)
2870 266 : minrti = rti;
2871 : }
2872 : else
2873 5034 : ptr++;
2874 : }
2875 242 : rtoffset = bms_next_member(plan->fs_base_relids, -1) - minrti;
2876 :
2877 : /* Now we can translate the string */
2878 242 : relations = makeStringInfo();
2879 242 : ptr = rawrelations;
2880 5754 : while (*ptr)
2881 : {
2882 5512 : if (isdigit((unsigned char) *ptr))
2883 : {
2884 478 : int rti = strtol(ptr, &ptr, 10);
2885 : RangeTblEntry *rte;
2886 : char *relname;
2887 : char *refname;
2888 :
2889 478 : rti += rtoffset;
2890 : Assert(bms_is_member(rti, plan->fs_base_relids));
2891 478 : rte = rt_fetch(rti, es->rtable);
2892 : Assert(rte->rtekind == RTE_RELATION);
2893 : /* This logic should agree with explain.c's ExplainTargetRel */
2894 478 : relname = get_rel_name(rte->relid);
2895 478 : if (es->verbose)
2896 : {
2897 : char *namespace;
2898 :
2899 452 : namespace = get_namespace_name_or_temp(get_rel_namespace(rte->relid));
2900 452 : appendStringInfo(relations, "%s.%s",
2901 : quote_identifier(namespace),
2902 : quote_identifier(relname));
2903 : }
2904 : else
2905 26 : appendStringInfoString(relations,
2906 : quote_identifier(relname));
2907 478 : refname = (char *) list_nth(es->rtable_names, rti - 1);
2908 478 : if (refname == NULL)
2909 0 : refname = rte->eref->aliasname;
2910 478 : if (strcmp(refname, relname) != 0)
2911 298 : appendStringInfo(relations, " %s",
2912 : quote_identifier(refname));
2913 : }
2914 : else
2915 5034 : appendStringInfoChar(relations, *ptr++);
2916 : }
2917 242 : ExplainPropertyText("Relations", relations->data, es);
2918 : }
2919 :
2920 : /*
2921 : * Add remote query, when VERBOSE option is specified.
2922 : */
2923 780 : if (es->verbose)
2924 : {
2925 : char *sql;
2926 :
2927 708 : sql = strVal(list_nth(fdw_private, FdwScanPrivateSelectSql));
2928 708 : ExplainPropertyText("Remote SQL", sql, es);
2929 : }
2930 780 : }
2931 :
2932 : /*
2933 : * postgresExplainForeignModify
2934 : * Produce extra output for EXPLAIN of a ModifyTable on a foreign table
2935 : */
2936 : static void
2937 92 : postgresExplainForeignModify(ModifyTableState *mtstate,
2938 : ResultRelInfo *rinfo,
2939 : List *fdw_private,
2940 : int subplan_index,
2941 : ExplainState *es)
2942 : {
2943 92 : if (es->verbose)
2944 : {
2945 92 : char *sql = strVal(list_nth(fdw_private,
2946 : FdwModifyPrivateUpdateSql));
2947 :
2948 92 : ExplainPropertyText("Remote SQL", sql, es);
2949 :
2950 : /*
2951 : * For INSERT we should always have batch size >= 1, but UPDATE and
2952 : * DELETE don't support batching so don't show the property.
2953 : */
2954 92 : if (rinfo->ri_BatchSize > 0)
2955 26 : ExplainPropertyInteger("Batch Size", NULL, rinfo->ri_BatchSize, es);
2956 : }
2957 92 : }
2958 :
2959 : /*
2960 : * postgresExplainDirectModify
2961 : * Produce extra output for EXPLAIN of a ForeignScan that modifies a
2962 : * foreign table directly
2963 : */
2964 : static void
2965 64 : postgresExplainDirectModify(ForeignScanState *node, ExplainState *es)
2966 : {
2967 : List *fdw_private;
2968 : char *sql;
2969 :
2970 64 : if (es->verbose)
2971 : {
2972 64 : fdw_private = ((ForeignScan *) node->ss.ps.plan)->fdw_private;
2973 64 : sql = strVal(list_nth(fdw_private, FdwDirectModifyPrivateUpdateSql));
2974 64 : ExplainPropertyText("Remote SQL", sql, es);
2975 : }
2976 64 : }
2977 :
2978 : /*
2979 : * postgresExecForeignTruncate
2980 : * Truncate one or more foreign tables
2981 : */
2982 : static void
2983 30 : postgresExecForeignTruncate(List *rels,
2984 : DropBehavior behavior,
2985 : bool restart_seqs)
2986 : {
2987 30 : Oid serverid = InvalidOid;
2988 30 : UserMapping *user = NULL;
2989 30 : PGconn *conn = NULL;
2990 : StringInfoData sql;
2991 : ListCell *lc;
2992 30 : bool server_truncatable = true;
2993 :
2994 : /*
2995 : * By default, all postgres_fdw foreign tables are assumed truncatable.
2996 : * This can be overridden by a per-server setting, which in turn can be
2997 : * overridden by a per-table setting.
2998 : */
2999 58 : foreach(lc, rels)
3000 : {
3001 34 : ForeignServer *server = NULL;
3002 34 : Relation rel = lfirst(lc);
3003 34 : ForeignTable *table = GetForeignTable(RelationGetRelid(rel));
3004 : ListCell *cell;
3005 : bool truncatable;
3006 :
3007 : /*
3008 : * First time through, determine whether the foreign server allows
3009 : * truncates. Since all specified foreign tables are assumed to belong
3010 : * to the same foreign server, this result can be used for other
3011 : * foreign tables.
3012 : */
3013 34 : if (!OidIsValid(serverid))
3014 : {
3015 30 : serverid = table->serverid;
3016 30 : server = GetForeignServer(serverid);
3017 :
3018 120 : foreach(cell, server->options)
3019 : {
3020 96 : DefElem *defel = (DefElem *) lfirst(cell);
3021 :
3022 96 : if (strcmp(defel->defname, "truncatable") == 0)
3023 : {
3024 6 : server_truncatable = defGetBoolean(defel);
3025 6 : break;
3026 : }
3027 : }
3028 : }
3029 :
3030 : /*
3031 : * Confirm that all specified foreign tables belong to the same
3032 : * foreign server.
3033 : */
3034 : Assert(table->serverid == serverid);
3035 :
3036 : /* Determine whether this foreign table allows truncations */
3037 34 : truncatable = server_truncatable;
3038 68 : foreach(cell, table->options)
3039 : {
3040 48 : DefElem *defel = (DefElem *) lfirst(cell);
3041 :
3042 48 : if (strcmp(defel->defname, "truncatable") == 0)
3043 : {
3044 14 : truncatable = defGetBoolean(defel);
3045 14 : break;
3046 : }
3047 : }
3048 :
3049 34 : if (!truncatable)
3050 6 : ereport(ERROR,
3051 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3052 : errmsg("foreign table \"%s\" does not allow truncates",
3053 : RelationGetRelationName(rel))));
3054 : }
3055 : Assert(OidIsValid(serverid));
3056 :
3057 : /*
3058 : * Get connection to the foreign server. Connection manager will
3059 : * establish new connection if necessary.
3060 : */
3061 24 : user = GetUserMapping(GetUserId(), serverid);
3062 24 : conn = GetConnection(user, false, NULL);
3063 :
3064 : /* Construct the TRUNCATE command string */
3065 24 : initStringInfo(&sql);
3066 24 : deparseTruncateSql(&sql, rels, behavior, restart_seqs);
3067 :
3068 : /* Issue the TRUNCATE command to remote server */
3069 24 : do_sql_command(conn, sql.data);
3070 :
3071 22 : pfree(sql.data);
3072 22 : }
3073 :
3074 : /*
3075 : * estimate_path_cost_size
3076 : * Get cost and size estimates for a foreign scan on given foreign relation
3077 : * either a base relation or a join between foreign relations or an upper
3078 : * relation containing foreign relations.
3079 : *
3080 : * param_join_conds are the parameterization clauses with outer relations.
3081 : * pathkeys specify the expected sort order if any for given path being costed.
3082 : * fpextra specifies additional post-scan/join-processing steps such as the
3083 : * final sort and the LIMIT restriction.
3084 : *
3085 : * The function returns the cost and size estimates in p_rows, p_width,
3086 : * p_disabled_nodes, p_startup_cost and p_total_cost variables.
3087 : */
3088 : static void
3089 5366 : estimate_path_cost_size(PlannerInfo *root,
3090 : RelOptInfo *foreignrel,
3091 : List *param_join_conds,
3092 : List *pathkeys,
3093 : PgFdwPathExtraData *fpextra,
3094 : double *p_rows, int *p_width,
3095 : int *p_disabled_nodes,
3096 : Cost *p_startup_cost, Cost *p_total_cost)
3097 : {
3098 5366 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) foreignrel->fdw_private;
3099 : double rows;
3100 : double retrieved_rows;
3101 : int width;
3102 5366 : int disabled_nodes = 0;
3103 : Cost startup_cost;
3104 : Cost total_cost;
3105 :
3106 : /* Make sure the core code has set up the relation's reltarget */
3107 : Assert(foreignrel->reltarget);
3108 :
3109 : /*
3110 : * If the table or the server is configured to use remote estimates,
3111 : * connect to the foreign server and execute EXPLAIN to estimate the
3112 : * number of rows selected by the restriction+join clauses. Otherwise,
3113 : * estimate rows using whatever statistics we have locally, in a way
3114 : * similar to ordinary tables.
3115 : */
3116 5366 : if (fpinfo->use_remote_estimate)
3117 : {
3118 : List *remote_param_join_conds;
3119 : List *local_param_join_conds;
3120 : StringInfoData sql;
3121 : PGconn *conn;
3122 : Selectivity local_sel;
3123 : QualCost local_cost;
3124 2570 : List *fdw_scan_tlist = NIL;
3125 : List *remote_conds;
3126 :
3127 : /* Required only to be passed to deparseSelectStmtForRel */
3128 : List *retrieved_attrs;
3129 :
3130 : /*
3131 : * param_join_conds might contain both clauses that are safe to send
3132 : * across, and clauses that aren't.
3133 : */
3134 2570 : classifyConditions(root, foreignrel, param_join_conds,
3135 : &remote_param_join_conds, &local_param_join_conds);
3136 :
3137 : /* Build the list of columns to be fetched from the foreign server. */
3138 2570 : if (IS_JOIN_REL(foreignrel) || IS_UPPER_REL(foreignrel))
3139 1042 : fdw_scan_tlist = build_tlist_to_deparse(foreignrel);
3140 : else
3141 1528 : fdw_scan_tlist = NIL;
3142 :
3143 : /*
3144 : * The complete list of remote conditions includes everything from
3145 : * baserestrictinfo plus any extra join_conds relevant to this
3146 : * particular path.
3147 : */
3148 2570 : remote_conds = list_concat(remote_param_join_conds,
3149 2570 : fpinfo->remote_conds);
3150 :
3151 : /*
3152 : * Construct EXPLAIN query including the desired SELECT, FROM, and
3153 : * WHERE clauses. Params and other-relation Vars are replaced by dummy
3154 : * values, so don't request params_list.
3155 : */
3156 2570 : initStringInfo(&sql);
3157 2570 : appendStringInfoString(&sql, "EXPLAIN ");
3158 2570 : deparseSelectStmtForRel(&sql, root, foreignrel, fdw_scan_tlist,
3159 : remote_conds, pathkeys,
3160 2570 : fpextra ? fpextra->has_final_sort : false,
3161 2570 : fpextra ? fpextra->has_limit : false,
3162 : false, &retrieved_attrs, NULL);
3163 :
3164 : /* Get the remote estimate */
3165 2570 : conn = GetConnection(fpinfo->user, false, NULL);
3166 2570 : get_remote_estimate(sql.data, conn, &rows, &width,
3167 : &startup_cost, &total_cost);
3168 2570 : ReleaseConnection(conn);
3169 :
3170 2570 : retrieved_rows = rows;
3171 :
3172 : /* Factor in the selectivity of the locally-checked quals */
3173 2570 : local_sel = clauselist_selectivity(root,
3174 : local_param_join_conds,
3175 2570 : foreignrel->relid,
3176 : JOIN_INNER,
3177 : NULL);
3178 2570 : local_sel *= fpinfo->local_conds_sel;
3179 :
3180 2570 : rows = clamp_row_est(rows * local_sel);
3181 :
3182 : /* Add in the eval cost of the locally-checked quals */
3183 2570 : startup_cost += fpinfo->local_conds_cost.startup;
3184 2570 : total_cost += fpinfo->local_conds_cost.per_tuple * retrieved_rows;
3185 2570 : cost_qual_eval(&local_cost, local_param_join_conds, root);
3186 2570 : startup_cost += local_cost.startup;
3187 2570 : total_cost += local_cost.per_tuple * retrieved_rows;
3188 :
3189 : /*
3190 : * Add in tlist eval cost for each output row. In case of an
3191 : * aggregate, some of the tlist expressions such as grouping
3192 : * expressions will be evaluated remotely, so adjust the costs.
3193 : */
3194 2570 : startup_cost += foreignrel->reltarget->cost.startup;
3195 2570 : total_cost += foreignrel->reltarget->cost.startup;
3196 2570 : total_cost += foreignrel->reltarget->cost.per_tuple * rows;
3197 2570 : if (IS_UPPER_REL(foreignrel))
3198 : {
3199 : QualCost tlist_cost;
3200 :
3201 80 : cost_qual_eval(&tlist_cost, fdw_scan_tlist, root);
3202 80 : startup_cost -= tlist_cost.startup;
3203 80 : total_cost -= tlist_cost.startup;
3204 80 : total_cost -= tlist_cost.per_tuple * rows;
3205 : }
3206 : }
3207 : else
3208 : {
3209 2796 : Cost run_cost = 0;
3210 :
3211 : /*
3212 : * We don't support join conditions in this mode (hence, no
3213 : * parameterized paths can be made).
3214 : */
3215 : Assert(param_join_conds == NIL);
3216 :
3217 : /*
3218 : * We will come here again and again with different set of pathkeys or
3219 : * additional post-scan/join-processing steps that caller wants to
3220 : * cost. We don't need to calculate the cost/size estimates for the
3221 : * underlying scan, join, or grouping each time. Instead, use those
3222 : * estimates if we have cached them already.
3223 : */
3224 2796 : if (fpinfo->rel_startup_cost >= 0 && fpinfo->rel_total_cost >= 0)
3225 : {
3226 : Assert(fpinfo->retrieved_rows >= 0);
3227 :
3228 620 : rows = fpinfo->rows;
3229 620 : retrieved_rows = fpinfo->retrieved_rows;
3230 620 : width = fpinfo->width;
3231 620 : startup_cost = fpinfo->rel_startup_cost;
3232 620 : run_cost = fpinfo->rel_total_cost - fpinfo->rel_startup_cost;
3233 :
3234 : /*
3235 : * If we estimate the costs of a foreign scan or a foreign join
3236 : * with additional post-scan/join-processing steps, the scan or
3237 : * join costs obtained from the cache wouldn't yet contain the
3238 : * eval costs for the final scan/join target, which would've been
3239 : * updated by apply_scanjoin_target_to_paths(); add the eval costs
3240 : * now.
3241 : */
3242 620 : if (fpextra && !IS_UPPER_REL(foreignrel))
3243 : {
3244 : /* Shouldn't get here unless we have LIMIT */
3245 : Assert(fpextra->has_limit);
3246 : Assert(foreignrel->reloptkind == RELOPT_BASEREL ||
3247 : foreignrel->reloptkind == RELOPT_JOINREL);
3248 180 : startup_cost += foreignrel->reltarget->cost.startup;
3249 180 : run_cost += foreignrel->reltarget->cost.per_tuple * rows;
3250 : }
3251 : }
3252 2176 : else if (IS_JOIN_REL(foreignrel))
3253 224 : {
3254 : PgFdwRelationInfo *fpinfo_i;
3255 : PgFdwRelationInfo *fpinfo_o;
3256 : QualCost join_cost;
3257 : QualCost remote_conds_cost;
3258 : double nrows;
3259 :
3260 : /* Use rows/width estimates made by the core code. */
3261 224 : rows = foreignrel->rows;
3262 224 : width = foreignrel->reltarget->width;
3263 :
3264 : /* For join we expect inner and outer relations set */
3265 : Assert(fpinfo->innerrel && fpinfo->outerrel);
3266 :
3267 224 : fpinfo_i = (PgFdwRelationInfo *) fpinfo->innerrel->fdw_private;
3268 224 : fpinfo_o = (PgFdwRelationInfo *) fpinfo->outerrel->fdw_private;
3269 :
3270 : /* Estimate of number of rows in cross product */
3271 224 : nrows = fpinfo_i->rows * fpinfo_o->rows;
3272 :
3273 : /*
3274 : * Back into an estimate of the number of retrieved rows. Just in
3275 : * case this is nuts, clamp to at most nrows.
3276 : */
3277 224 : retrieved_rows = clamp_row_est(rows / fpinfo->local_conds_sel);
3278 224 : retrieved_rows = Min(retrieved_rows, nrows);
3279 :
3280 : /*
3281 : * The cost of foreign join is estimated as cost of generating
3282 : * rows for the joining relations + cost for applying quals on the
3283 : * rows.
3284 : */
3285 :
3286 : /*
3287 : * Calculate the cost of clauses pushed down to the foreign server
3288 : */
3289 224 : cost_qual_eval(&remote_conds_cost, fpinfo->remote_conds, root);
3290 : /* Calculate the cost of applying join clauses */
3291 224 : cost_qual_eval(&join_cost, fpinfo->joinclauses, root);
3292 :
3293 : /*
3294 : * Startup cost includes startup cost of joining relations and the
3295 : * startup cost for join and other clauses. We do not include the
3296 : * startup cost specific to join strategy (e.g. setting up hash
3297 : * tables) since we do not know what strategy the foreign server
3298 : * is going to use.
3299 : */
3300 224 : startup_cost = fpinfo_i->rel_startup_cost + fpinfo_o->rel_startup_cost;
3301 224 : startup_cost += join_cost.startup;
3302 224 : startup_cost += remote_conds_cost.startup;
3303 224 : startup_cost += fpinfo->local_conds_cost.startup;
3304 :
3305 : /*
3306 : * Run time cost includes:
3307 : *
3308 : * 1. Run time cost (total_cost - startup_cost) of relations being
3309 : * joined
3310 : *
3311 : * 2. Run time cost of applying join clauses on the cross product
3312 : * of the joining relations.
3313 : *
3314 : * 3. Run time cost of applying pushed down other clauses on the
3315 : * result of join
3316 : *
3317 : * 4. Run time cost of applying nonpushable other clauses locally
3318 : * on the result fetched from the foreign server.
3319 : */
3320 224 : run_cost = fpinfo_i->rel_total_cost - fpinfo_i->rel_startup_cost;
3321 224 : run_cost += fpinfo_o->rel_total_cost - fpinfo_o->rel_startup_cost;
3322 224 : run_cost += nrows * join_cost.per_tuple;
3323 224 : nrows = clamp_row_est(nrows * fpinfo->joinclause_sel);
3324 224 : run_cost += nrows * remote_conds_cost.per_tuple;
3325 224 : run_cost += fpinfo->local_conds_cost.per_tuple * retrieved_rows;
3326 :
3327 : /* Add in tlist eval cost for each output row */
3328 224 : startup_cost += foreignrel->reltarget->cost.startup;
3329 224 : run_cost += foreignrel->reltarget->cost.per_tuple * rows;
3330 : }
3331 1952 : else if (IS_UPPER_REL(foreignrel))
3332 198 : {
3333 198 : RelOptInfo *outerrel = fpinfo->outerrel;
3334 : PgFdwRelationInfo *ofpinfo;
3335 198 : AggClauseCosts aggcosts = {0};
3336 : double input_rows;
3337 : int numGroupCols;
3338 198 : double numGroups = 1;
3339 :
3340 : /* The upper relation should have its outer relation set */
3341 : Assert(outerrel);
3342 : /* and that outer relation should have its reltarget set */
3343 : Assert(outerrel->reltarget);
3344 :
3345 : /*
3346 : * This cost model is mixture of costing done for sorted and
3347 : * hashed aggregates in cost_agg(). We are not sure which
3348 : * strategy will be considered at remote side, thus for
3349 : * simplicity, we put all startup related costs in startup_cost
3350 : * and all finalization and run cost are added in total_cost.
3351 : */
3352 :
3353 198 : ofpinfo = (PgFdwRelationInfo *) outerrel->fdw_private;
3354 :
3355 : /* Get rows from input rel */
3356 198 : input_rows = ofpinfo->rows;
3357 :
3358 : /* Collect statistics about aggregates for estimating costs. */
3359 198 : if (root->parse->hasAggs)
3360 : {
3361 190 : get_agg_clause_costs(root, AGGSPLIT_SIMPLE, &aggcosts);
3362 : }
3363 :
3364 : /* Get number of grouping columns and possible number of groups */
3365 198 : numGroupCols = list_length(root->processed_groupClause);
3366 198 : numGroups = estimate_num_groups(root,
3367 : get_sortgrouplist_exprs(root->processed_groupClause,
3368 : fpinfo->grouped_tlist),
3369 : input_rows, NULL, NULL);
3370 :
3371 : /*
3372 : * Get the retrieved_rows and rows estimates. If there are HAVING
3373 : * quals, account for their selectivity.
3374 : */
3375 198 : if (root->hasHavingQual)
3376 : {
3377 : /* Factor in the selectivity of the remotely-checked quals */
3378 : retrieved_rows =
3379 28 : clamp_row_est(numGroups *
3380 28 : clauselist_selectivity(root,
3381 : fpinfo->remote_conds,
3382 : 0,
3383 : JOIN_INNER,
3384 : NULL));
3385 : /* Factor in the selectivity of the locally-checked quals */
3386 28 : rows = clamp_row_est(retrieved_rows * fpinfo->local_conds_sel);
3387 : }
3388 : else
3389 : {
3390 170 : rows = retrieved_rows = numGroups;
3391 : }
3392 :
3393 : /* Use width estimate made by the core code. */
3394 198 : width = foreignrel->reltarget->width;
3395 :
3396 : /*-----
3397 : * Startup cost includes:
3398 : * 1. Startup cost for underneath input relation, adjusted for
3399 : * tlist replacement by apply_scanjoin_target_to_paths()
3400 : * 2. Cost of performing aggregation, per cost_agg()
3401 : *-----
3402 : */
3403 198 : startup_cost = ofpinfo->rel_startup_cost;
3404 198 : startup_cost += outerrel->reltarget->cost.startup;
3405 198 : startup_cost += aggcosts.transCost.startup;
3406 198 : startup_cost += aggcosts.transCost.per_tuple * input_rows;
3407 198 : startup_cost += aggcosts.finalCost.startup;
3408 198 : startup_cost += (cpu_operator_cost * numGroupCols) * input_rows;
3409 :
3410 : /*-----
3411 : * Run time cost includes:
3412 : * 1. Run time cost of underneath input relation, adjusted for
3413 : * tlist replacement by apply_scanjoin_target_to_paths()
3414 : * 2. Run time cost of performing aggregation, per cost_agg()
3415 : *-----
3416 : */
3417 198 : run_cost = ofpinfo->rel_total_cost - ofpinfo->rel_startup_cost;
3418 198 : run_cost += outerrel->reltarget->cost.per_tuple * input_rows;
3419 198 : run_cost += aggcosts.finalCost.per_tuple * numGroups;
3420 198 : run_cost += cpu_tuple_cost * numGroups;
3421 :
3422 : /* Account for the eval cost of HAVING quals, if any */
3423 198 : if (root->hasHavingQual)
3424 : {
3425 : QualCost remote_cost;
3426 :
3427 : /* Add in the eval cost of the remotely-checked quals */
3428 28 : cost_qual_eval(&remote_cost, fpinfo->remote_conds, root);
3429 28 : startup_cost += remote_cost.startup;
3430 28 : run_cost += remote_cost.per_tuple * numGroups;
3431 : /* Add in the eval cost of the locally-checked quals */
3432 28 : startup_cost += fpinfo->local_conds_cost.startup;
3433 28 : run_cost += fpinfo->local_conds_cost.per_tuple * retrieved_rows;
3434 : }
3435 :
3436 : /* Add in tlist eval cost for each output row */
3437 198 : startup_cost += foreignrel->reltarget->cost.startup;
3438 198 : run_cost += foreignrel->reltarget->cost.per_tuple * rows;
3439 : }
3440 : else
3441 : {
3442 : Cost cpu_per_tuple;
3443 :
3444 : /* Use rows/width estimates made by set_baserel_size_estimates. */
3445 1754 : rows = foreignrel->rows;
3446 1754 : width = foreignrel->reltarget->width;
3447 :
3448 : /*
3449 : * Back into an estimate of the number of retrieved rows. Just in
3450 : * case this is nuts, clamp to at most foreignrel->tuples.
3451 : */
3452 1754 : retrieved_rows = clamp_row_est(rows / fpinfo->local_conds_sel);
3453 1754 : retrieved_rows = Min(retrieved_rows, foreignrel->tuples);
3454 :
3455 : /*
3456 : * Cost as though this were a seqscan, which is pessimistic. We
3457 : * effectively imagine the local_conds are being evaluated
3458 : * remotely, too.
3459 : */
3460 1754 : startup_cost = 0;
3461 1754 : run_cost = 0;
3462 1754 : run_cost += seq_page_cost * foreignrel->pages;
3463 :
3464 1754 : startup_cost += foreignrel->baserestrictcost.startup;
3465 1754 : cpu_per_tuple = cpu_tuple_cost + foreignrel->baserestrictcost.per_tuple;
3466 1754 : run_cost += cpu_per_tuple * foreignrel->tuples;
3467 :
3468 : /* Add in tlist eval cost for each output row */
3469 1754 : startup_cost += foreignrel->reltarget->cost.startup;
3470 1754 : run_cost += foreignrel->reltarget->cost.per_tuple * rows;
3471 : }
3472 :
3473 : /*
3474 : * Without remote estimates, we have no real way to estimate the cost
3475 : * of generating sorted output. It could be free if the query plan
3476 : * the remote side would have chosen generates properly-sorted output
3477 : * anyway, but in most cases it will cost something. Estimate a value
3478 : * high enough that we won't pick the sorted path when the ordering
3479 : * isn't locally useful, but low enough that we'll err on the side of
3480 : * pushing down the ORDER BY clause when it's useful to do so.
3481 : */
3482 2796 : if (pathkeys != NIL)
3483 : {
3484 508 : if (IS_UPPER_REL(foreignrel))
3485 : {
3486 : Assert(foreignrel->reloptkind == RELOPT_UPPER_REL &&
3487 : fpinfo->stage == UPPERREL_GROUP_AGG);
3488 :
3489 : /*
3490 : * We can only get here when this function is called from
3491 : * add_foreign_ordered_paths() or add_foreign_final_paths();
3492 : * in which cases, the passed-in fpextra should not be NULL.
3493 : */
3494 : Assert(fpextra);
3495 60 : adjust_foreign_grouping_path_cost(root, pathkeys,
3496 : retrieved_rows, width,
3497 : fpextra->limit_tuples,
3498 : &disabled_nodes,
3499 : &startup_cost, &run_cost);
3500 : }
3501 : else
3502 : {
3503 448 : startup_cost *= DEFAULT_FDW_SORT_MULTIPLIER;
3504 448 : run_cost *= DEFAULT_FDW_SORT_MULTIPLIER;
3505 : }
3506 : }
3507 :
3508 2796 : total_cost = startup_cost + run_cost;
3509 :
3510 : /* Adjust the cost estimates if we have LIMIT */
3511 2796 : if (fpextra && fpextra->has_limit)
3512 : {
3513 184 : adjust_limit_rows_costs(&rows, &startup_cost, &total_cost,
3514 : fpextra->offset_est, fpextra->count_est);
3515 184 : retrieved_rows = rows;
3516 : }
3517 : }
3518 :
3519 : /*
3520 : * If this includes the final sort step, the given target, which will be
3521 : * applied to the resulting path, might have different expressions from
3522 : * the foreignrel's reltarget (see make_sort_input_target()); adjust tlist
3523 : * eval costs.
3524 : */
3525 5366 : if (fpextra && fpextra->has_final_sort &&
3526 218 : fpextra->target != foreignrel->reltarget)
3527 : {
3528 12 : QualCost oldcost = foreignrel->reltarget->cost;
3529 12 : QualCost newcost = fpextra->target->cost;
3530 :
3531 12 : startup_cost += newcost.startup - oldcost.startup;
3532 12 : total_cost += newcost.startup - oldcost.startup;
3533 12 : total_cost += (newcost.per_tuple - oldcost.per_tuple) * rows;
3534 : }
3535 :
3536 : /*
3537 : * Cache the retrieved rows and cost estimates for scans, joins, or
3538 : * groupings without any parameterization, pathkeys, or additional
3539 : * post-scan/join-processing steps, before adding the costs for
3540 : * transferring data from the foreign server. These estimates are useful
3541 : * for costing remote joins involving this relation or costing other
3542 : * remote operations on this relation such as remote sorts and remote
3543 : * LIMIT restrictions, when the costs can not be obtained from the foreign
3544 : * server. This function will be called at least once for every foreign
3545 : * relation without any parameterization, pathkeys, or additional
3546 : * post-scan/join-processing steps.
3547 : */
3548 5366 : if (pathkeys == NIL && param_join_conds == NIL && fpextra == NULL)
3549 : {
3550 3276 : fpinfo->retrieved_rows = retrieved_rows;
3551 3276 : fpinfo->rel_startup_cost = startup_cost;
3552 3276 : fpinfo->rel_total_cost = total_cost;
3553 : }
3554 :
3555 : /*
3556 : * Add some additional cost factors to account for connection overhead
3557 : * (fdw_startup_cost), transferring data across the network
3558 : * (fdw_tuple_cost per retrieved row), and local manipulation of the data
3559 : * (cpu_tuple_cost per retrieved row).
3560 : */
3561 5366 : startup_cost += fpinfo->fdw_startup_cost;
3562 5366 : total_cost += fpinfo->fdw_startup_cost;
3563 5366 : total_cost += fpinfo->fdw_tuple_cost * retrieved_rows;
3564 5366 : total_cost += cpu_tuple_cost * retrieved_rows;
3565 :
3566 : /*
3567 : * If we have LIMIT, we should prefer performing the restriction remotely
3568 : * rather than locally, as the former avoids extra row fetches from the
3569 : * remote that the latter might cause. But since the core code doesn't
3570 : * account for such fetches when estimating the costs of the local
3571 : * restriction (see create_limit_path()), there would be no difference
3572 : * between the costs of the local restriction and the costs of the remote
3573 : * restriction estimated above if we don't use remote estimates (except
3574 : * for the case where the foreignrel is a grouping relation, the given
3575 : * pathkeys is not NIL, and the effects of a bounded sort for that rel is
3576 : * accounted for in costing the remote restriction). Tweak the costs of
3577 : * the remote restriction to ensure we'll prefer it if LIMIT is a useful
3578 : * one.
3579 : */
3580 5366 : if (!fpinfo->use_remote_estimate &&
3581 244 : fpextra && fpextra->has_limit &&
3582 184 : fpextra->limit_tuples > 0 &&
3583 184 : fpextra->limit_tuples < fpinfo->rows)
3584 : {
3585 : Assert(fpinfo->rows > 0);
3586 172 : total_cost -= (total_cost - startup_cost) * 0.05 *
3587 172 : (fpinfo->rows - fpextra->limit_tuples) / fpinfo->rows;
3588 : }
3589 :
3590 : /* Return results. */
3591 5366 : *p_rows = rows;
3592 5366 : *p_width = width;
3593 5366 : *p_disabled_nodes = disabled_nodes;
3594 5366 : *p_startup_cost = startup_cost;
3595 5366 : *p_total_cost = total_cost;
3596 5366 : }
3597 :
3598 : /*
3599 : * Estimate costs of executing a SQL statement remotely.
3600 : * The given "sql" must be an EXPLAIN command.
3601 : */
3602 : static void
3603 2570 : get_remote_estimate(const char *sql, PGconn *conn,
3604 : double *rows, int *width,
3605 : Cost *startup_cost, Cost *total_cost)
3606 : {
3607 : PGresult *res;
3608 : char *line;
3609 : char *p;
3610 : int n;
3611 :
3612 : /*
3613 : * Execute EXPLAIN remotely.
3614 : */
3615 2570 : res = pgfdw_exec_query(conn, sql, NULL);
3616 2570 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
3617 0 : pgfdw_report_error(res, conn, sql);
3618 :
3619 : /*
3620 : * Extract cost numbers for topmost plan node. Note we search for a left
3621 : * paren from the end of the line to avoid being confused by other uses of
3622 : * parentheses.
3623 : */
3624 2570 : line = PQgetvalue(res, 0, 0);
3625 2570 : p = strrchr(line, '(');
3626 2570 : if (p == NULL)
3627 0 : elog(ERROR, "could not interpret EXPLAIN output: \"%s\"", line);
3628 2570 : n = sscanf(p, "(cost=%lf..%lf rows=%lf width=%d)",
3629 : startup_cost, total_cost, rows, width);
3630 2570 : if (n != 4)
3631 0 : elog(ERROR, "could not interpret EXPLAIN output: \"%s\"", line);
3632 2570 : PQclear(res);
3633 2570 : }
3634 :
3635 : /*
3636 : * Adjust the cost estimates of a foreign grouping path to include the cost of
3637 : * generating properly-sorted output.
3638 : */
3639 : static void
3640 60 : adjust_foreign_grouping_path_cost(PlannerInfo *root,
3641 : List *pathkeys,
3642 : double retrieved_rows,
3643 : double width,
3644 : double limit_tuples,
3645 : int *p_disabled_nodes,
3646 : Cost *p_startup_cost,
3647 : Cost *p_run_cost)
3648 : {
3649 : /*
3650 : * If the GROUP BY clause isn't sort-able, the plan chosen by the remote
3651 : * side is unlikely to generate properly-sorted output, so it would need
3652 : * an explicit sort; adjust the given costs with cost_sort(). Likewise,
3653 : * if the GROUP BY clause is sort-able but isn't a superset of the given
3654 : * pathkeys, adjust the costs with that function. Otherwise, adjust the
3655 : * costs by applying the same heuristic as for the scan or join case.
3656 : */
3657 60 : if (!grouping_is_sortable(root->processed_groupClause) ||
3658 60 : !pathkeys_contained_in(pathkeys, root->group_pathkeys))
3659 44 : {
3660 : Path sort_path; /* dummy for result of cost_sort */
3661 :
3662 44 : cost_sort(&sort_path,
3663 : root,
3664 : pathkeys,
3665 : 0,
3666 44 : *p_startup_cost + *p_run_cost,
3667 : retrieved_rows,
3668 : width,
3669 : 0.0,
3670 : work_mem,
3671 : limit_tuples);
3672 :
3673 44 : *p_startup_cost = sort_path.startup_cost;
3674 44 : *p_run_cost = sort_path.total_cost - sort_path.startup_cost;
3675 : }
3676 : else
3677 : {
3678 : /*
3679 : * The default extra cost seems too large for foreign-grouping cases;
3680 : * add 1/4th of that default.
3681 : */
3682 16 : double sort_multiplier = 1.0 + (DEFAULT_FDW_SORT_MULTIPLIER
3683 : - 1.0) * 0.25;
3684 :
3685 16 : *p_startup_cost *= sort_multiplier;
3686 16 : *p_run_cost *= sort_multiplier;
3687 : }
3688 60 : }
3689 :
3690 : /*
3691 : * Detect whether we want to process an EquivalenceClass member.
3692 : *
3693 : * This is a callback for use by generate_implied_equalities_for_column.
3694 : */
3695 : static bool
3696 596 : ec_member_matches_foreign(PlannerInfo *root, RelOptInfo *rel,
3697 : EquivalenceClass *ec, EquivalenceMember *em,
3698 : void *arg)
3699 : {
3700 596 : ec_member_foreign_arg *state = (ec_member_foreign_arg *) arg;
3701 596 : Expr *expr = em->em_expr;
3702 :
3703 : /*
3704 : * If we've identified what we're processing in the current scan, we only
3705 : * want to match that expression.
3706 : */
3707 596 : if (state->current != NULL)
3708 0 : return equal(expr, state->current);
3709 :
3710 : /*
3711 : * Otherwise, ignore anything we've already processed.
3712 : */
3713 596 : if (list_member(state->already_used, expr))
3714 314 : return false;
3715 :
3716 : /* This is the new target to process. */
3717 282 : state->current = expr;
3718 282 : return true;
3719 : }
3720 :
3721 : /*
3722 : * Create cursor for node's query with current parameter values.
3723 : */
3724 : static void
3725 1654 : create_cursor(ForeignScanState *node)
3726 : {
3727 1654 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
3728 1654 : ExprContext *econtext = node->ss.ps.ps_ExprContext;
3729 1654 : int numParams = fsstate->numParams;
3730 1654 : const char **values = fsstate->param_values;
3731 1654 : PGconn *conn = fsstate->conn;
3732 : StringInfoData buf;
3733 : PGresult *res;
3734 :
3735 : /* First, process a pending asynchronous request, if any. */
3736 1654 : if (fsstate->conn_state->pendingAreq)
3737 2 : process_pending_request(fsstate->conn_state->pendingAreq);
3738 :
3739 : /*
3740 : * Construct array of query parameter values in text format. We do the
3741 : * conversions in the short-lived per-tuple context, so as not to cause a
3742 : * memory leak over repeated scans.
3743 : */
3744 1654 : if (numParams > 0)
3745 : {
3746 : MemoryContext oldcontext;
3747 :
3748 694 : oldcontext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory);
3749 :
3750 694 : process_query_params(econtext,
3751 : fsstate->param_flinfo,
3752 : fsstate->param_exprs,
3753 : values);
3754 :
3755 694 : MemoryContextSwitchTo(oldcontext);
3756 : }
3757 :
3758 : /* Construct the DECLARE CURSOR command */
3759 1654 : initStringInfo(&buf);
3760 1654 : appendStringInfo(&buf, "DECLARE c%u CURSOR FOR\n%s",
3761 : fsstate->cursor_number, fsstate->query);
3762 :
3763 : /*
3764 : * Notice that we pass NULL for paramTypes, thus forcing the remote server
3765 : * to infer types for all parameters. Since we explicitly cast every
3766 : * parameter (see deparse.c), the "inference" is trivial and will produce
3767 : * the desired result. This allows us to avoid assuming that the remote
3768 : * server has the same OIDs we do for the parameters' types.
3769 : */
3770 1654 : if (!PQsendQueryParams(conn, buf.data, numParams,
3771 : NULL, values, NULL, NULL, 0))
3772 0 : pgfdw_report_error(NULL, conn, buf.data);
3773 :
3774 : /*
3775 : * Get the result, and check for success.
3776 : */
3777 1654 : res = pgfdw_get_result(conn);
3778 1652 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
3779 6 : pgfdw_report_error(res, conn, fsstate->query);
3780 1646 : PQclear(res);
3781 :
3782 : /* Mark the cursor as created, and show no tuples have been retrieved */
3783 1646 : fsstate->cursor_exists = true;
3784 1646 : fsstate->tuples = NULL;
3785 1646 : fsstate->num_tuples = 0;
3786 1646 : fsstate->next_tuple = 0;
3787 1646 : fsstate->fetch_ct_2 = 0;
3788 1646 : fsstate->eof_reached = false;
3789 :
3790 : /* Clean up */
3791 1646 : pfree(buf.data);
3792 1646 : }
3793 :
3794 : /*
3795 : * Fetch some more rows from the node's cursor.
3796 : */
3797 : static void
3798 2990 : fetch_more_data(ForeignScanState *node)
3799 : {
3800 2990 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
3801 2990 : PGconn *conn = fsstate->conn;
3802 : PGresult *res;
3803 : int numrows;
3804 : int i;
3805 : MemoryContext oldcontext;
3806 :
3807 : /*
3808 : * We'll store the tuples in the batch_cxt. First, flush the previous
3809 : * batch.
3810 : */
3811 2990 : fsstate->tuples = NULL;
3812 2990 : MemoryContextReset(fsstate->batch_cxt);
3813 2990 : oldcontext = MemoryContextSwitchTo(fsstate->batch_cxt);
3814 :
3815 2990 : if (fsstate->async_capable)
3816 : {
3817 : Assert(fsstate->conn_state->pendingAreq);
3818 :
3819 : /*
3820 : * The query was already sent by an earlier call to
3821 : * fetch_more_data_begin. So now we just fetch the result.
3822 : */
3823 318 : res = pgfdw_get_result(conn);
3824 : /* On error, report the original query, not the FETCH. */
3825 318 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
3826 0 : pgfdw_report_error(res, conn, fsstate->query);
3827 :
3828 : /* Reset per-connection state */
3829 318 : fsstate->conn_state->pendingAreq = NULL;
3830 : }
3831 : else
3832 : {
3833 : char sql[64];
3834 :
3835 : /* This is a regular synchronous fetch. */
3836 2672 : snprintf(sql, sizeof(sql), "FETCH %d FROM c%u",
3837 : fsstate->fetch_size, fsstate->cursor_number);
3838 :
3839 2672 : res = pgfdw_exec_query(conn, sql, fsstate->conn_state);
3840 : /* On error, report the original query, not the FETCH. */
3841 2672 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
3842 2 : pgfdw_report_error(res, conn, fsstate->query);
3843 : }
3844 :
3845 : /* Convert the data into HeapTuples */
3846 2988 : numrows = PQntuples(res);
3847 2988 : fsstate->tuples = (HeapTuple *) palloc0(numrows * sizeof(HeapTuple));
3848 2988 : fsstate->num_tuples = numrows;
3849 2988 : fsstate->next_tuple = 0;
3850 :
3851 145436 : for (i = 0; i < numrows; i++)
3852 : {
3853 : Assert(IsA(node->ss.ps.plan, ForeignScan));
3854 :
3855 142448 : fsstate->tuples[i] =
3856 142456 : make_tuple_from_result_row(res, i,
3857 : fsstate->rel,
3858 : fsstate->attinmeta,
3859 : fsstate->retrieved_attrs,
3860 : node,
3861 : fsstate->temp_cxt);
3862 : }
3863 :
3864 : /* Update fetch_ct_2 */
3865 2980 : if (fsstate->fetch_ct_2 < 2)
3866 1874 : fsstate->fetch_ct_2++;
3867 :
3868 : /* Must be EOF if we didn't get as many tuples as we asked for. */
3869 2980 : fsstate->eof_reached = (numrows < fsstate->fetch_size);
3870 :
3871 2980 : PQclear(res);
3872 :
3873 2980 : MemoryContextSwitchTo(oldcontext);
3874 2980 : }
3875 :
3876 : /*
3877 : * Force assorted GUC parameters to settings that ensure that we'll output
3878 : * data values in a form that is unambiguous to the remote server.
3879 : *
3880 : * This is rather expensive and annoying to do once per row, but there's
3881 : * little choice if we want to be sure values are transmitted accurately;
3882 : * we can't leave the settings in place between rows for fear of affecting
3883 : * user-visible computations.
3884 : *
3885 : * We use the equivalent of a function SET option to allow the settings to
3886 : * persist only until the caller calls reset_transmission_modes(). If an
3887 : * error is thrown in between, guc.c will take care of undoing the settings.
3888 : *
3889 : * The return value is the nestlevel that must be passed to
3890 : * reset_transmission_modes() to undo things.
3891 : */
3892 : int
3893 8404 : set_transmission_modes(void)
3894 : {
3895 8404 : int nestlevel = NewGUCNestLevel();
3896 :
3897 : /*
3898 : * The values set here should match what pg_dump does. See also
3899 : * configure_remote_session in connection.c.
3900 : */
3901 8404 : if (DateStyle != USE_ISO_DATES)
3902 8400 : (void) set_config_option("datestyle", "ISO",
3903 : PGC_USERSET, PGC_S_SESSION,
3904 : GUC_ACTION_SAVE, true, 0, false);
3905 8404 : if (IntervalStyle != INTSTYLE_POSTGRES)
3906 8400 : (void) set_config_option("intervalstyle", "postgres",
3907 : PGC_USERSET, PGC_S_SESSION,
3908 : GUC_ACTION_SAVE, true, 0, false);
3909 8404 : if (extra_float_digits < 3)
3910 8400 : (void) set_config_option("extra_float_digits", "3",
3911 : PGC_USERSET, PGC_S_SESSION,
3912 : GUC_ACTION_SAVE, true, 0, false);
3913 :
3914 : /*
3915 : * In addition force restrictive search_path, in case there are any
3916 : * regproc or similar constants to be printed.
3917 : */
3918 8404 : (void) set_config_option("search_path", "pg_catalog",
3919 : PGC_USERSET, PGC_S_SESSION,
3920 : GUC_ACTION_SAVE, true, 0, false);
3921 :
3922 8404 : return nestlevel;
3923 : }
3924 :
3925 : /*
3926 : * Undo the effects of set_transmission_modes().
3927 : */
3928 : void
3929 8404 : reset_transmission_modes(int nestlevel)
3930 : {
3931 8404 : AtEOXact_GUC(true, nestlevel);
3932 8404 : }
3933 :
3934 : /*
3935 : * Utility routine to close a cursor.
3936 : */
3937 : static void
3938 982 : close_cursor(PGconn *conn, unsigned int cursor_number,
3939 : PgFdwConnState *conn_state)
3940 : {
3941 : char sql[64];
3942 : PGresult *res;
3943 :
3944 982 : snprintf(sql, sizeof(sql), "CLOSE c%u", cursor_number);
3945 982 : res = pgfdw_exec_query(conn, sql, conn_state);
3946 982 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
3947 0 : pgfdw_report_error(res, conn, sql);
3948 982 : PQclear(res);
3949 982 : }
3950 :
3951 : /*
3952 : * create_foreign_modify
3953 : * Construct an execution state of a foreign insert/update/delete
3954 : * operation
3955 : */
3956 : static PgFdwModifyState *
3957 364 : create_foreign_modify(EState *estate,
3958 : RangeTblEntry *rte,
3959 : ResultRelInfo *resultRelInfo,
3960 : CmdType operation,
3961 : Plan *subplan,
3962 : char *query,
3963 : List *target_attrs,
3964 : int values_end,
3965 : bool has_returning,
3966 : List *retrieved_attrs)
3967 : {
3968 : PgFdwModifyState *fmstate;
3969 364 : Relation rel = resultRelInfo->ri_RelationDesc;
3970 364 : TupleDesc tupdesc = RelationGetDescr(rel);
3971 : Oid userid;
3972 : ForeignTable *table;
3973 : UserMapping *user;
3974 : AttrNumber n_params;
3975 : Oid typefnoid;
3976 : bool isvarlena;
3977 : ListCell *lc;
3978 :
3979 : /* Begin constructing PgFdwModifyState. */
3980 364 : fmstate = (PgFdwModifyState *) palloc0(sizeof(PgFdwModifyState));
3981 364 : fmstate->rel = rel;
3982 :
3983 : /* Identify which user to do the remote access as. */
3984 364 : userid = ExecGetResultRelCheckAsUser(resultRelInfo, estate);
3985 :
3986 : /* Get info about foreign table. */
3987 364 : table = GetForeignTable(RelationGetRelid(rel));
3988 364 : user = GetUserMapping(userid, table->serverid);
3989 :
3990 : /* Open connection; report that we'll create a prepared statement. */
3991 364 : fmstate->conn = GetConnection(user, true, &fmstate->conn_state);
3992 364 : fmstate->p_name = NULL; /* prepared statement not made yet */
3993 :
3994 : /* Set up remote query information. */
3995 364 : fmstate->query = query;
3996 364 : if (operation == CMD_INSERT)
3997 : {
3998 266 : fmstate->query = pstrdup(fmstate->query);
3999 266 : fmstate->orig_query = pstrdup(fmstate->query);
4000 : }
4001 364 : fmstate->target_attrs = target_attrs;
4002 364 : fmstate->values_end = values_end;
4003 364 : fmstate->has_returning = has_returning;
4004 364 : fmstate->retrieved_attrs = retrieved_attrs;
4005 :
4006 : /* Create context for per-tuple temp workspace. */
4007 364 : fmstate->temp_cxt = AllocSetContextCreate(estate->es_query_cxt,
4008 : "postgres_fdw temporary data",
4009 : ALLOCSET_SMALL_SIZES);
4010 :
4011 : /* Prepare for input conversion of RETURNING results. */
4012 364 : if (fmstate->has_returning)
4013 124 : fmstate->attinmeta = TupleDescGetAttInMetadata(tupdesc);
4014 :
4015 : /* Prepare for output conversion of parameters used in prepared stmt. */
4016 364 : n_params = list_length(fmstate->target_attrs) + 1;
4017 364 : fmstate->p_flinfo = (FmgrInfo *) palloc0(sizeof(FmgrInfo) * n_params);
4018 364 : fmstate->p_nums = 0;
4019 :
4020 364 : if (operation == CMD_UPDATE || operation == CMD_DELETE)
4021 : {
4022 : Assert(subplan != NULL);
4023 :
4024 : /* Find the ctid resjunk column in the subplan's result */
4025 98 : fmstate->ctidAttno = ExecFindJunkAttributeInTlist(subplan->targetlist,
4026 : "ctid");
4027 98 : if (!AttributeNumberIsValid(fmstate->ctidAttno))
4028 0 : elog(ERROR, "could not find junk ctid column");
4029 :
4030 : /* First transmittable parameter will be ctid */
4031 98 : getTypeOutputInfo(TIDOID, &typefnoid, &isvarlena);
4032 98 : fmgr_info(typefnoid, &fmstate->p_flinfo[fmstate->p_nums]);
4033 98 : fmstate->p_nums++;
4034 : }
4035 :
4036 364 : if (operation == CMD_INSERT || operation == CMD_UPDATE)
4037 : {
4038 : /* Set up for remaining transmittable parameters */
4039 1136 : foreach(lc, fmstate->target_attrs)
4040 : {
4041 798 : int attnum = lfirst_int(lc);
4042 798 : Form_pg_attribute attr = TupleDescAttr(tupdesc, attnum - 1);
4043 :
4044 : Assert(!attr->attisdropped);
4045 :
4046 : /* Ignore generated columns; they are set to DEFAULT */
4047 798 : if (attr->attgenerated)
4048 16 : continue;
4049 782 : getTypeOutputInfo(attr->atttypid, &typefnoid, &isvarlena);
4050 782 : fmgr_info(typefnoid, &fmstate->p_flinfo[fmstate->p_nums]);
4051 782 : fmstate->p_nums++;
4052 : }
4053 : }
4054 :
4055 : Assert(fmstate->p_nums <= n_params);
4056 :
4057 : /* Set batch_size from foreign server/table options. */
4058 364 : if (operation == CMD_INSERT)
4059 266 : fmstate->batch_size = get_batch_size_option(rel);
4060 :
4061 364 : fmstate->num_slots = 1;
4062 :
4063 : /* Initialize auxiliary state */
4064 364 : fmstate->aux_fmstate = NULL;
4065 :
4066 364 : return fmstate;
4067 : }
4068 :
4069 : /*
4070 : * execute_foreign_modify
4071 : * Perform foreign-table modification as required, and fetch RETURNING
4072 : * result if any. (This is the shared guts of postgresExecForeignInsert,
4073 : * postgresExecForeignBatchInsert, postgresExecForeignUpdate, and
4074 : * postgresExecForeignDelete.)
4075 : */
4076 : static TupleTableSlot **
4077 2104 : execute_foreign_modify(EState *estate,
4078 : ResultRelInfo *resultRelInfo,
4079 : CmdType operation,
4080 : TupleTableSlot **slots,
4081 : TupleTableSlot **planSlots,
4082 : int *numSlots)
4083 : {
4084 2104 : PgFdwModifyState *fmstate = (PgFdwModifyState *) resultRelInfo->ri_FdwState;
4085 2104 : ItemPointer ctid = NULL;
4086 : const char **p_values;
4087 : PGresult *res;
4088 : int n_rows;
4089 : StringInfoData sql;
4090 :
4091 : /* The operation should be INSERT, UPDATE, or DELETE */
4092 : Assert(operation == CMD_INSERT ||
4093 : operation == CMD_UPDATE ||
4094 : operation == CMD_DELETE);
4095 :
4096 : /* First, process a pending asynchronous request, if any. */
4097 2104 : if (fmstate->conn_state->pendingAreq)
4098 2 : process_pending_request(fmstate->conn_state->pendingAreq);
4099 :
4100 : /*
4101 : * If the existing query was deparsed and prepared for a different number
4102 : * of rows, rebuild it for the proper number.
4103 : */
4104 2104 : if (operation == CMD_INSERT && fmstate->num_slots != *numSlots)
4105 : {
4106 : /* Destroy the prepared statement created previously */
4107 52 : if (fmstate->p_name)
4108 22 : deallocate_query(fmstate);
4109 :
4110 : /* Build INSERT string with numSlots records in its VALUES clause. */
4111 52 : initStringInfo(&sql);
4112 52 : rebuildInsertSql(&sql, fmstate->rel,
4113 : fmstate->orig_query, fmstate->target_attrs,
4114 : fmstate->values_end, fmstate->p_nums,
4115 52 : *numSlots - 1);
4116 52 : pfree(fmstate->query);
4117 52 : fmstate->query = sql.data;
4118 52 : fmstate->num_slots = *numSlots;
4119 : }
4120 :
4121 : /* Set up the prepared statement on the remote server, if we didn't yet */
4122 2104 : if (!fmstate->p_name)
4123 374 : prepare_foreign_modify(fmstate);
4124 :
4125 : /*
4126 : * For UPDATE/DELETE, get the ctid that was passed up as a resjunk column
4127 : */
4128 2104 : if (operation == CMD_UPDATE || operation == CMD_DELETE)
4129 : {
4130 : Datum datum;
4131 : bool isNull;
4132 :
4133 236 : datum = ExecGetJunkAttribute(planSlots[0],
4134 236 : fmstate->ctidAttno,
4135 : &isNull);
4136 : /* shouldn't ever get a null result... */
4137 236 : if (isNull)
4138 0 : elog(ERROR, "ctid is NULL");
4139 236 : ctid = (ItemPointer) DatumGetPointer(datum);
4140 : }
4141 :
4142 : /* Convert parameters needed by prepared statement to text form */
4143 2104 : p_values = convert_prep_stmt_params(fmstate, ctid, slots, *numSlots);
4144 :
4145 : /*
4146 : * Execute the prepared statement.
4147 : */
4148 2104 : if (!PQsendQueryPrepared(fmstate->conn,
4149 2104 : fmstate->p_name,
4150 2104 : fmstate->p_nums * (*numSlots),
4151 : p_values,
4152 : NULL,
4153 : NULL,
4154 : 0))
4155 0 : pgfdw_report_error(NULL, fmstate->conn, fmstate->query);
4156 :
4157 : /*
4158 : * Get the result, and check for success.
4159 : */
4160 2104 : res = pgfdw_get_result(fmstate->conn);
4161 4208 : if (PQresultStatus(res) !=
4162 2104 : (fmstate->has_returning ? PGRES_TUPLES_OK : PGRES_COMMAND_OK))
4163 10 : pgfdw_report_error(res, fmstate->conn, fmstate->query);
4164 :
4165 : /* Check number of rows affected, and fetch RETURNING tuple if any */
4166 2094 : if (fmstate->has_returning)
4167 : {
4168 : Assert(*numSlots == 1);
4169 216 : n_rows = PQntuples(res);
4170 216 : if (n_rows > 0)
4171 214 : store_returning_result(fmstate, slots[0], res);
4172 : }
4173 : else
4174 1878 : n_rows = atoi(PQcmdTuples(res));
4175 :
4176 : /* And clean up */
4177 2094 : PQclear(res);
4178 :
4179 2094 : MemoryContextReset(fmstate->temp_cxt);
4180 :
4181 2094 : *numSlots = n_rows;
4182 :
4183 : /*
4184 : * Return NULL if nothing was inserted/updated/deleted on the remote end
4185 : */
4186 2094 : return (n_rows > 0) ? slots : NULL;
4187 : }
4188 :
4189 : /*
4190 : * prepare_foreign_modify
4191 : * Establish a prepared statement for execution of INSERT/UPDATE/DELETE
4192 : */
4193 : static void
4194 374 : prepare_foreign_modify(PgFdwModifyState *fmstate)
4195 : {
4196 : char prep_name[NAMEDATALEN];
4197 : char *p_name;
4198 : PGresult *res;
4199 :
4200 : /*
4201 : * The caller would already have processed a pending asynchronous request
4202 : * if any, so no need to do it here.
4203 : */
4204 :
4205 : /* Construct name we'll use for the prepared statement. */
4206 374 : snprintf(prep_name, sizeof(prep_name), "pgsql_fdw_prep_%u",
4207 : GetPrepStmtNumber(fmstate->conn));
4208 374 : p_name = pstrdup(prep_name);
4209 :
4210 : /*
4211 : * We intentionally do not specify parameter types here, but leave the
4212 : * remote server to derive them by default. This avoids possible problems
4213 : * with the remote server using different type OIDs than we do. All of
4214 : * the prepared statements we use in this module are simple enough that
4215 : * the remote server will make the right choices.
4216 : */
4217 374 : if (!PQsendPrepare(fmstate->conn,
4218 : p_name,
4219 374 : fmstate->query,
4220 : 0,
4221 : NULL))
4222 0 : pgfdw_report_error(NULL, fmstate->conn, fmstate->query);
4223 :
4224 : /*
4225 : * Get the result, and check for success.
4226 : */
4227 374 : res = pgfdw_get_result(fmstate->conn);
4228 374 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
4229 0 : pgfdw_report_error(res, fmstate->conn, fmstate->query);
4230 374 : PQclear(res);
4231 :
4232 : /* This action shows that the prepare has been done. */
4233 374 : fmstate->p_name = p_name;
4234 374 : }
4235 :
4236 : /*
4237 : * convert_prep_stmt_params
4238 : * Create array of text strings representing parameter values
4239 : *
4240 : * tupleid is ctid to send, or NULL if none
4241 : * slot is slot to get remaining parameters from, or NULL if none
4242 : *
4243 : * Data is constructed in temp_cxt; caller should reset that after use.
4244 : */
4245 : static const char **
4246 2104 : convert_prep_stmt_params(PgFdwModifyState *fmstate,
4247 : ItemPointer tupleid,
4248 : TupleTableSlot **slots,
4249 : int numSlots)
4250 : {
4251 : const char **p_values;
4252 : int i;
4253 : int j;
4254 2104 : int pindex = 0;
4255 : MemoryContext oldcontext;
4256 :
4257 2104 : oldcontext = MemoryContextSwitchTo(fmstate->temp_cxt);
4258 :
4259 2104 : p_values = (const char **) palloc(sizeof(char *) * fmstate->p_nums * numSlots);
4260 :
4261 : /* ctid is provided only for UPDATE/DELETE, which don't allow batching */
4262 : Assert(!(tupleid != NULL && numSlots > 1));
4263 :
4264 : /* 1st parameter should be ctid, if it's in use */
4265 2104 : if (tupleid != NULL)
4266 : {
4267 : Assert(numSlots == 1);
4268 : /* don't need set_transmission_modes for TID output */
4269 236 : p_values[pindex] = OutputFunctionCall(&fmstate->p_flinfo[pindex],
4270 : PointerGetDatum(tupleid));
4271 236 : pindex++;
4272 : }
4273 :
4274 : /* get following parameters from slots */
4275 2104 : if (slots != NULL && fmstate->target_attrs != NIL)
4276 : {
4277 2052 : TupleDesc tupdesc = RelationGetDescr(fmstate->rel);
4278 : int nestlevel;
4279 : ListCell *lc;
4280 :
4281 2052 : nestlevel = set_transmission_modes();
4282 :
4283 4348 : for (i = 0; i < numSlots; i++)
4284 : {
4285 2296 : j = (tupleid != NULL) ? 1 : 0;
4286 9594 : foreach(lc, fmstate->target_attrs)
4287 : {
4288 7298 : int attnum = lfirst_int(lc);
4289 7298 : CompactAttribute *attr = TupleDescCompactAttr(tupdesc, attnum - 1);
4290 : Datum value;
4291 : bool isnull;
4292 :
4293 : /* Ignore generated columns; they are set to DEFAULT */
4294 7298 : if (attr->attgenerated)
4295 28 : continue;
4296 7270 : value = slot_getattr(slots[i], attnum, &isnull);
4297 7270 : if (isnull)
4298 1166 : p_values[pindex] = NULL;
4299 : else
4300 6104 : p_values[pindex] = OutputFunctionCall(&fmstate->p_flinfo[j],
4301 : value);
4302 7270 : pindex++;
4303 7270 : j++;
4304 : }
4305 : }
4306 :
4307 2052 : reset_transmission_modes(nestlevel);
4308 : }
4309 :
4310 : Assert(pindex == fmstate->p_nums * numSlots);
4311 :
4312 2104 : MemoryContextSwitchTo(oldcontext);
4313 :
4314 2104 : return p_values;
4315 : }
4316 :
4317 : /*
4318 : * store_returning_result
4319 : * Store the result of a RETURNING clause
4320 : */
4321 : static void
4322 214 : store_returning_result(PgFdwModifyState *fmstate,
4323 : TupleTableSlot *slot, PGresult *res)
4324 : {
4325 : HeapTuple newtup;
4326 :
4327 214 : newtup = make_tuple_from_result_row(res, 0,
4328 : fmstate->rel,
4329 : fmstate->attinmeta,
4330 : fmstate->retrieved_attrs,
4331 : NULL,
4332 : fmstate->temp_cxt);
4333 :
4334 : /*
4335 : * The returning slot will not necessarily be suitable to store heaptuples
4336 : * directly, so allow for conversion.
4337 : */
4338 214 : ExecForceStoreHeapTuple(newtup, slot, true);
4339 214 : }
4340 :
4341 : /*
4342 : * finish_foreign_modify
4343 : * Release resources for a foreign insert/update/delete operation
4344 : */
4345 : static void
4346 320 : finish_foreign_modify(PgFdwModifyState *fmstate)
4347 : {
4348 : Assert(fmstate != NULL);
4349 :
4350 : /* If we created a prepared statement, destroy it */
4351 320 : deallocate_query(fmstate);
4352 :
4353 : /* Release remote connection */
4354 320 : ReleaseConnection(fmstate->conn);
4355 320 : fmstate->conn = NULL;
4356 320 : }
4357 :
4358 : /*
4359 : * deallocate_query
4360 : * Deallocate a prepared statement for a foreign insert/update/delete
4361 : * operation
4362 : */
4363 : static void
4364 342 : deallocate_query(PgFdwModifyState *fmstate)
4365 : {
4366 : char sql[64];
4367 : PGresult *res;
4368 :
4369 : /* do nothing if the query is not allocated */
4370 342 : if (!fmstate->p_name)
4371 8 : return;
4372 :
4373 334 : snprintf(sql, sizeof(sql), "DEALLOCATE %s", fmstate->p_name);
4374 334 : res = pgfdw_exec_query(fmstate->conn, sql, fmstate->conn_state);
4375 334 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
4376 0 : pgfdw_report_error(res, fmstate->conn, sql);
4377 334 : PQclear(res);
4378 334 : pfree(fmstate->p_name);
4379 334 : fmstate->p_name = NULL;
4380 : }
4381 :
4382 : /*
4383 : * build_remote_returning
4384 : * Build a RETURNING targetlist of a remote query for performing an
4385 : * UPDATE/DELETE .. RETURNING on a join directly
4386 : */
4387 : static List *
4388 8 : build_remote_returning(Index rtindex, Relation rel, List *returningList)
4389 : {
4390 8 : bool have_wholerow = false;
4391 8 : List *tlist = NIL;
4392 : List *vars;
4393 : ListCell *lc;
4394 :
4395 : Assert(returningList);
4396 :
4397 8 : vars = pull_var_clause((Node *) returningList, PVC_INCLUDE_PLACEHOLDERS);
4398 :
4399 : /*
4400 : * If there's a whole-row reference to the target relation, then we'll
4401 : * need all the columns of the relation.
4402 : */
4403 8 : foreach(lc, vars)
4404 : {
4405 4 : Var *var = (Var *) lfirst(lc);
4406 :
4407 4 : if (IsA(var, Var) &&
4408 4 : var->varno == rtindex &&
4409 4 : var->varattno == InvalidAttrNumber)
4410 : {
4411 4 : have_wholerow = true;
4412 4 : break;
4413 : }
4414 : }
4415 :
4416 8 : if (have_wholerow)
4417 : {
4418 4 : TupleDesc tupdesc = RelationGetDescr(rel);
4419 : int i;
4420 :
4421 40 : for (i = 1; i <= tupdesc->natts; i++)
4422 : {
4423 36 : Form_pg_attribute attr = TupleDescAttr(tupdesc, i - 1);
4424 : Var *var;
4425 :
4426 : /* Ignore dropped attributes. */
4427 36 : if (attr->attisdropped)
4428 4 : continue;
4429 :
4430 32 : var = makeVar(rtindex,
4431 : i,
4432 : attr->atttypid,
4433 : attr->atttypmod,
4434 : attr->attcollation,
4435 : 0);
4436 :
4437 32 : tlist = lappend(tlist,
4438 32 : makeTargetEntry((Expr *) var,
4439 32 : list_length(tlist) + 1,
4440 : NULL,
4441 : false));
4442 : }
4443 : }
4444 :
4445 : /* Now add any remaining columns to tlist. */
4446 60 : foreach(lc, vars)
4447 : {
4448 52 : Var *var = (Var *) lfirst(lc);
4449 :
4450 : /*
4451 : * No need for whole-row references to the target relation. We don't
4452 : * need system columns other than ctid and oid either, since those are
4453 : * set locally.
4454 : */
4455 52 : if (IsA(var, Var) &&
4456 52 : var->varno == rtindex &&
4457 36 : var->varattno <= InvalidAttrNumber &&
4458 4 : var->varattno != SelfItemPointerAttributeNumber)
4459 4 : continue; /* don't need it */
4460 :
4461 48 : if (tlist_member((Expr *) var, tlist))
4462 32 : continue; /* already got it */
4463 :
4464 16 : tlist = lappend(tlist,
4465 16 : makeTargetEntry((Expr *) var,
4466 16 : list_length(tlist) + 1,
4467 : NULL,
4468 : false));
4469 : }
4470 :
4471 8 : list_free(vars);
4472 :
4473 8 : return tlist;
4474 : }
4475 :
4476 : /*
4477 : * rebuild_fdw_scan_tlist
4478 : * Build new fdw_scan_tlist of given foreign-scan plan node from given
4479 : * tlist
4480 : *
4481 : * There might be columns that the fdw_scan_tlist of the given foreign-scan
4482 : * plan node contains that the given tlist doesn't. The fdw_scan_tlist would
4483 : * have contained resjunk columns such as 'ctid' of the target relation and
4484 : * 'wholerow' of non-target relations, but the tlist might not contain them,
4485 : * for example. So, adjust the tlist so it contains all the columns specified
4486 : * in the fdw_scan_tlist; else setrefs.c will get confused.
4487 : */
4488 : static void
4489 4 : rebuild_fdw_scan_tlist(ForeignScan *fscan, List *tlist)
4490 : {
4491 4 : List *new_tlist = tlist;
4492 4 : List *old_tlist = fscan->fdw_scan_tlist;
4493 : ListCell *lc;
4494 :
4495 32 : foreach(lc, old_tlist)
4496 : {
4497 28 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
4498 :
4499 28 : if (tlist_member(tle->expr, new_tlist))
4500 16 : continue; /* already got it */
4501 :
4502 12 : new_tlist = lappend(new_tlist,
4503 12 : makeTargetEntry(tle->expr,
4504 12 : list_length(new_tlist) + 1,
4505 : NULL,
4506 : false));
4507 : }
4508 4 : fscan->fdw_scan_tlist = new_tlist;
4509 4 : }
4510 :
4511 : /*
4512 : * Execute a direct UPDATE/DELETE statement.
4513 : */
4514 : static void
4515 142 : execute_dml_stmt(ForeignScanState *node)
4516 : {
4517 142 : PgFdwDirectModifyState *dmstate = (PgFdwDirectModifyState *) node->fdw_state;
4518 142 : ExprContext *econtext = node->ss.ps.ps_ExprContext;
4519 142 : int numParams = dmstate->numParams;
4520 142 : const char **values = dmstate->param_values;
4521 :
4522 : /* First, process a pending asynchronous request, if any. */
4523 142 : if (dmstate->conn_state->pendingAreq)
4524 2 : process_pending_request(dmstate->conn_state->pendingAreq);
4525 :
4526 : /*
4527 : * Construct array of query parameter values in text format.
4528 : */
4529 142 : if (numParams > 0)
4530 0 : process_query_params(econtext,
4531 : dmstate->param_flinfo,
4532 : dmstate->param_exprs,
4533 : values);
4534 :
4535 : /*
4536 : * Notice that we pass NULL for paramTypes, thus forcing the remote server
4537 : * to infer types for all parameters. Since we explicitly cast every
4538 : * parameter (see deparse.c), the "inference" is trivial and will produce
4539 : * the desired result. This allows us to avoid assuming that the remote
4540 : * server has the same OIDs we do for the parameters' types.
4541 : */
4542 142 : if (!PQsendQueryParams(dmstate->conn, dmstate->query, numParams,
4543 : NULL, values, NULL, NULL, 0))
4544 0 : pgfdw_report_error(NULL, dmstate->conn, dmstate->query);
4545 :
4546 : /*
4547 : * Get the result, and check for success.
4548 : */
4549 142 : dmstate->result = pgfdw_get_result(dmstate->conn);
4550 284 : if (PQresultStatus(dmstate->result) !=
4551 142 : (dmstate->has_returning ? PGRES_TUPLES_OK : PGRES_COMMAND_OK))
4552 8 : pgfdw_report_error(dmstate->result, dmstate->conn,
4553 8 : dmstate->query);
4554 :
4555 : /*
4556 : * The result potentially needs to survive across multiple executor row
4557 : * cycles, so move it to the context where the dmstate is.
4558 : */
4559 134 : dmstate->result = libpqsrv_PGresultSetParent(dmstate->result,
4560 : GetMemoryChunkContext(dmstate));
4561 :
4562 : /* Get the number of rows affected. */
4563 134 : if (dmstate->has_returning)
4564 28 : dmstate->num_tuples = PQntuples(dmstate->result);
4565 : else
4566 106 : dmstate->num_tuples = atoi(PQcmdTuples(dmstate->result));
4567 134 : }
4568 :
4569 : /*
4570 : * Get the result of a RETURNING clause.
4571 : */
4572 : static TupleTableSlot *
4573 728 : get_returning_data(ForeignScanState *node)
4574 : {
4575 728 : PgFdwDirectModifyState *dmstate = (PgFdwDirectModifyState *) node->fdw_state;
4576 728 : EState *estate = node->ss.ps.state;
4577 728 : ResultRelInfo *resultRelInfo = node->resultRelInfo;
4578 728 : TupleTableSlot *slot = node->ss.ss_ScanTupleSlot;
4579 : TupleTableSlot *resultSlot;
4580 :
4581 : Assert(resultRelInfo->ri_projectReturning);
4582 :
4583 : /* If we didn't get any tuples, must be end of data. */
4584 728 : if (dmstate->next_tuple >= dmstate->num_tuples)
4585 34 : return ExecClearTuple(slot);
4586 :
4587 : /* Increment the command es_processed count if necessary. */
4588 694 : if (dmstate->set_processed)
4589 692 : estate->es_processed += 1;
4590 :
4591 : /*
4592 : * Store a RETURNING tuple. If has_returning is false, just emit a dummy
4593 : * tuple. (has_returning is false when the local query is of the form
4594 : * "UPDATE/DELETE .. RETURNING 1" for example.)
4595 : */
4596 694 : if (!dmstate->has_returning)
4597 : {
4598 24 : ExecStoreAllNullTuple(slot);
4599 24 : resultSlot = slot;
4600 : }
4601 : else
4602 : {
4603 : HeapTuple newtup;
4604 :
4605 670 : newtup = make_tuple_from_result_row(dmstate->result,
4606 : dmstate->next_tuple,
4607 : dmstate->rel,
4608 : dmstate->attinmeta,
4609 : dmstate->retrieved_attrs,
4610 : node,
4611 : dmstate->temp_cxt);
4612 670 : ExecStoreHeapTuple(newtup, slot, false);
4613 : /* Get the updated/deleted tuple. */
4614 670 : if (dmstate->rel)
4615 638 : resultSlot = slot;
4616 : else
4617 32 : resultSlot = apply_returning_filter(dmstate, resultRelInfo, slot, estate);
4618 : }
4619 694 : dmstate->next_tuple++;
4620 :
4621 : /* Make slot available for evaluation of the local query RETURNING list. */
4622 694 : resultRelInfo->ri_projectReturning->pi_exprContext->ecxt_scantuple =
4623 : resultSlot;
4624 :
4625 694 : return slot;
4626 : }
4627 :
4628 : /*
4629 : * Initialize a filter to extract an updated/deleted tuple from a scan tuple.
4630 : */
4631 : static void
4632 2 : init_returning_filter(PgFdwDirectModifyState *dmstate,
4633 : List *fdw_scan_tlist,
4634 : Index rtindex)
4635 : {
4636 2 : TupleDesc resultTupType = RelationGetDescr(dmstate->resultRel);
4637 : ListCell *lc;
4638 : int i;
4639 :
4640 : /*
4641 : * Calculate the mapping between the fdw_scan_tlist's entries and the
4642 : * result tuple's attributes.
4643 : *
4644 : * The "map" is an array of indexes of the result tuple's attributes in
4645 : * fdw_scan_tlist, i.e., one entry for every attribute of the result
4646 : * tuple. We store zero for any attributes that don't have the
4647 : * corresponding entries in that list, marking that a NULL is needed in
4648 : * the result tuple.
4649 : *
4650 : * Also get the indexes of the entries for ctid and oid if any.
4651 : */
4652 2 : dmstate->attnoMap = (AttrNumber *)
4653 2 : palloc0(resultTupType->natts * sizeof(AttrNumber));
4654 :
4655 2 : dmstate->ctidAttno = dmstate->oidAttno = 0;
4656 :
4657 2 : i = 1;
4658 2 : dmstate->hasSystemCols = false;
4659 32 : foreach(lc, fdw_scan_tlist)
4660 : {
4661 30 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
4662 30 : Var *var = (Var *) tle->expr;
4663 :
4664 : Assert(IsA(var, Var));
4665 :
4666 : /*
4667 : * If the Var is a column of the target relation to be retrieved from
4668 : * the foreign server, get the index of the entry.
4669 : */
4670 50 : if (var->varno == rtindex &&
4671 20 : list_member_int(dmstate->retrieved_attrs, i))
4672 : {
4673 16 : int attrno = var->varattno;
4674 :
4675 16 : if (attrno < 0)
4676 : {
4677 : /*
4678 : * We don't retrieve system columns other than ctid and oid.
4679 : */
4680 0 : if (attrno == SelfItemPointerAttributeNumber)
4681 0 : dmstate->ctidAttno = i;
4682 : else
4683 : Assert(false);
4684 0 : dmstate->hasSystemCols = true;
4685 : }
4686 : else
4687 : {
4688 : /*
4689 : * We don't retrieve whole-row references to the target
4690 : * relation either.
4691 : */
4692 : Assert(attrno > 0);
4693 :
4694 16 : dmstate->attnoMap[attrno - 1] = i;
4695 : }
4696 : }
4697 30 : i++;
4698 : }
4699 2 : }
4700 :
4701 : /*
4702 : * Extract and return an updated/deleted tuple from a scan tuple.
4703 : */
4704 : static TupleTableSlot *
4705 32 : apply_returning_filter(PgFdwDirectModifyState *dmstate,
4706 : ResultRelInfo *resultRelInfo,
4707 : TupleTableSlot *slot,
4708 : EState *estate)
4709 : {
4710 32 : TupleDesc resultTupType = RelationGetDescr(dmstate->resultRel);
4711 : TupleTableSlot *resultSlot;
4712 : Datum *values;
4713 : bool *isnull;
4714 : Datum *old_values;
4715 : bool *old_isnull;
4716 : int i;
4717 :
4718 : /*
4719 : * Use the return tuple slot as a place to store the result tuple.
4720 : */
4721 32 : resultSlot = ExecGetReturningSlot(estate, resultRelInfo);
4722 :
4723 : /*
4724 : * Extract all the values of the scan tuple.
4725 : */
4726 32 : slot_getallattrs(slot);
4727 32 : old_values = slot->tts_values;
4728 32 : old_isnull = slot->tts_isnull;
4729 :
4730 : /*
4731 : * Prepare to build the result tuple.
4732 : */
4733 32 : ExecClearTuple(resultSlot);
4734 32 : values = resultSlot->tts_values;
4735 32 : isnull = resultSlot->tts_isnull;
4736 :
4737 : /*
4738 : * Transpose data into proper fields of the result tuple.
4739 : */
4740 320 : for (i = 0; i < resultTupType->natts; i++)
4741 : {
4742 288 : int j = dmstate->attnoMap[i];
4743 :
4744 288 : if (j == 0)
4745 : {
4746 32 : values[i] = (Datum) 0;
4747 32 : isnull[i] = true;
4748 : }
4749 : else
4750 : {
4751 256 : values[i] = old_values[j - 1];
4752 256 : isnull[i] = old_isnull[j - 1];
4753 : }
4754 : }
4755 :
4756 : /*
4757 : * Build the virtual tuple.
4758 : */
4759 32 : ExecStoreVirtualTuple(resultSlot);
4760 :
4761 : /*
4762 : * If we have any system columns to return, materialize a heap tuple in
4763 : * the slot from column values set above and install system columns in
4764 : * that tuple.
4765 : */
4766 32 : if (dmstate->hasSystemCols)
4767 : {
4768 0 : HeapTuple resultTup = ExecFetchSlotHeapTuple(resultSlot, true, NULL);
4769 :
4770 : /* ctid */
4771 0 : if (dmstate->ctidAttno)
4772 : {
4773 0 : ItemPointer ctid = NULL;
4774 :
4775 0 : ctid = (ItemPointer) DatumGetPointer(old_values[dmstate->ctidAttno - 1]);
4776 0 : resultTup->t_self = *ctid;
4777 : }
4778 :
4779 : /*
4780 : * And remaining columns
4781 : *
4782 : * Note: since we currently don't allow the target relation to appear
4783 : * on the nullable side of an outer join, any system columns wouldn't
4784 : * go to NULL.
4785 : *
4786 : * Note: no need to care about tableoid here because it will be
4787 : * initialized in ExecProcessReturning().
4788 : */
4789 0 : HeapTupleHeaderSetXmin(resultTup->t_data, InvalidTransactionId);
4790 0 : HeapTupleHeaderSetXmax(resultTup->t_data, InvalidTransactionId);
4791 0 : HeapTupleHeaderSetCmin(resultTup->t_data, InvalidTransactionId);
4792 : }
4793 :
4794 : /*
4795 : * And return the result tuple.
4796 : */
4797 32 : return resultSlot;
4798 : }
4799 :
4800 : /*
4801 : * Prepare for processing of parameters used in remote query.
4802 : */
4803 : static void
4804 38 : prepare_query_params(PlanState *node,
4805 : List *fdw_exprs,
4806 : int numParams,
4807 : FmgrInfo **param_flinfo,
4808 : List **param_exprs,
4809 : const char ***param_values)
4810 : {
4811 : int i;
4812 : ListCell *lc;
4813 :
4814 : Assert(numParams > 0);
4815 :
4816 : /* Prepare for output conversion of parameters used in remote query. */
4817 38 : *param_flinfo = (FmgrInfo *) palloc0(sizeof(FmgrInfo) * numParams);
4818 :
4819 38 : i = 0;
4820 78 : foreach(lc, fdw_exprs)
4821 : {
4822 40 : Node *param_expr = (Node *) lfirst(lc);
4823 : Oid typefnoid;
4824 : bool isvarlena;
4825 :
4826 40 : getTypeOutputInfo(exprType(param_expr), &typefnoid, &isvarlena);
4827 40 : fmgr_info(typefnoid, &(*param_flinfo)[i]);
4828 40 : i++;
4829 : }
4830 :
4831 : /*
4832 : * Prepare remote-parameter expressions for evaluation. (Note: in
4833 : * practice, we expect that all these expressions will be just Params, so
4834 : * we could possibly do something more efficient than using the full
4835 : * expression-eval machinery for this. But probably there would be little
4836 : * benefit, and it'd require postgres_fdw to know more than is desirable
4837 : * about Param evaluation.)
4838 : */
4839 38 : *param_exprs = ExecInitExprList(fdw_exprs, node);
4840 :
4841 : /* Allocate buffer for text form of query parameters. */
4842 38 : *param_values = (const char **) palloc0(numParams * sizeof(char *));
4843 38 : }
4844 :
4845 : /*
4846 : * Construct array of query parameter values in text format.
4847 : */
4848 : static void
4849 694 : process_query_params(ExprContext *econtext,
4850 : FmgrInfo *param_flinfo,
4851 : List *param_exprs,
4852 : const char **param_values)
4853 : {
4854 : int nestlevel;
4855 : int i;
4856 : ListCell *lc;
4857 :
4858 694 : nestlevel = set_transmission_modes();
4859 :
4860 694 : i = 0;
4861 1788 : foreach(lc, param_exprs)
4862 : {
4863 1094 : ExprState *expr_state = (ExprState *) lfirst(lc);
4864 : Datum expr_value;
4865 : bool isNull;
4866 :
4867 : /* Evaluate the parameter expression */
4868 1094 : expr_value = ExecEvalExpr(expr_state, econtext, &isNull);
4869 :
4870 : /*
4871 : * Get string representation of each parameter value by invoking
4872 : * type-specific output function, unless the value is null.
4873 : */
4874 1094 : if (isNull)
4875 0 : param_values[i] = NULL;
4876 : else
4877 1094 : param_values[i] = OutputFunctionCall(¶m_flinfo[i], expr_value);
4878 :
4879 1094 : i++;
4880 : }
4881 :
4882 694 : reset_transmission_modes(nestlevel);
4883 694 : }
4884 :
4885 : /*
4886 : * postgresAnalyzeForeignTable
4887 : * Test whether analyzing this foreign table is supported
4888 : */
4889 : static bool
4890 84 : postgresAnalyzeForeignTable(Relation relation,
4891 : AcquireSampleRowsFunc *func,
4892 : BlockNumber *totalpages)
4893 : {
4894 : ForeignTable *table;
4895 : UserMapping *user;
4896 : PGconn *conn;
4897 : StringInfoData sql;
4898 : PGresult *res;
4899 :
4900 : /* Return the row-analysis function pointer */
4901 84 : *func = postgresAcquireSampleRowsFunc;
4902 :
4903 : /*
4904 : * Now we have to get the number of pages. It's annoying that the ANALYZE
4905 : * API requires us to return that now, because it forces some duplication
4906 : * of effort between this routine and postgresAcquireSampleRowsFunc. But
4907 : * it's probably not worth redefining that API at this point.
4908 : */
4909 :
4910 : /*
4911 : * Get the connection to use. We do the remote access as the table's
4912 : * owner, even if the ANALYZE was started by some other user.
4913 : */
4914 84 : table = GetForeignTable(RelationGetRelid(relation));
4915 84 : user = GetUserMapping(relation->rd_rel->relowner, table->serverid);
4916 84 : conn = GetConnection(user, false, NULL);
4917 :
4918 : /*
4919 : * Construct command to get page count for relation.
4920 : */
4921 84 : initStringInfo(&sql);
4922 84 : deparseAnalyzeSizeSql(&sql, relation);
4923 :
4924 84 : res = pgfdw_exec_query(conn, sql.data, NULL);
4925 84 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
4926 0 : pgfdw_report_error(res, conn, sql.data);
4927 :
4928 84 : if (PQntuples(res) != 1 || PQnfields(res) != 1)
4929 0 : elog(ERROR, "unexpected result from deparseAnalyzeSizeSql query");
4930 84 : *totalpages = strtoul(PQgetvalue(res, 0, 0), NULL, 10);
4931 84 : PQclear(res);
4932 :
4933 84 : ReleaseConnection(conn);
4934 :
4935 84 : return true;
4936 : }
4937 :
4938 : /*
4939 : * postgresGetAnalyzeInfoForForeignTable
4940 : * Count tuples in foreign table (just get pg_class.reltuples).
4941 : *
4942 : * can_tablesample determines if the remote relation supports acquiring the
4943 : * sample using TABLESAMPLE.
4944 : */
4945 : static double
4946 84 : postgresGetAnalyzeInfoForForeignTable(Relation relation, bool *can_tablesample)
4947 : {
4948 : ForeignTable *table;
4949 : UserMapping *user;
4950 : PGconn *conn;
4951 : StringInfoData sql;
4952 : PGresult *res;
4953 : double reltuples;
4954 : char relkind;
4955 :
4956 : /* assume the remote relation does not support TABLESAMPLE */
4957 84 : *can_tablesample = false;
4958 :
4959 : /*
4960 : * Get the connection to use. We do the remote access as the table's
4961 : * owner, even if the ANALYZE was started by some other user.
4962 : */
4963 84 : table = GetForeignTable(RelationGetRelid(relation));
4964 84 : user = GetUserMapping(relation->rd_rel->relowner, table->serverid);
4965 84 : conn = GetConnection(user, false, NULL);
4966 :
4967 : /*
4968 : * Construct command to get page count for relation.
4969 : */
4970 84 : initStringInfo(&sql);
4971 84 : deparseAnalyzeInfoSql(&sql, relation);
4972 :
4973 84 : res = pgfdw_exec_query(conn, sql.data, NULL);
4974 84 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
4975 0 : pgfdw_report_error(res, conn, sql.data);
4976 :
4977 84 : if (PQntuples(res) != 1 || PQnfields(res) != 2)
4978 0 : elog(ERROR, "unexpected result from deparseAnalyzeInfoSql query");
4979 84 : reltuples = strtod(PQgetvalue(res, 0, 0), NULL);
4980 84 : relkind = *(PQgetvalue(res, 0, 1));
4981 84 : PQclear(res);
4982 :
4983 84 : ReleaseConnection(conn);
4984 :
4985 : /* TABLESAMPLE is supported only for regular tables and matviews */
4986 0 : *can_tablesample = (relkind == RELKIND_RELATION ||
4987 84 : relkind == RELKIND_MATVIEW ||
4988 : relkind == RELKIND_PARTITIONED_TABLE);
4989 :
4990 84 : return reltuples;
4991 : }
4992 :
4993 : /*
4994 : * Acquire a random sample of rows from foreign table managed by postgres_fdw.
4995 : *
4996 : * Selected rows are returned in the caller-allocated array rows[],
4997 : * which must have at least targrows entries.
4998 : * The actual number of rows selected is returned as the function result.
4999 : * We also count the total number of rows in the table and return it into
5000 : * *totalrows. Note that *totaldeadrows is always set to 0.
5001 : *
5002 : * Note that the returned list of rows is not always in order by physical
5003 : * position in the table. Therefore, correlation estimates derived later
5004 : * may be meaningless, but it's OK because we don't use the estimates
5005 : * currently (the planner only pays attention to correlation for indexscans).
5006 : */
5007 : static int
5008 84 : postgresAcquireSampleRowsFunc(Relation relation, int elevel,
5009 : HeapTuple *rows, int targrows,
5010 : double *totalrows,
5011 : double *totaldeadrows)
5012 : {
5013 : PgFdwAnalyzeState astate;
5014 : ForeignTable *table;
5015 : ForeignServer *server;
5016 : UserMapping *user;
5017 : PGconn *conn;
5018 : int server_version_num;
5019 84 : PgFdwSamplingMethod method = ANALYZE_SAMPLE_AUTO; /* auto is default */
5020 84 : double sample_frac = -1.0;
5021 84 : double reltuples = -1.0;
5022 : unsigned int cursor_number;
5023 : StringInfoData sql;
5024 : PGresult *res;
5025 : char fetch_sql[64];
5026 : int fetch_size;
5027 : ListCell *lc;
5028 :
5029 : /* Initialize workspace state */
5030 84 : astate.rel = relation;
5031 84 : astate.attinmeta = TupleDescGetAttInMetadata(RelationGetDescr(relation));
5032 :
5033 84 : astate.rows = rows;
5034 84 : astate.targrows = targrows;
5035 84 : astate.numrows = 0;
5036 84 : astate.samplerows = 0;
5037 84 : astate.rowstoskip = -1; /* -1 means not set yet */
5038 84 : reservoir_init_selection_state(&astate.rstate, targrows);
5039 :
5040 : /* Remember ANALYZE context, and create a per-tuple temp context */
5041 84 : astate.anl_cxt = CurrentMemoryContext;
5042 84 : astate.temp_cxt = AllocSetContextCreate(CurrentMemoryContext,
5043 : "postgres_fdw temporary data",
5044 : ALLOCSET_SMALL_SIZES);
5045 :
5046 : /*
5047 : * Get the connection to use. We do the remote access as the table's
5048 : * owner, even if the ANALYZE was started by some other user.
5049 : */
5050 84 : table = GetForeignTable(RelationGetRelid(relation));
5051 84 : server = GetForeignServer(table->serverid);
5052 84 : user = GetUserMapping(relation->rd_rel->relowner, table->serverid);
5053 84 : conn = GetConnection(user, false, NULL);
5054 :
5055 : /* We'll need server version, so fetch it now. */
5056 84 : server_version_num = PQserverVersion(conn);
5057 :
5058 : /*
5059 : * What sampling method should we use?
5060 : */
5061 368 : foreach(lc, server->options)
5062 : {
5063 284 : DefElem *def = (DefElem *) lfirst(lc);
5064 :
5065 284 : if (strcmp(def->defname, "analyze_sampling") == 0)
5066 : {
5067 0 : char *value = defGetString(def);
5068 :
5069 0 : if (strcmp(value, "off") == 0)
5070 0 : method = ANALYZE_SAMPLE_OFF;
5071 0 : else if (strcmp(value, "auto") == 0)
5072 0 : method = ANALYZE_SAMPLE_AUTO;
5073 0 : else if (strcmp(value, "random") == 0)
5074 0 : method = ANALYZE_SAMPLE_RANDOM;
5075 0 : else if (strcmp(value, "system") == 0)
5076 0 : method = ANALYZE_SAMPLE_SYSTEM;
5077 0 : else if (strcmp(value, "bernoulli") == 0)
5078 0 : method = ANALYZE_SAMPLE_BERNOULLI;
5079 :
5080 0 : break;
5081 : }
5082 : }
5083 :
5084 196 : foreach(lc, table->options)
5085 : {
5086 112 : DefElem *def = (DefElem *) lfirst(lc);
5087 :
5088 112 : if (strcmp(def->defname, "analyze_sampling") == 0)
5089 : {
5090 0 : char *value = defGetString(def);
5091 :
5092 0 : if (strcmp(value, "off") == 0)
5093 0 : method = ANALYZE_SAMPLE_OFF;
5094 0 : else if (strcmp(value, "auto") == 0)
5095 0 : method = ANALYZE_SAMPLE_AUTO;
5096 0 : else if (strcmp(value, "random") == 0)
5097 0 : method = ANALYZE_SAMPLE_RANDOM;
5098 0 : else if (strcmp(value, "system") == 0)
5099 0 : method = ANALYZE_SAMPLE_SYSTEM;
5100 0 : else if (strcmp(value, "bernoulli") == 0)
5101 0 : method = ANALYZE_SAMPLE_BERNOULLI;
5102 :
5103 0 : break;
5104 : }
5105 : }
5106 :
5107 : /*
5108 : * Error-out if explicitly required one of the TABLESAMPLE methods, but
5109 : * the server does not support it.
5110 : */
5111 84 : if ((server_version_num < 95000) &&
5112 0 : (method == ANALYZE_SAMPLE_SYSTEM ||
5113 : method == ANALYZE_SAMPLE_BERNOULLI))
5114 0 : ereport(ERROR,
5115 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
5116 : errmsg("remote server does not support TABLESAMPLE feature")));
5117 :
5118 : /*
5119 : * If we've decided to do remote sampling, calculate the sampling rate. We
5120 : * need to get the number of tuples from the remote server, but skip that
5121 : * network round-trip if not needed.
5122 : */
5123 84 : if (method != ANALYZE_SAMPLE_OFF)
5124 : {
5125 : bool can_tablesample;
5126 :
5127 84 : reltuples = postgresGetAnalyzeInfoForForeignTable(relation,
5128 : &can_tablesample);
5129 :
5130 : /*
5131 : * Make sure we're not choosing TABLESAMPLE when the remote relation
5132 : * does not support that. But only do this for "auto" - if the user
5133 : * explicitly requested BERNOULLI/SYSTEM, it's better to fail.
5134 : */
5135 84 : if (!can_tablesample && (method == ANALYZE_SAMPLE_AUTO))
5136 0 : method = ANALYZE_SAMPLE_RANDOM;
5137 :
5138 : /*
5139 : * Remote's reltuples could be 0 or -1 if the table has never been
5140 : * vacuumed/analyzed. In that case, disable sampling after all.
5141 : */
5142 84 : if ((reltuples <= 0) || (targrows >= reltuples))
5143 84 : method = ANALYZE_SAMPLE_OFF;
5144 : else
5145 : {
5146 : /*
5147 : * All supported sampling methods require sampling rate, not
5148 : * target rows directly, so we calculate that using the remote
5149 : * reltuples value. That's imperfect, because it might be off a
5150 : * good deal, but that's not something we can (or should) address
5151 : * here.
5152 : *
5153 : * If reltuples is too low (i.e. when table grew), we'll end up
5154 : * sampling more rows - but then we'll apply the local sampling,
5155 : * so we get the expected sample size. This is the same outcome as
5156 : * without remote sampling.
5157 : *
5158 : * If reltuples is too high (e.g. after bulk DELETE), we will end
5159 : * up sampling too few rows.
5160 : *
5161 : * We can't really do much better here - we could try sampling a
5162 : * bit more rows, but we don't know how off the reltuples value is
5163 : * so how much is "a bit more"?
5164 : *
5165 : * Furthermore, the targrows value for partitions is determined
5166 : * based on table size (relpages), which can be off in different
5167 : * ways too. Adjusting the sampling rate here might make the issue
5168 : * worse.
5169 : */
5170 0 : sample_frac = targrows / reltuples;
5171 :
5172 : /*
5173 : * We should never get sampling rate outside the valid range
5174 : * (between 0.0 and 1.0), because those cases should be covered by
5175 : * the previous branch that sets ANALYZE_SAMPLE_OFF.
5176 : */
5177 : Assert(sample_frac >= 0.0 && sample_frac <= 1.0);
5178 : }
5179 : }
5180 :
5181 : /*
5182 : * For "auto" method, pick the one we believe is best. For servers with
5183 : * TABLESAMPLE support we pick BERNOULLI, for old servers we fall-back to
5184 : * random() to at least reduce network transfer.
5185 : */
5186 84 : if (method == ANALYZE_SAMPLE_AUTO)
5187 : {
5188 0 : if (server_version_num < 95000)
5189 0 : method = ANALYZE_SAMPLE_RANDOM;
5190 : else
5191 0 : method = ANALYZE_SAMPLE_BERNOULLI;
5192 : }
5193 :
5194 : /*
5195 : * Construct cursor that retrieves whole rows from remote.
5196 : */
5197 84 : cursor_number = GetCursorNumber(conn);
5198 84 : initStringInfo(&sql);
5199 84 : appendStringInfo(&sql, "DECLARE c%u CURSOR FOR ", cursor_number);
5200 :
5201 84 : deparseAnalyzeSql(&sql, relation, method, sample_frac, &astate.retrieved_attrs);
5202 :
5203 84 : res = pgfdw_exec_query(conn, sql.data, NULL);
5204 84 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
5205 0 : pgfdw_report_error(res, conn, sql.data);
5206 84 : PQclear(res);
5207 :
5208 : /*
5209 : * Determine the fetch size. The default is arbitrary, but shouldn't be
5210 : * enormous.
5211 : */
5212 84 : fetch_size = 100;
5213 368 : foreach(lc, server->options)
5214 : {
5215 284 : DefElem *def = (DefElem *) lfirst(lc);
5216 :
5217 284 : if (strcmp(def->defname, "fetch_size") == 0)
5218 : {
5219 0 : (void) parse_int(defGetString(def), &fetch_size, 0, NULL);
5220 0 : break;
5221 : }
5222 : }
5223 196 : foreach(lc, table->options)
5224 : {
5225 112 : DefElem *def = (DefElem *) lfirst(lc);
5226 :
5227 112 : if (strcmp(def->defname, "fetch_size") == 0)
5228 : {
5229 0 : (void) parse_int(defGetString(def), &fetch_size, 0, NULL);
5230 0 : break;
5231 : }
5232 : }
5233 :
5234 : /* Construct command to fetch rows from remote. */
5235 84 : snprintf(fetch_sql, sizeof(fetch_sql), "FETCH %d FROM c%u",
5236 : fetch_size, cursor_number);
5237 :
5238 : /* Retrieve and process rows a batch at a time. */
5239 : for (;;)
5240 344 : {
5241 : int numrows;
5242 : int i;
5243 :
5244 : /* Allow users to cancel long query */
5245 428 : CHECK_FOR_INTERRUPTS();
5246 :
5247 : /*
5248 : * XXX possible future improvement: if rowstoskip is large, we could
5249 : * issue a MOVE rather than physically fetching the rows, then just
5250 : * adjust rowstoskip and samplerows appropriately.
5251 : */
5252 :
5253 : /* Fetch some rows */
5254 428 : res = pgfdw_exec_query(conn, fetch_sql, NULL);
5255 : /* On error, report the original query, not the FETCH. */
5256 428 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
5257 0 : pgfdw_report_error(res, conn, sql.data);
5258 :
5259 : /* Process whatever we got. */
5260 428 : numrows = PQntuples(res);
5261 35882 : for (i = 0; i < numrows; i++)
5262 35456 : analyze_row_processor(res, i, &astate);
5263 :
5264 426 : PQclear(res);
5265 :
5266 : /* Must be EOF if we didn't get all the rows requested. */
5267 426 : if (numrows < fetch_size)
5268 82 : break;
5269 : }
5270 :
5271 : /* Close the cursor, just to be tidy. */
5272 82 : close_cursor(conn, cursor_number, NULL);
5273 :
5274 82 : ReleaseConnection(conn);
5275 :
5276 : /* We assume that we have no dead tuple. */
5277 82 : *totaldeadrows = 0.0;
5278 :
5279 : /*
5280 : * Without sampling, we've retrieved all living tuples from foreign
5281 : * server, so report that as totalrows. Otherwise use the reltuples
5282 : * estimate we got from the remote side.
5283 : */
5284 82 : if (method == ANALYZE_SAMPLE_OFF)
5285 82 : *totalrows = astate.samplerows;
5286 : else
5287 0 : *totalrows = reltuples;
5288 :
5289 : /*
5290 : * Emit some interesting relation info
5291 : */
5292 82 : ereport(elevel,
5293 : (errmsg("\"%s\": table contains %.0f rows, %d rows in sample",
5294 : RelationGetRelationName(relation),
5295 : *totalrows, astate.numrows)));
5296 :
5297 82 : return astate.numrows;
5298 : }
5299 :
5300 : /*
5301 : * Collect sample rows from the result of query.
5302 : * - Use all tuples in sample until target # of samples are collected.
5303 : * - Subsequently, replace already-sampled tuples randomly.
5304 : */
5305 : static void
5306 35456 : analyze_row_processor(PGresult *res, int row, PgFdwAnalyzeState *astate)
5307 : {
5308 35456 : int targrows = astate->targrows;
5309 : int pos; /* array index to store tuple in */
5310 : MemoryContext oldcontext;
5311 :
5312 : /* Always increment sample row counter. */
5313 35456 : astate->samplerows += 1;
5314 :
5315 : /*
5316 : * Determine the slot where this sample row should be stored. Set pos to
5317 : * negative value to indicate the row should be skipped.
5318 : */
5319 35456 : if (astate->numrows < targrows)
5320 : {
5321 : /* First targrows rows are always included into the sample */
5322 35456 : pos = astate->numrows++;
5323 : }
5324 : else
5325 : {
5326 : /*
5327 : * Now we start replacing tuples in the sample until we reach the end
5328 : * of the relation. Same algorithm as in acquire_sample_rows in
5329 : * analyze.c; see Jeff Vitter's paper.
5330 : */
5331 0 : if (astate->rowstoskip < 0)
5332 0 : astate->rowstoskip = reservoir_get_next_S(&astate->rstate, astate->samplerows, targrows);
5333 :
5334 0 : if (astate->rowstoskip <= 0)
5335 : {
5336 : /* Choose a random reservoir element to replace. */
5337 0 : pos = (int) (targrows * sampler_random_fract(&astate->rstate.randstate));
5338 : Assert(pos >= 0 && pos < targrows);
5339 0 : heap_freetuple(astate->rows[pos]);
5340 : }
5341 : else
5342 : {
5343 : /* Skip this tuple. */
5344 0 : pos = -1;
5345 : }
5346 :
5347 0 : astate->rowstoskip -= 1;
5348 : }
5349 :
5350 35456 : if (pos >= 0)
5351 : {
5352 : /*
5353 : * Create sample tuple from current result row, and store it in the
5354 : * position determined above. The tuple has to be created in anl_cxt.
5355 : */
5356 35456 : oldcontext = MemoryContextSwitchTo(astate->anl_cxt);
5357 :
5358 35456 : astate->rows[pos] = make_tuple_from_result_row(res, row,
5359 : astate->rel,
5360 : astate->attinmeta,
5361 : astate->retrieved_attrs,
5362 : NULL,
5363 : astate->temp_cxt);
5364 :
5365 35454 : MemoryContextSwitchTo(oldcontext);
5366 : }
5367 35454 : }
5368 :
5369 : /*
5370 : * Import a foreign schema
5371 : */
5372 : static List *
5373 20 : postgresImportForeignSchema(ImportForeignSchemaStmt *stmt, Oid serverOid)
5374 : {
5375 20 : List *commands = NIL;
5376 20 : bool import_collate = true;
5377 20 : bool import_default = false;
5378 20 : bool import_generated = true;
5379 20 : bool import_not_null = true;
5380 : ForeignServer *server;
5381 : UserMapping *mapping;
5382 : PGconn *conn;
5383 : StringInfoData buf;
5384 : PGresult *res;
5385 : int numrows,
5386 : i;
5387 : ListCell *lc;
5388 :
5389 : /* Parse statement options */
5390 28 : foreach(lc, stmt->options)
5391 : {
5392 8 : DefElem *def = (DefElem *) lfirst(lc);
5393 :
5394 8 : if (strcmp(def->defname, "import_collate") == 0)
5395 2 : import_collate = defGetBoolean(def);
5396 6 : else if (strcmp(def->defname, "import_default") == 0)
5397 2 : import_default = defGetBoolean(def);
5398 4 : else if (strcmp(def->defname, "import_generated") == 0)
5399 2 : import_generated = defGetBoolean(def);
5400 2 : else if (strcmp(def->defname, "import_not_null") == 0)
5401 2 : import_not_null = defGetBoolean(def);
5402 : else
5403 0 : ereport(ERROR,
5404 : (errcode(ERRCODE_FDW_INVALID_OPTION_NAME),
5405 : errmsg("invalid option \"%s\"", def->defname)));
5406 : }
5407 :
5408 : /*
5409 : * Get connection to the foreign server. Connection manager will
5410 : * establish new connection if necessary.
5411 : */
5412 20 : server = GetForeignServer(serverOid);
5413 20 : mapping = GetUserMapping(GetUserId(), server->serverid);
5414 20 : conn = GetConnection(mapping, false, NULL);
5415 :
5416 : /* Don't attempt to import collation if remote server hasn't got it */
5417 20 : if (PQserverVersion(conn) < 90100)
5418 0 : import_collate = false;
5419 :
5420 : /* Create workspace for strings */
5421 20 : initStringInfo(&buf);
5422 :
5423 : /* Check that the schema really exists */
5424 20 : appendStringInfoString(&buf, "SELECT 1 FROM pg_catalog.pg_namespace WHERE nspname = ");
5425 20 : deparseStringLiteral(&buf, stmt->remote_schema);
5426 :
5427 20 : res = pgfdw_exec_query(conn, buf.data, NULL);
5428 20 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
5429 0 : pgfdw_report_error(res, conn, buf.data);
5430 :
5431 20 : if (PQntuples(res) != 1)
5432 2 : ereport(ERROR,
5433 : (errcode(ERRCODE_FDW_SCHEMA_NOT_FOUND),
5434 : errmsg("schema \"%s\" is not present on foreign server \"%s\"",
5435 : stmt->remote_schema, server->servername)));
5436 :
5437 18 : PQclear(res);
5438 18 : resetStringInfo(&buf);
5439 :
5440 : /*
5441 : * Fetch all table data from this schema, possibly restricted by EXCEPT or
5442 : * LIMIT TO. (We don't actually need to pay any attention to EXCEPT/LIMIT
5443 : * TO here, because the core code will filter the statements we return
5444 : * according to those lists anyway. But it should save a few cycles to
5445 : * not process excluded tables in the first place.)
5446 : *
5447 : * Import table data for partitions only when they are explicitly
5448 : * specified in LIMIT TO clause. Otherwise ignore them and only include
5449 : * the definitions of the root partitioned tables to allow access to the
5450 : * complete remote data set locally in the schema imported.
5451 : *
5452 : * Note: because we run the connection with search_path restricted to
5453 : * pg_catalog, the format_type() and pg_get_expr() outputs will always
5454 : * include a schema name for types/functions in other schemas, which is
5455 : * what we want.
5456 : */
5457 18 : appendStringInfoString(&buf,
5458 : "SELECT relname, "
5459 : " attname, "
5460 : " format_type(atttypid, atttypmod), "
5461 : " attnotnull, "
5462 : " pg_get_expr(adbin, adrelid), ");
5463 :
5464 : /* Generated columns are supported since Postgres 12 */
5465 18 : if (PQserverVersion(conn) >= 120000)
5466 18 : appendStringInfoString(&buf,
5467 : " attgenerated, ");
5468 : else
5469 0 : appendStringInfoString(&buf,
5470 : " NULL, ");
5471 :
5472 18 : if (import_collate)
5473 16 : appendStringInfoString(&buf,
5474 : " collname, "
5475 : " collnsp.nspname ");
5476 : else
5477 2 : appendStringInfoString(&buf,
5478 : " NULL, NULL ");
5479 :
5480 18 : appendStringInfoString(&buf,
5481 : "FROM pg_class c "
5482 : " JOIN pg_namespace n ON "
5483 : " relnamespace = n.oid "
5484 : " LEFT JOIN pg_attribute a ON "
5485 : " attrelid = c.oid AND attnum > 0 "
5486 : " AND NOT attisdropped "
5487 : " LEFT JOIN pg_attrdef ad ON "
5488 : " adrelid = c.oid AND adnum = attnum ");
5489 :
5490 18 : if (import_collate)
5491 16 : appendStringInfoString(&buf,
5492 : " LEFT JOIN pg_collation coll ON "
5493 : " coll.oid = attcollation "
5494 : " LEFT JOIN pg_namespace collnsp ON "
5495 : " collnsp.oid = collnamespace ");
5496 :
5497 18 : appendStringInfoString(&buf,
5498 : "WHERE c.relkind IN ("
5499 : CppAsString2(RELKIND_RELATION) ","
5500 : CppAsString2(RELKIND_VIEW) ","
5501 : CppAsString2(RELKIND_FOREIGN_TABLE) ","
5502 : CppAsString2(RELKIND_MATVIEW) ","
5503 : CppAsString2(RELKIND_PARTITIONED_TABLE) ") "
5504 : " AND n.nspname = ");
5505 18 : deparseStringLiteral(&buf, stmt->remote_schema);
5506 :
5507 : /* Partitions are supported since Postgres 10 */
5508 18 : if (PQserverVersion(conn) >= 100000 &&
5509 18 : stmt->list_type != FDW_IMPORT_SCHEMA_LIMIT_TO)
5510 10 : appendStringInfoString(&buf, " AND NOT c.relispartition ");
5511 :
5512 : /* Apply restrictions for LIMIT TO and EXCEPT */
5513 18 : if (stmt->list_type == FDW_IMPORT_SCHEMA_LIMIT_TO ||
5514 10 : stmt->list_type == FDW_IMPORT_SCHEMA_EXCEPT)
5515 : {
5516 10 : bool first_item = true;
5517 :
5518 10 : appendStringInfoString(&buf, " AND c.relname ");
5519 10 : if (stmt->list_type == FDW_IMPORT_SCHEMA_EXCEPT)
5520 2 : appendStringInfoString(&buf, "NOT ");
5521 10 : appendStringInfoString(&buf, "IN (");
5522 :
5523 : /* Append list of table names within IN clause */
5524 30 : foreach(lc, stmt->table_list)
5525 : {
5526 20 : RangeVar *rv = (RangeVar *) lfirst(lc);
5527 :
5528 20 : if (first_item)
5529 10 : first_item = false;
5530 : else
5531 10 : appendStringInfoString(&buf, ", ");
5532 20 : deparseStringLiteral(&buf, rv->relname);
5533 : }
5534 10 : appendStringInfoChar(&buf, ')');
5535 : }
5536 :
5537 : /* Append ORDER BY at the end of query to ensure output ordering */
5538 18 : appendStringInfoString(&buf, " ORDER BY c.relname, a.attnum");
5539 :
5540 : /* Fetch the data */
5541 18 : res = pgfdw_exec_query(conn, buf.data, NULL);
5542 18 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
5543 0 : pgfdw_report_error(res, conn, buf.data);
5544 :
5545 : /* Process results */
5546 18 : numrows = PQntuples(res);
5547 : /* note: incrementation of i happens in inner loop's while() test */
5548 94 : for (i = 0; i < numrows;)
5549 : {
5550 76 : char *tablename = PQgetvalue(res, i, 0);
5551 76 : bool first_item = true;
5552 :
5553 76 : resetStringInfo(&buf);
5554 76 : appendStringInfo(&buf, "CREATE FOREIGN TABLE %s (\n",
5555 : quote_identifier(tablename));
5556 :
5557 : /* Scan all rows for this table */
5558 : do
5559 : {
5560 : char *attname;
5561 : char *typename;
5562 : char *attnotnull;
5563 : char *attgenerated;
5564 : char *attdefault;
5565 : char *collname;
5566 : char *collnamespace;
5567 :
5568 : /* If table has no columns, we'll see nulls here */
5569 150 : if (PQgetisnull(res, i, 1))
5570 10 : continue;
5571 :
5572 140 : attname = PQgetvalue(res, i, 1);
5573 140 : typename = PQgetvalue(res, i, 2);
5574 140 : attnotnull = PQgetvalue(res, i, 3);
5575 140 : attdefault = PQgetisnull(res, i, 4) ? NULL :
5576 30 : PQgetvalue(res, i, 4);
5577 140 : attgenerated = PQgetisnull(res, i, 5) ? NULL :
5578 140 : PQgetvalue(res, i, 5);
5579 140 : collname = PQgetisnull(res, i, 6) ? NULL :
5580 38 : PQgetvalue(res, i, 6);
5581 140 : collnamespace = PQgetisnull(res, i, 7) ? NULL :
5582 38 : PQgetvalue(res, i, 7);
5583 :
5584 140 : if (first_item)
5585 66 : first_item = false;
5586 : else
5587 74 : appendStringInfoString(&buf, ",\n");
5588 :
5589 : /* Print column name and type */
5590 140 : appendStringInfo(&buf, " %s %s",
5591 : quote_identifier(attname),
5592 : typename);
5593 :
5594 : /*
5595 : * Add column_name option so that renaming the foreign table's
5596 : * column doesn't break the association to the underlying column.
5597 : */
5598 140 : appendStringInfoString(&buf, " OPTIONS (column_name ");
5599 140 : deparseStringLiteral(&buf, attname);
5600 140 : appendStringInfoChar(&buf, ')');
5601 :
5602 : /* Add COLLATE if needed */
5603 140 : if (import_collate && collname != NULL && collnamespace != NULL)
5604 38 : appendStringInfo(&buf, " COLLATE %s.%s",
5605 : quote_identifier(collnamespace),
5606 : quote_identifier(collname));
5607 :
5608 : /* Add DEFAULT if needed */
5609 140 : if (import_default && attdefault != NULL &&
5610 6 : (!attgenerated || !attgenerated[0]))
5611 4 : appendStringInfo(&buf, " DEFAULT %s", attdefault);
5612 :
5613 : /* Add GENERATED if needed */
5614 140 : if (import_generated && attgenerated != NULL &&
5615 114 : attgenerated[0] == ATTRIBUTE_GENERATED_STORED)
5616 : {
5617 : Assert(attdefault != NULL);
5618 8 : appendStringInfo(&buf,
5619 : " GENERATED ALWAYS AS (%s) STORED",
5620 : attdefault);
5621 : }
5622 :
5623 : /* Add NOT NULL if needed */
5624 140 : if (import_not_null && attnotnull[0] == 't')
5625 8 : appendStringInfoString(&buf, " NOT NULL");
5626 : }
5627 150 : while (++i < numrows &&
5628 132 : strcmp(PQgetvalue(res, i, 0), tablename) == 0);
5629 :
5630 : /*
5631 : * Add server name and table-level options. We specify remote schema
5632 : * and table name as options (the latter to ensure that renaming the
5633 : * foreign table doesn't break the association).
5634 : */
5635 76 : appendStringInfo(&buf, "\n) SERVER %s\nOPTIONS (",
5636 76 : quote_identifier(server->servername));
5637 :
5638 76 : appendStringInfoString(&buf, "schema_name ");
5639 76 : deparseStringLiteral(&buf, stmt->remote_schema);
5640 76 : appendStringInfoString(&buf, ", table_name ");
5641 76 : deparseStringLiteral(&buf, tablename);
5642 :
5643 76 : appendStringInfoString(&buf, ");");
5644 :
5645 76 : commands = lappend(commands, pstrdup(buf.data));
5646 : }
5647 18 : PQclear(res);
5648 :
5649 18 : ReleaseConnection(conn);
5650 :
5651 18 : return commands;
5652 : }
5653 :
5654 : /*
5655 : * Check if reltarget is safe enough to push down semi-join. Reltarget is not
5656 : * safe, if it contains references to inner rel relids, which do not belong to
5657 : * outer rel.
5658 : */
5659 : static bool
5660 128 : semijoin_target_ok(PlannerInfo *root, RelOptInfo *joinrel, RelOptInfo *outerrel, RelOptInfo *innerrel)
5661 : {
5662 : List *vars;
5663 : ListCell *lc;
5664 128 : bool ok = true;
5665 :
5666 : Assert(joinrel->reltarget);
5667 :
5668 128 : vars = pull_var_clause((Node *) joinrel->reltarget->exprs, PVC_INCLUDE_PLACEHOLDERS);
5669 :
5670 886 : foreach(lc, vars)
5671 : {
5672 788 : Var *var = (Var *) lfirst(lc);
5673 :
5674 788 : if (!IsA(var, Var))
5675 0 : continue;
5676 :
5677 788 : if (bms_is_member(var->varno, innerrel->relids))
5678 : {
5679 : /*
5680 : * The planner can create semi-join, which refers to inner rel
5681 : * vars in its target list. However, we deparse semi-join as an
5682 : * exists() subquery, so can't handle references to inner rel in
5683 : * the target list.
5684 : */
5685 : Assert(!bms_is_member(var->varno, outerrel->relids));
5686 30 : ok = false;
5687 30 : break;
5688 : }
5689 : }
5690 128 : return ok;
5691 : }
5692 :
5693 : /*
5694 : * Assess whether the join between inner and outer relations can be pushed down
5695 : * to the foreign server. As a side effect, save information we obtain in this
5696 : * function to PgFdwRelationInfo passed in.
5697 : */
5698 : static bool
5699 784 : foreign_join_ok(PlannerInfo *root, RelOptInfo *joinrel, JoinType jointype,
5700 : RelOptInfo *outerrel, RelOptInfo *innerrel,
5701 : JoinPathExtraData *extra)
5702 : {
5703 : PgFdwRelationInfo *fpinfo;
5704 : PgFdwRelationInfo *fpinfo_o;
5705 : PgFdwRelationInfo *fpinfo_i;
5706 : ListCell *lc;
5707 : List *joinclauses;
5708 :
5709 : /*
5710 : * We support pushing down INNER, LEFT, RIGHT, FULL OUTER and SEMI joins.
5711 : * Constructing queries representing ANTI joins is hard, hence not
5712 : * considered right now.
5713 : */
5714 784 : if (jointype != JOIN_INNER && jointype != JOIN_LEFT &&
5715 258 : jointype != JOIN_RIGHT && jointype != JOIN_FULL &&
5716 : jointype != JOIN_SEMI)
5717 38 : return false;
5718 :
5719 : /*
5720 : * We can't push down semi-join if its reltarget is not safe
5721 : */
5722 746 : if ((jointype == JOIN_SEMI) && !semijoin_target_ok(root, joinrel, outerrel, innerrel))
5723 30 : return false;
5724 :
5725 : /*
5726 : * If either of the joining relations is marked as unsafe to pushdown, the
5727 : * join can not be pushed down.
5728 : */
5729 716 : fpinfo = (PgFdwRelationInfo *) joinrel->fdw_private;
5730 716 : fpinfo_o = (PgFdwRelationInfo *) outerrel->fdw_private;
5731 716 : fpinfo_i = (PgFdwRelationInfo *) innerrel->fdw_private;
5732 716 : if (!fpinfo_o || !fpinfo_o->pushdown_safe ||
5733 706 : !fpinfo_i || !fpinfo_i->pushdown_safe)
5734 10 : return false;
5735 :
5736 : /*
5737 : * If joining relations have local conditions, those conditions are
5738 : * required to be applied before joining the relations. Hence the join can
5739 : * not be pushed down.
5740 : */
5741 706 : if (fpinfo_o->local_conds || fpinfo_i->local_conds)
5742 18 : return false;
5743 :
5744 : /*
5745 : * Merge FDW options. We might be tempted to do this after we have deemed
5746 : * the foreign join to be OK. But we must do this beforehand so that we
5747 : * know which quals can be evaluated on the foreign server, which might
5748 : * depend on shippable_extensions.
5749 : */
5750 688 : fpinfo->server = fpinfo_o->server;
5751 688 : merge_fdw_options(fpinfo, fpinfo_o, fpinfo_i);
5752 :
5753 : /*
5754 : * Separate restrict list into join quals and pushed-down (other) quals.
5755 : *
5756 : * Join quals belonging to an outer join must all be shippable, else we
5757 : * cannot execute the join remotely. Add such quals to 'joinclauses'.
5758 : *
5759 : * Add other quals to fpinfo->remote_conds if they are shippable, else to
5760 : * fpinfo->local_conds. In an inner join it's okay to execute conditions
5761 : * either locally or remotely; the same is true for pushed-down conditions
5762 : * at an outer join.
5763 : *
5764 : * Note we might return failure after having already scribbled on
5765 : * fpinfo->remote_conds and fpinfo->local_conds. That's okay because we
5766 : * won't consult those lists again if we deem the join unshippable.
5767 : */
5768 688 : joinclauses = NIL;
5769 1370 : foreach(lc, extra->restrictlist)
5770 : {
5771 688 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc);
5772 688 : bool is_remote_clause = is_foreign_expr(root, joinrel,
5773 : rinfo->clause);
5774 :
5775 688 : if (IS_OUTER_JOIN(jointype) &&
5776 258 : !RINFO_IS_PUSHED_DOWN(rinfo, joinrel->relids))
5777 : {
5778 226 : if (!is_remote_clause)
5779 6 : return false;
5780 220 : joinclauses = lappend(joinclauses, rinfo);
5781 : }
5782 : else
5783 : {
5784 462 : if (is_remote_clause)
5785 438 : fpinfo->remote_conds = lappend(fpinfo->remote_conds, rinfo);
5786 : else
5787 24 : fpinfo->local_conds = lappend(fpinfo->local_conds, rinfo);
5788 : }
5789 : }
5790 :
5791 : /*
5792 : * deparseExplicitTargetList() isn't smart enough to handle anything other
5793 : * than a Var. In particular, if there's some PlaceHolderVar that would
5794 : * need to be evaluated within this join tree (because there's an upper
5795 : * reference to a quantity that may go to NULL as a result of an outer
5796 : * join), then we can't try to push the join down because we'll fail when
5797 : * we get to deparseExplicitTargetList(). However, a PlaceHolderVar that
5798 : * needs to be evaluated *at the top* of this join tree is OK, because we
5799 : * can do that locally after fetching the results from the remote side.
5800 : */
5801 688 : foreach(lc, root->placeholder_list)
5802 : {
5803 22 : PlaceHolderInfo *phinfo = lfirst(lc);
5804 : Relids relids;
5805 :
5806 : /* PlaceHolderInfo refers to parent relids, not child relids. */
5807 22 : relids = IS_OTHER_REL(joinrel) ?
5808 44 : joinrel->top_parent_relids : joinrel->relids;
5809 :
5810 44 : if (bms_is_subset(phinfo->ph_eval_at, relids) &&
5811 22 : bms_nonempty_difference(relids, phinfo->ph_eval_at))
5812 16 : return false;
5813 : }
5814 :
5815 : /* Save the join clauses, for later use. */
5816 666 : fpinfo->joinclauses = joinclauses;
5817 :
5818 666 : fpinfo->outerrel = outerrel;
5819 666 : fpinfo->innerrel = innerrel;
5820 666 : fpinfo->jointype = jointype;
5821 :
5822 : /*
5823 : * By default, both the input relations are not required to be deparsed as
5824 : * subqueries, but there might be some relations covered by the input
5825 : * relations that are required to be deparsed as subqueries, so save the
5826 : * relids of those relations for later use by the deparser.
5827 : */
5828 666 : fpinfo->make_outerrel_subquery = false;
5829 666 : fpinfo->make_innerrel_subquery = false;
5830 : Assert(bms_is_subset(fpinfo_o->lower_subquery_rels, outerrel->relids));
5831 : Assert(bms_is_subset(fpinfo_i->lower_subquery_rels, innerrel->relids));
5832 1332 : fpinfo->lower_subquery_rels = bms_union(fpinfo_o->lower_subquery_rels,
5833 666 : fpinfo_i->lower_subquery_rels);
5834 1332 : fpinfo->hidden_subquery_rels = bms_union(fpinfo_o->hidden_subquery_rels,
5835 666 : fpinfo_i->hidden_subquery_rels);
5836 :
5837 : /*
5838 : * Pull the other remote conditions from the joining relations into join
5839 : * clauses or other remote clauses (remote_conds) of this relation
5840 : * wherever possible. This avoids building subqueries at every join step.
5841 : *
5842 : * For an inner join, clauses from both the relations are added to the
5843 : * other remote clauses. For LEFT and RIGHT OUTER join, the clauses from
5844 : * the outer side are added to remote_conds since those can be evaluated
5845 : * after the join is evaluated. The clauses from inner side are added to
5846 : * the joinclauses, since they need to be evaluated while constructing the
5847 : * join.
5848 : *
5849 : * For SEMI-JOIN clauses from inner relation can not be added to
5850 : * remote_conds, but should be treated as join clauses (as they are
5851 : * deparsed to EXISTS subquery, where inner relation can be referred). A
5852 : * list of relation ids, which can't be referred to from higher levels, is
5853 : * preserved as a hidden_subquery_rels list.
5854 : *
5855 : * For a FULL OUTER JOIN, the other clauses from either relation can not
5856 : * be added to the joinclauses or remote_conds, since each relation acts
5857 : * as an outer relation for the other.
5858 : *
5859 : * The joining sides can not have local conditions, thus no need to test
5860 : * shippability of the clauses being pulled up.
5861 : */
5862 666 : switch (jointype)
5863 : {
5864 374 : case JOIN_INNER:
5865 748 : fpinfo->remote_conds = list_concat(fpinfo->remote_conds,
5866 374 : fpinfo_i->remote_conds);
5867 748 : fpinfo->remote_conds = list_concat(fpinfo->remote_conds,
5868 374 : fpinfo_o->remote_conds);
5869 374 : break;
5870 :
5871 120 : case JOIN_LEFT:
5872 :
5873 : /*
5874 : * When semi-join is involved in the inner or outer part of the
5875 : * left join, it's deparsed as a subquery, and we can't refer to
5876 : * its vars on the upper level.
5877 : */
5878 120 : if (bms_is_empty(fpinfo_i->hidden_subquery_rels))
5879 112 : fpinfo->joinclauses = list_concat(fpinfo->joinclauses,
5880 112 : fpinfo_i->remote_conds);
5881 120 : if (bms_is_empty(fpinfo_o->hidden_subquery_rels))
5882 120 : fpinfo->remote_conds = list_concat(fpinfo->remote_conds,
5883 120 : fpinfo_o->remote_conds);
5884 120 : break;
5885 :
5886 0 : case JOIN_RIGHT:
5887 :
5888 : /*
5889 : * When semi-join is involved in the inner or outer part of the
5890 : * right join, it's deparsed as a subquery, and we can't refer to
5891 : * its vars on the upper level.
5892 : */
5893 0 : if (bms_is_empty(fpinfo_o->hidden_subquery_rels))
5894 0 : fpinfo->joinclauses = list_concat(fpinfo->joinclauses,
5895 0 : fpinfo_o->remote_conds);
5896 0 : if (bms_is_empty(fpinfo_i->hidden_subquery_rels))
5897 0 : fpinfo->remote_conds = list_concat(fpinfo->remote_conds,
5898 0 : fpinfo_i->remote_conds);
5899 0 : break;
5900 :
5901 88 : case JOIN_SEMI:
5902 176 : fpinfo->joinclauses = list_concat(fpinfo->joinclauses,
5903 88 : fpinfo_i->remote_conds);
5904 176 : fpinfo->joinclauses = list_concat(fpinfo->joinclauses,
5905 88 : fpinfo->remote_conds);
5906 88 : fpinfo->remote_conds = list_copy(fpinfo_o->remote_conds);
5907 176 : fpinfo->hidden_subquery_rels = bms_union(fpinfo->hidden_subquery_rels,
5908 88 : innerrel->relids);
5909 88 : break;
5910 :
5911 84 : case JOIN_FULL:
5912 :
5913 : /*
5914 : * In this case, if any of the input relations has conditions, we
5915 : * need to deparse that relation as a subquery so that the
5916 : * conditions can be evaluated before the join. Remember it in
5917 : * the fpinfo of this relation so that the deparser can take
5918 : * appropriate action. Also, save the relids of base relations
5919 : * covered by that relation for later use by the deparser.
5920 : */
5921 84 : if (fpinfo_o->remote_conds)
5922 : {
5923 28 : fpinfo->make_outerrel_subquery = true;
5924 28 : fpinfo->lower_subquery_rels =
5925 28 : bms_add_members(fpinfo->lower_subquery_rels,
5926 28 : outerrel->relids);
5927 : }
5928 84 : if (fpinfo_i->remote_conds)
5929 : {
5930 28 : fpinfo->make_innerrel_subquery = true;
5931 28 : fpinfo->lower_subquery_rels =
5932 28 : bms_add_members(fpinfo->lower_subquery_rels,
5933 28 : innerrel->relids);
5934 : }
5935 84 : break;
5936 :
5937 0 : default:
5938 : /* Should not happen, we have just checked this above */
5939 0 : elog(ERROR, "unsupported join type %d", jointype);
5940 : }
5941 :
5942 : /*
5943 : * For an inner join, all restrictions can be treated alike. Treating the
5944 : * pushed down conditions as join conditions allows a top level full outer
5945 : * join to be deparsed without requiring subqueries.
5946 : */
5947 666 : if (jointype == JOIN_INNER)
5948 : {
5949 : Assert(!fpinfo->joinclauses);
5950 374 : fpinfo->joinclauses = fpinfo->remote_conds;
5951 374 : fpinfo->remote_conds = NIL;
5952 : }
5953 292 : else if (jointype == JOIN_LEFT || jointype == JOIN_RIGHT || jointype == JOIN_FULL)
5954 : {
5955 : /*
5956 : * Conditions, generated from semi-joins, should be evaluated before
5957 : * LEFT/RIGHT/FULL join.
5958 : */
5959 204 : if (!bms_is_empty(fpinfo_o->hidden_subquery_rels))
5960 : {
5961 0 : fpinfo->make_outerrel_subquery = true;
5962 0 : fpinfo->lower_subquery_rels = bms_add_members(fpinfo->lower_subquery_rels, outerrel->relids);
5963 : }
5964 :
5965 204 : if (!bms_is_empty(fpinfo_i->hidden_subquery_rels))
5966 : {
5967 8 : fpinfo->make_innerrel_subquery = true;
5968 8 : fpinfo->lower_subquery_rels = bms_add_members(fpinfo->lower_subquery_rels, innerrel->relids);
5969 : }
5970 : }
5971 :
5972 : /* Mark that this join can be pushed down safely */
5973 666 : fpinfo->pushdown_safe = true;
5974 :
5975 : /* Get user mapping */
5976 666 : if (fpinfo->use_remote_estimate)
5977 : {
5978 442 : if (fpinfo_o->use_remote_estimate)
5979 310 : fpinfo->user = fpinfo_o->user;
5980 : else
5981 132 : fpinfo->user = fpinfo_i->user;
5982 : }
5983 : else
5984 224 : fpinfo->user = NULL;
5985 :
5986 : /*
5987 : * Set # of retrieved rows and cached relation costs to some negative
5988 : * value, so that we can detect when they are set to some sensible values,
5989 : * during one (usually the first) of the calls to estimate_path_cost_size.
5990 : */
5991 666 : fpinfo->retrieved_rows = -1;
5992 666 : fpinfo->rel_startup_cost = -1;
5993 666 : fpinfo->rel_total_cost = -1;
5994 :
5995 : /*
5996 : * Set the string describing this join relation to be used in EXPLAIN
5997 : * output of corresponding ForeignScan. Note that the decoration we add
5998 : * to the base relation names mustn't include any digits, or it'll confuse
5999 : * postgresExplainForeignScan.
6000 : */
6001 666 : fpinfo->relation_name = psprintf("(%s) %s JOIN (%s)",
6002 : fpinfo_o->relation_name,
6003 : get_jointype_name(fpinfo->jointype),
6004 : fpinfo_i->relation_name);
6005 :
6006 : /*
6007 : * Set the relation index. This is defined as the position of this
6008 : * joinrel in the join_rel_list list plus the length of the rtable list.
6009 : * Note that since this joinrel is at the end of the join_rel_list list
6010 : * when we are called, we can get the position by list_length.
6011 : */
6012 : Assert(fpinfo->relation_index == 0); /* shouldn't be set yet */
6013 666 : fpinfo->relation_index =
6014 666 : list_length(root->parse->rtable) + list_length(root->join_rel_list);
6015 :
6016 666 : return true;
6017 : }
6018 :
6019 : static void
6020 3022 : add_paths_with_pathkeys_for_rel(PlannerInfo *root, RelOptInfo *rel,
6021 : Path *epq_path, List *restrictlist)
6022 : {
6023 3022 : List *useful_pathkeys_list = NIL; /* List of all pathkeys */
6024 : ListCell *lc;
6025 :
6026 3022 : useful_pathkeys_list = get_useful_pathkeys_for_relation(root, rel);
6027 :
6028 : /*
6029 : * Before creating sorted paths, arrange for the passed-in EPQ path, if
6030 : * any, to return columns needed by the parent ForeignScan node so that
6031 : * they will propagate up through Sort nodes injected below, if necessary.
6032 : */
6033 3022 : if (epq_path != NULL && useful_pathkeys_list != NIL)
6034 : {
6035 64 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) rel->fdw_private;
6036 64 : PathTarget *target = copy_pathtarget(epq_path->pathtarget);
6037 :
6038 : /* Include columns required for evaluating PHVs in the tlist. */
6039 64 : add_new_columns_to_pathtarget(target,
6040 64 : pull_var_clause((Node *) target->exprs,
6041 : PVC_RECURSE_PLACEHOLDERS));
6042 :
6043 : /* Include columns required for evaluating the local conditions. */
6044 70 : foreach(lc, fpinfo->local_conds)
6045 : {
6046 6 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc);
6047 :
6048 6 : add_new_columns_to_pathtarget(target,
6049 6 : pull_var_clause((Node *) rinfo->clause,
6050 : PVC_RECURSE_PLACEHOLDERS));
6051 : }
6052 :
6053 : /*
6054 : * If we have added any new columns, adjust the tlist of the EPQ path.
6055 : *
6056 : * Note: the plan created using this path will only be used to execute
6057 : * EPQ checks, where accuracy of the plan cost and width estimates
6058 : * would not be important, so we do not do set_pathtarget_cost_width()
6059 : * for the new pathtarget here. See also postgresGetForeignPlan().
6060 : */
6061 64 : if (list_length(target->exprs) > list_length(epq_path->pathtarget->exprs))
6062 : {
6063 : /* The EPQ path is a join path, so it is projection-capable. */
6064 : Assert(is_projection_capable_path(epq_path));
6065 :
6066 : /*
6067 : * Use create_projection_path() here, so as to avoid modifying it
6068 : * in place.
6069 : */
6070 8 : epq_path = (Path *) create_projection_path(root,
6071 : rel,
6072 : epq_path,
6073 : target);
6074 : }
6075 : }
6076 :
6077 : /* Create one path for each set of pathkeys we found above. */
6078 4398 : foreach(lc, useful_pathkeys_list)
6079 : {
6080 : double rows;
6081 : int width;
6082 : int disabled_nodes;
6083 : Cost startup_cost;
6084 : Cost total_cost;
6085 1376 : List *useful_pathkeys = lfirst(lc);
6086 : Path *sorted_epq_path;
6087 :
6088 1376 : estimate_path_cost_size(root, rel, NIL, useful_pathkeys, NULL,
6089 : &rows, &width, &disabled_nodes,
6090 : &startup_cost, &total_cost);
6091 :
6092 : /*
6093 : * The EPQ path must be at least as well sorted as the path itself, in
6094 : * case it gets used as input to a mergejoin.
6095 : */
6096 1376 : sorted_epq_path = epq_path;
6097 1376 : if (sorted_epq_path != NULL &&
6098 64 : !pathkeys_contained_in(useful_pathkeys,
6099 : sorted_epq_path->pathkeys))
6100 : sorted_epq_path = (Path *)
6101 52 : create_sort_path(root,
6102 : rel,
6103 : sorted_epq_path,
6104 : useful_pathkeys,
6105 : -1.0);
6106 :
6107 1376 : if (IS_SIMPLE_REL(rel))
6108 842 : add_path(rel, (Path *)
6109 842 : create_foreignscan_path(root, rel,
6110 : NULL,
6111 : rows,
6112 : disabled_nodes,
6113 : startup_cost,
6114 : total_cost,
6115 : useful_pathkeys,
6116 : rel->lateral_relids,
6117 : sorted_epq_path,
6118 : NIL, /* no fdw_restrictinfo
6119 : * list */
6120 : NIL));
6121 : else
6122 534 : add_path(rel, (Path *)
6123 534 : create_foreign_join_path(root, rel,
6124 : NULL,
6125 : rows,
6126 : disabled_nodes,
6127 : startup_cost,
6128 : total_cost,
6129 : useful_pathkeys,
6130 : rel->lateral_relids,
6131 : sorted_epq_path,
6132 : restrictlist,
6133 : NIL));
6134 : }
6135 3022 : }
6136 :
6137 : /*
6138 : * Parse options from foreign server and apply them to fpinfo.
6139 : *
6140 : * New options might also require tweaking merge_fdw_options().
6141 : */
6142 : static void
6143 2360 : apply_server_options(PgFdwRelationInfo *fpinfo)
6144 : {
6145 : ListCell *lc;
6146 :
6147 9958 : foreach(lc, fpinfo->server->options)
6148 : {
6149 7598 : DefElem *def = (DefElem *) lfirst(lc);
6150 :
6151 7598 : if (strcmp(def->defname, "use_remote_estimate") == 0)
6152 208 : fpinfo->use_remote_estimate = defGetBoolean(def);
6153 7390 : else if (strcmp(def->defname, "fdw_startup_cost") == 0)
6154 12 : (void) parse_real(defGetString(def), &fpinfo->fdw_startup_cost, 0,
6155 : NULL);
6156 7378 : else if (strcmp(def->defname, "fdw_tuple_cost") == 0)
6157 4 : (void) parse_real(defGetString(def), &fpinfo->fdw_tuple_cost, 0,
6158 : NULL);
6159 7374 : else if (strcmp(def->defname, "extensions") == 0)
6160 1856 : fpinfo->shippable_extensions =
6161 1856 : ExtractExtensionList(defGetString(def), false);
6162 5518 : else if (strcmp(def->defname, "fetch_size") == 0)
6163 0 : (void) parse_int(defGetString(def), &fpinfo->fetch_size, 0, NULL);
6164 5518 : else if (strcmp(def->defname, "async_capable") == 0)
6165 242 : fpinfo->async_capable = defGetBoolean(def);
6166 : }
6167 2360 : }
6168 :
6169 : /*
6170 : * Parse options from foreign table and apply them to fpinfo.
6171 : *
6172 : * New options might also require tweaking merge_fdw_options().
6173 : */
6174 : static void
6175 2360 : apply_table_options(PgFdwRelationInfo *fpinfo)
6176 : {
6177 : ListCell *lc;
6178 :
6179 6890 : foreach(lc, fpinfo->table->options)
6180 : {
6181 4530 : DefElem *def = (DefElem *) lfirst(lc);
6182 :
6183 4530 : if (strcmp(def->defname, "use_remote_estimate") == 0)
6184 696 : fpinfo->use_remote_estimate = defGetBoolean(def);
6185 3834 : else if (strcmp(def->defname, "fetch_size") == 0)
6186 0 : (void) parse_int(defGetString(def), &fpinfo->fetch_size, 0, NULL);
6187 3834 : else if (strcmp(def->defname, "async_capable") == 0)
6188 0 : fpinfo->async_capable = defGetBoolean(def);
6189 : }
6190 2360 : }
6191 :
6192 : /*
6193 : * Merge FDW options from input relations into a new set of options for a join
6194 : * or an upper rel.
6195 : *
6196 : * For a join relation, FDW-specific information about the inner and outer
6197 : * relations is provided using fpinfo_i and fpinfo_o. For an upper relation,
6198 : * fpinfo_o provides the information for the input relation; fpinfo_i is
6199 : * expected to NULL.
6200 : */
6201 : static void
6202 1580 : merge_fdw_options(PgFdwRelationInfo *fpinfo,
6203 : const PgFdwRelationInfo *fpinfo_o,
6204 : const PgFdwRelationInfo *fpinfo_i)
6205 : {
6206 : /* We must always have fpinfo_o. */
6207 : Assert(fpinfo_o);
6208 :
6209 : /* fpinfo_i may be NULL, but if present the servers must both match. */
6210 : Assert(!fpinfo_i ||
6211 : fpinfo_i->server->serverid == fpinfo_o->server->serverid);
6212 :
6213 : /*
6214 : * Copy the server specific FDW options. (For a join, both relations come
6215 : * from the same server, so the server options should have the same value
6216 : * for both relations.)
6217 : */
6218 1580 : fpinfo->fdw_startup_cost = fpinfo_o->fdw_startup_cost;
6219 1580 : fpinfo->fdw_tuple_cost = fpinfo_o->fdw_tuple_cost;
6220 1580 : fpinfo->shippable_extensions = fpinfo_o->shippable_extensions;
6221 1580 : fpinfo->use_remote_estimate = fpinfo_o->use_remote_estimate;
6222 1580 : fpinfo->fetch_size = fpinfo_o->fetch_size;
6223 1580 : fpinfo->async_capable = fpinfo_o->async_capable;
6224 :
6225 : /* Merge the table level options from either side of the join. */
6226 1580 : if (fpinfo_i)
6227 : {
6228 : /*
6229 : * We'll prefer to use remote estimates for this join if any table
6230 : * from either side of the join is using remote estimates. This is
6231 : * most likely going to be preferred since they're already willing to
6232 : * pay the price of a round trip to get the remote EXPLAIN. In any
6233 : * case it's not entirely clear how we might otherwise handle this
6234 : * best.
6235 : */
6236 1056 : fpinfo->use_remote_estimate = fpinfo_o->use_remote_estimate ||
6237 368 : fpinfo_i->use_remote_estimate;
6238 :
6239 : /*
6240 : * Set fetch size to maximum of the joining sides, since we are
6241 : * expecting the rows returned by the join to be proportional to the
6242 : * relation sizes.
6243 : */
6244 688 : fpinfo->fetch_size = Max(fpinfo_o->fetch_size, fpinfo_i->fetch_size);
6245 :
6246 : /*
6247 : * We'll prefer to consider this join async-capable if any table from
6248 : * either side of the join is considered async-capable. This would be
6249 : * reasonable because in that case the foreign server would have its
6250 : * own resources to scan that table asynchronously, and the join could
6251 : * also be computed asynchronously using the resources.
6252 : */
6253 1360 : fpinfo->async_capable = fpinfo_o->async_capable ||
6254 672 : fpinfo_i->async_capable;
6255 : }
6256 1580 : }
6257 :
6258 : /*
6259 : * postgresGetForeignJoinPaths
6260 : * Add possible ForeignPath to joinrel, if join is safe to push down.
6261 : */
6262 : static void
6263 2704 : postgresGetForeignJoinPaths(PlannerInfo *root,
6264 : RelOptInfo *joinrel,
6265 : RelOptInfo *outerrel,
6266 : RelOptInfo *innerrel,
6267 : JoinType jointype,
6268 : JoinPathExtraData *extra)
6269 : {
6270 : PgFdwRelationInfo *fpinfo;
6271 : ForeignPath *joinpath;
6272 : double rows;
6273 : int width;
6274 : int disabled_nodes;
6275 : Cost startup_cost;
6276 : Cost total_cost;
6277 : Path *epq_path; /* Path to create plan to be executed when
6278 : * EvalPlanQual gets triggered. */
6279 :
6280 : /*
6281 : * Skip if this join combination has been considered already.
6282 : */
6283 2704 : if (joinrel->fdw_private)
6284 2038 : return;
6285 :
6286 : /*
6287 : * This code does not work for joins with lateral references, since those
6288 : * must have parameterized paths, which we don't generate yet.
6289 : */
6290 792 : if (!bms_is_empty(joinrel->lateral_relids))
6291 8 : return;
6292 :
6293 : /*
6294 : * Create unfinished PgFdwRelationInfo entry which is used to indicate
6295 : * that the join relation is already considered, so that we won't waste
6296 : * time in judging safety of join pushdown and adding the same paths again
6297 : * if found safe. Once we know that this join can be pushed down, we fill
6298 : * the entry.
6299 : */
6300 784 : fpinfo = (PgFdwRelationInfo *) palloc0(sizeof(PgFdwRelationInfo));
6301 784 : fpinfo->pushdown_safe = false;
6302 784 : joinrel->fdw_private = fpinfo;
6303 : /* attrs_used is only for base relations. */
6304 784 : fpinfo->attrs_used = NULL;
6305 :
6306 : /*
6307 : * If there is a possibility that EvalPlanQual will be executed, we need
6308 : * to be able to reconstruct the row using scans of the base relations.
6309 : * GetExistingLocalJoinPath will find a suitable path for this purpose in
6310 : * the path list of the joinrel, if one exists. We must be careful to
6311 : * call it before adding any ForeignPath, since the ForeignPath might
6312 : * dominate the only suitable local path available. We also do it before
6313 : * calling foreign_join_ok(), since that function updates fpinfo and marks
6314 : * it as pushable if the join is found to be pushable.
6315 : */
6316 784 : if (root->parse->commandType == CMD_DELETE ||
6317 756 : root->parse->commandType == CMD_UPDATE ||
6318 704 : root->rowMarks)
6319 : {
6320 152 : epq_path = GetExistingLocalJoinPath(joinrel);
6321 152 : if (!epq_path)
6322 : {
6323 0 : elog(DEBUG3, "could not push down foreign join because a local path suitable for EPQ checks was not found");
6324 0 : return;
6325 : }
6326 : }
6327 : else
6328 632 : epq_path = NULL;
6329 :
6330 784 : if (!foreign_join_ok(root, joinrel, jointype, outerrel, innerrel, extra))
6331 : {
6332 : /* Free path required for EPQ if we copied one; we don't need it now */
6333 118 : if (epq_path)
6334 4 : pfree(epq_path);
6335 118 : return;
6336 : }
6337 :
6338 : /*
6339 : * Compute the selectivity and cost of the local_conds, so we don't have
6340 : * to do it over again for each path. The best we can do for these
6341 : * conditions is to estimate selectivity on the basis of local statistics.
6342 : * The local conditions are applied after the join has been computed on
6343 : * the remote side like quals in WHERE clause, so pass jointype as
6344 : * JOIN_INNER.
6345 : */
6346 666 : fpinfo->local_conds_sel = clauselist_selectivity(root,
6347 : fpinfo->local_conds,
6348 : 0,
6349 : JOIN_INNER,
6350 : NULL);
6351 666 : cost_qual_eval(&fpinfo->local_conds_cost, fpinfo->local_conds, root);
6352 :
6353 : /*
6354 : * If we are going to estimate costs locally, estimate the join clause
6355 : * selectivity here while we have special join info.
6356 : */
6357 666 : if (!fpinfo->use_remote_estimate)
6358 224 : fpinfo->joinclause_sel = clauselist_selectivity(root, fpinfo->joinclauses,
6359 : 0, fpinfo->jointype,
6360 : extra->sjinfo);
6361 :
6362 : /* Estimate costs for bare join relation */
6363 666 : estimate_path_cost_size(root, joinrel, NIL, NIL, NULL,
6364 : &rows, &width, &disabled_nodes,
6365 : &startup_cost, &total_cost);
6366 : /* Now update this information in the joinrel */
6367 666 : joinrel->rows = rows;
6368 666 : joinrel->reltarget->width = width;
6369 666 : fpinfo->rows = rows;
6370 666 : fpinfo->width = width;
6371 666 : fpinfo->disabled_nodes = disabled_nodes;
6372 666 : fpinfo->startup_cost = startup_cost;
6373 666 : fpinfo->total_cost = total_cost;
6374 :
6375 : /*
6376 : * Create a new join path and add it to the joinrel which represents a
6377 : * join between foreign tables.
6378 : */
6379 666 : joinpath = create_foreign_join_path(root,
6380 : joinrel,
6381 : NULL, /* default pathtarget */
6382 : rows,
6383 : disabled_nodes,
6384 : startup_cost,
6385 : total_cost,
6386 : NIL, /* no pathkeys */
6387 : joinrel->lateral_relids,
6388 : epq_path,
6389 : extra->restrictlist,
6390 : NIL); /* no fdw_private */
6391 :
6392 : /* Add generated path into joinrel by add_path(). */
6393 666 : add_path(joinrel, (Path *) joinpath);
6394 :
6395 : /* Consider pathkeys for the join relation */
6396 666 : add_paths_with_pathkeys_for_rel(root, joinrel, epq_path,
6397 : extra->restrictlist);
6398 :
6399 : /* XXX Consider parameterized paths for the join relation */
6400 : }
6401 :
6402 : /*
6403 : * Assess whether the aggregation, grouping and having operations can be pushed
6404 : * down to the foreign server. As a side effect, save information we obtain in
6405 : * this function to PgFdwRelationInfo of the input relation.
6406 : */
6407 : static bool
6408 322 : foreign_grouping_ok(PlannerInfo *root, RelOptInfo *grouped_rel,
6409 : Node *havingQual)
6410 : {
6411 322 : Query *query = root->parse;
6412 322 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) grouped_rel->fdw_private;
6413 322 : PathTarget *grouping_target = grouped_rel->reltarget;
6414 : PgFdwRelationInfo *ofpinfo;
6415 : ListCell *lc;
6416 : int i;
6417 322 : List *tlist = NIL;
6418 :
6419 : /* We currently don't support pushing Grouping Sets. */
6420 322 : if (query->groupingSets)
6421 12 : return false;
6422 :
6423 : /* Get the fpinfo of the underlying scan relation. */
6424 310 : ofpinfo = (PgFdwRelationInfo *) fpinfo->outerrel->fdw_private;
6425 :
6426 : /*
6427 : * If underlying scan relation has any local conditions, those conditions
6428 : * are required to be applied before performing aggregation. Hence the
6429 : * aggregate cannot be pushed down.
6430 : */
6431 310 : if (ofpinfo->local_conds)
6432 18 : return false;
6433 :
6434 : /*
6435 : * Examine grouping expressions, as well as other expressions we'd need to
6436 : * compute, and check whether they are safe to push down to the foreign
6437 : * server. All GROUP BY expressions will be part of the grouping target
6438 : * and thus there is no need to search for them separately. Add grouping
6439 : * expressions into target list which will be passed to foreign server.
6440 : *
6441 : * A tricky fine point is that we must not put any expression into the
6442 : * target list that is just a foreign param (that is, something that
6443 : * deparse.c would conclude has to be sent to the foreign server). If we
6444 : * do, the expression will also appear in the fdw_exprs list of the plan
6445 : * node, and setrefs.c will get confused and decide that the fdw_exprs
6446 : * entry is actually a reference to the fdw_scan_tlist entry, resulting in
6447 : * a broken plan. Somewhat oddly, it's OK if the expression contains such
6448 : * a node, as long as it's not at top level; then no match is possible.
6449 : */
6450 292 : i = 0;
6451 854 : foreach(lc, grouping_target->exprs)
6452 : {
6453 598 : Expr *expr = (Expr *) lfirst(lc);
6454 598 : Index sgref = get_pathtarget_sortgroupref(grouping_target, i);
6455 : ListCell *l;
6456 :
6457 : /*
6458 : * Check whether this expression is part of GROUP BY clause. Note we
6459 : * check the whole GROUP BY clause not just processed_groupClause,
6460 : * because we will ship all of it, cf. appendGroupByClause.
6461 : */
6462 598 : if (sgref && get_sortgroupref_clause_noerr(sgref, query->groupClause))
6463 184 : {
6464 : TargetEntry *tle;
6465 :
6466 : /*
6467 : * If any GROUP BY expression is not shippable, then we cannot
6468 : * push down aggregation to the foreign server.
6469 : */
6470 190 : if (!is_foreign_expr(root, grouped_rel, expr))
6471 36 : return false;
6472 :
6473 : /*
6474 : * If it would be a foreign param, we can't put it into the tlist,
6475 : * so we have to fail.
6476 : */
6477 188 : if (is_foreign_param(root, grouped_rel, expr))
6478 4 : return false;
6479 :
6480 : /*
6481 : * Pushable, so add to tlist. We need to create a TLE for this
6482 : * expression and apply the sortgroupref to it. We cannot use
6483 : * add_to_flat_tlist() here because that avoids making duplicate
6484 : * entries in the tlist. If there are duplicate entries with
6485 : * distinct sortgrouprefs, we have to duplicate that situation in
6486 : * the output tlist.
6487 : */
6488 184 : tle = makeTargetEntry(expr, list_length(tlist) + 1, NULL, false);
6489 184 : tle->ressortgroupref = sgref;
6490 184 : tlist = lappend(tlist, tle);
6491 : }
6492 : else
6493 : {
6494 : /*
6495 : * Non-grouping expression we need to compute. Can we ship it
6496 : * as-is to the foreign server?
6497 : */
6498 408 : if (is_foreign_expr(root, grouped_rel, expr) &&
6499 366 : !is_foreign_param(root, grouped_rel, expr))
6500 362 : {
6501 : /* Yes, so add to tlist as-is; OK to suppress duplicates */
6502 362 : tlist = add_to_flat_tlist(tlist, list_make1(expr));
6503 : }
6504 : else
6505 : {
6506 : /* Not pushable as a whole; extract its Vars and aggregates */
6507 : List *aggvars;
6508 :
6509 46 : aggvars = pull_var_clause((Node *) expr,
6510 : PVC_INCLUDE_AGGREGATES);
6511 :
6512 : /*
6513 : * If any aggregate expression is not shippable, then we
6514 : * cannot push down aggregation to the foreign server. (We
6515 : * don't have to check is_foreign_param, since that certainly
6516 : * won't return true for any such expression.)
6517 : */
6518 46 : if (!is_foreign_expr(root, grouped_rel, (Expr *) aggvars))
6519 30 : return false;
6520 :
6521 : /*
6522 : * Add aggregates, if any, into the targetlist. Plain Vars
6523 : * outside an aggregate can be ignored, because they should be
6524 : * either same as some GROUP BY column or part of some GROUP
6525 : * BY expression. In either case, they are already part of
6526 : * the targetlist and thus no need to add them again. In fact
6527 : * including plain Vars in the tlist when they do not match a
6528 : * GROUP BY column would cause the foreign server to complain
6529 : * that the shipped query is invalid.
6530 : */
6531 28 : foreach(l, aggvars)
6532 : {
6533 12 : Expr *aggref = (Expr *) lfirst(l);
6534 :
6535 12 : if (IsA(aggref, Aggref))
6536 8 : tlist = add_to_flat_tlist(tlist, list_make1(aggref));
6537 : }
6538 : }
6539 : }
6540 :
6541 562 : i++;
6542 : }
6543 :
6544 : /*
6545 : * Classify the pushable and non-pushable HAVING clauses and save them in
6546 : * remote_conds and local_conds of the grouped rel's fpinfo.
6547 : */
6548 256 : if (havingQual)
6549 : {
6550 68 : foreach(lc, (List *) havingQual)
6551 : {
6552 38 : Expr *expr = (Expr *) lfirst(lc);
6553 : RestrictInfo *rinfo;
6554 :
6555 : /*
6556 : * Currently, the core code doesn't wrap havingQuals in
6557 : * RestrictInfos, so we must make our own.
6558 : */
6559 : Assert(!IsA(expr, RestrictInfo));
6560 38 : rinfo = make_restrictinfo(root,
6561 : expr,
6562 : true,
6563 : false,
6564 : false,
6565 : false,
6566 : root->qual_security_level,
6567 : grouped_rel->relids,
6568 : NULL,
6569 : NULL);
6570 38 : if (is_foreign_expr(root, grouped_rel, expr))
6571 32 : fpinfo->remote_conds = lappend(fpinfo->remote_conds, rinfo);
6572 : else
6573 6 : fpinfo->local_conds = lappend(fpinfo->local_conds, rinfo);
6574 : }
6575 : }
6576 :
6577 : /*
6578 : * If there are any local conditions, pull Vars and aggregates from it and
6579 : * check whether they are safe to pushdown or not.
6580 : */
6581 256 : if (fpinfo->local_conds)
6582 : {
6583 6 : List *aggvars = NIL;
6584 :
6585 12 : foreach(lc, fpinfo->local_conds)
6586 : {
6587 6 : RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc);
6588 :
6589 6 : aggvars = list_concat(aggvars,
6590 6 : pull_var_clause((Node *) rinfo->clause,
6591 : PVC_INCLUDE_AGGREGATES));
6592 : }
6593 :
6594 14 : foreach(lc, aggvars)
6595 : {
6596 10 : Expr *expr = (Expr *) lfirst(lc);
6597 :
6598 : /*
6599 : * If aggregates within local conditions are not safe to push
6600 : * down, then we cannot push down the query. Vars are already
6601 : * part of GROUP BY clause which are checked above, so no need to
6602 : * access them again here. Again, we need not check
6603 : * is_foreign_param for a foreign aggregate.
6604 : */
6605 10 : if (IsA(expr, Aggref))
6606 : {
6607 10 : if (!is_foreign_expr(root, grouped_rel, expr))
6608 2 : return false;
6609 :
6610 8 : tlist = add_to_flat_tlist(tlist, list_make1(expr));
6611 : }
6612 : }
6613 : }
6614 :
6615 : /* Store generated targetlist */
6616 254 : fpinfo->grouped_tlist = tlist;
6617 :
6618 : /* Safe to pushdown */
6619 254 : fpinfo->pushdown_safe = true;
6620 :
6621 : /*
6622 : * Set # of retrieved rows and cached relation costs to some negative
6623 : * value, so that we can detect when they are set to some sensible values,
6624 : * during one (usually the first) of the calls to estimate_path_cost_size.
6625 : */
6626 254 : fpinfo->retrieved_rows = -1;
6627 254 : fpinfo->rel_startup_cost = -1;
6628 254 : fpinfo->rel_total_cost = -1;
6629 :
6630 : /*
6631 : * Set the string describing this grouped relation to be used in EXPLAIN
6632 : * output of corresponding ForeignScan. Note that the decoration we add
6633 : * to the base relation name mustn't include any digits, or it'll confuse
6634 : * postgresExplainForeignScan.
6635 : */
6636 254 : fpinfo->relation_name = psprintf("Aggregate on (%s)",
6637 : ofpinfo->relation_name);
6638 :
6639 254 : return true;
6640 : }
6641 :
6642 : /*
6643 : * postgresGetForeignUpperPaths
6644 : * Add paths for post-join operations like aggregation, grouping etc. if
6645 : * corresponding operations are safe to push down.
6646 : */
6647 : static void
6648 1928 : postgresGetForeignUpperPaths(PlannerInfo *root, UpperRelationKind stage,
6649 : RelOptInfo *input_rel, RelOptInfo *output_rel,
6650 : void *extra)
6651 : {
6652 : PgFdwRelationInfo *fpinfo;
6653 :
6654 : /*
6655 : * If input rel is not safe to pushdown, then simply return as we cannot
6656 : * perform any post-join operations on the foreign server.
6657 : */
6658 1928 : if (!input_rel->fdw_private ||
6659 1796 : !((PgFdwRelationInfo *) input_rel->fdw_private)->pushdown_safe)
6660 244 : return;
6661 :
6662 : /* Ignore stages we don't support; and skip any duplicate calls. */
6663 1684 : if ((stage != UPPERREL_GROUP_AGG &&
6664 1062 : stage != UPPERREL_ORDERED &&
6665 1650 : stage != UPPERREL_FINAL) ||
6666 1650 : output_rel->fdw_private)
6667 34 : return;
6668 :
6669 1650 : fpinfo = (PgFdwRelationInfo *) palloc0(sizeof(PgFdwRelationInfo));
6670 1650 : fpinfo->pushdown_safe = false;
6671 1650 : fpinfo->stage = stage;
6672 1650 : output_rel->fdw_private = fpinfo;
6673 :
6674 1650 : switch (stage)
6675 : {
6676 322 : case UPPERREL_GROUP_AGG:
6677 322 : add_foreign_grouping_paths(root, input_rel, output_rel,
6678 : (GroupPathExtraData *) extra);
6679 322 : break;
6680 300 : case UPPERREL_ORDERED:
6681 300 : add_foreign_ordered_paths(root, input_rel, output_rel);
6682 300 : break;
6683 1028 : case UPPERREL_FINAL:
6684 1028 : add_foreign_final_paths(root, input_rel, output_rel,
6685 : (FinalPathExtraData *) extra);
6686 1028 : break;
6687 0 : default:
6688 0 : elog(ERROR, "unexpected upper relation: %d", (int) stage);
6689 : break;
6690 : }
6691 : }
6692 :
6693 : /*
6694 : * add_foreign_grouping_paths
6695 : * Add foreign path for grouping and/or aggregation.
6696 : *
6697 : * Given input_rel represents the underlying scan. The paths are added to the
6698 : * given grouped_rel.
6699 : */
6700 : static void
6701 322 : add_foreign_grouping_paths(PlannerInfo *root, RelOptInfo *input_rel,
6702 : RelOptInfo *grouped_rel,
6703 : GroupPathExtraData *extra)
6704 : {
6705 322 : Query *parse = root->parse;
6706 322 : PgFdwRelationInfo *ifpinfo = input_rel->fdw_private;
6707 322 : PgFdwRelationInfo *fpinfo = grouped_rel->fdw_private;
6708 : ForeignPath *grouppath;
6709 : double rows;
6710 : int width;
6711 : int disabled_nodes;
6712 : Cost startup_cost;
6713 : Cost total_cost;
6714 :
6715 : /* Nothing to be done, if there is no grouping or aggregation required. */
6716 322 : if (!parse->groupClause && !parse->groupingSets && !parse->hasAggs &&
6717 0 : !root->hasHavingQual)
6718 68 : return;
6719 :
6720 : Assert(extra->patype == PARTITIONWISE_AGGREGATE_NONE ||
6721 : extra->patype == PARTITIONWISE_AGGREGATE_FULL);
6722 :
6723 : /* save the input_rel as outerrel in fpinfo */
6724 322 : fpinfo->outerrel = input_rel;
6725 :
6726 : /*
6727 : * Copy foreign table, foreign server, user mapping, FDW options etc.
6728 : * details from the input relation's fpinfo.
6729 : */
6730 322 : fpinfo->table = ifpinfo->table;
6731 322 : fpinfo->server = ifpinfo->server;
6732 322 : fpinfo->user = ifpinfo->user;
6733 322 : merge_fdw_options(fpinfo, ifpinfo, NULL);
6734 :
6735 : /*
6736 : * Assess if it is safe to push down aggregation and grouping.
6737 : *
6738 : * Use HAVING qual from extra. In case of child partition, it will have
6739 : * translated Vars.
6740 : */
6741 322 : if (!foreign_grouping_ok(root, grouped_rel, extra->havingQual))
6742 68 : return;
6743 :
6744 : /*
6745 : * Compute the selectivity and cost of the local_conds, so we don't have
6746 : * to do it over again for each path. (Currently we create just a single
6747 : * path here, but in future it would be possible that we build more paths
6748 : * such as pre-sorted paths as in postgresGetForeignPaths and
6749 : * postgresGetForeignJoinPaths.) The best we can do for these conditions
6750 : * is to estimate selectivity on the basis of local statistics.
6751 : */
6752 254 : fpinfo->local_conds_sel = clauselist_selectivity(root,
6753 : fpinfo->local_conds,
6754 : 0,
6755 : JOIN_INNER,
6756 : NULL);
6757 :
6758 254 : cost_qual_eval(&fpinfo->local_conds_cost, fpinfo->local_conds, root);
6759 :
6760 : /* Estimate the cost of push down */
6761 254 : estimate_path_cost_size(root, grouped_rel, NIL, NIL, NULL,
6762 : &rows, &width, &disabled_nodes,
6763 : &startup_cost, &total_cost);
6764 :
6765 : /* Now update this information in the fpinfo */
6766 254 : fpinfo->rows = rows;
6767 254 : fpinfo->width = width;
6768 254 : fpinfo->disabled_nodes = disabled_nodes;
6769 254 : fpinfo->startup_cost = startup_cost;
6770 254 : fpinfo->total_cost = total_cost;
6771 :
6772 : /* Create and add foreign path to the grouping relation. */
6773 254 : grouppath = create_foreign_upper_path(root,
6774 : grouped_rel,
6775 254 : grouped_rel->reltarget,
6776 : rows,
6777 : disabled_nodes,
6778 : startup_cost,
6779 : total_cost,
6780 : NIL, /* no pathkeys */
6781 : NULL,
6782 : NIL, /* no fdw_restrictinfo list */
6783 : NIL); /* no fdw_private */
6784 :
6785 : /* Add generated path into grouped_rel by add_path(). */
6786 254 : add_path(grouped_rel, (Path *) grouppath);
6787 : }
6788 :
6789 : /*
6790 : * add_foreign_ordered_paths
6791 : * Add foreign paths for performing the final sort remotely.
6792 : *
6793 : * Given input_rel contains the source-data Paths. The paths are added to the
6794 : * given ordered_rel.
6795 : */
6796 : static void
6797 300 : add_foreign_ordered_paths(PlannerInfo *root, RelOptInfo *input_rel,
6798 : RelOptInfo *ordered_rel)
6799 : {
6800 300 : Query *parse = root->parse;
6801 300 : PgFdwRelationInfo *ifpinfo = input_rel->fdw_private;
6802 300 : PgFdwRelationInfo *fpinfo = ordered_rel->fdw_private;
6803 : PgFdwPathExtraData *fpextra;
6804 : double rows;
6805 : int width;
6806 : int disabled_nodes;
6807 : Cost startup_cost;
6808 : Cost total_cost;
6809 : List *fdw_private;
6810 : ForeignPath *ordered_path;
6811 : ListCell *lc;
6812 :
6813 : /* Shouldn't get here unless the query has ORDER BY */
6814 : Assert(parse->sortClause);
6815 :
6816 : /* We don't support cases where there are any SRFs in the targetlist */
6817 300 : if (parse->hasTargetSRFs)
6818 216 : return;
6819 :
6820 : /* Save the input_rel as outerrel in fpinfo */
6821 300 : fpinfo->outerrel = input_rel;
6822 :
6823 : /*
6824 : * Copy foreign table, foreign server, user mapping, FDW options etc.
6825 : * details from the input relation's fpinfo.
6826 : */
6827 300 : fpinfo->table = ifpinfo->table;
6828 300 : fpinfo->server = ifpinfo->server;
6829 300 : fpinfo->user = ifpinfo->user;
6830 300 : merge_fdw_options(fpinfo, ifpinfo, NULL);
6831 :
6832 : /*
6833 : * If the input_rel is a base or join relation, we would already have
6834 : * considered pushing down the final sort to the remote server when
6835 : * creating pre-sorted foreign paths for that relation, because the
6836 : * query_pathkeys is set to the root->sort_pathkeys in that case (see
6837 : * standard_qp_callback()).
6838 : */
6839 300 : if (input_rel->reloptkind == RELOPT_BASEREL ||
6840 218 : input_rel->reloptkind == RELOPT_JOINREL)
6841 : {
6842 : Assert(root->query_pathkeys == root->sort_pathkeys);
6843 :
6844 : /* Safe to push down if the query_pathkeys is safe to push down */
6845 208 : fpinfo->pushdown_safe = ifpinfo->qp_is_pushdown_safe;
6846 :
6847 208 : return;
6848 : }
6849 :
6850 : /* The input_rel should be a grouping relation */
6851 : Assert(input_rel->reloptkind == RELOPT_UPPER_REL &&
6852 : ifpinfo->stage == UPPERREL_GROUP_AGG);
6853 :
6854 : /*
6855 : * We try to create a path below by extending a simple foreign path for
6856 : * the underlying grouping relation to perform the final sort remotely,
6857 : * which is stored into the fdw_private list of the resulting path.
6858 : */
6859 :
6860 : /* Assess if it is safe to push down the final sort */
6861 188 : foreach(lc, root->sort_pathkeys)
6862 : {
6863 104 : PathKey *pathkey = (PathKey *) lfirst(lc);
6864 104 : EquivalenceClass *pathkey_ec = pathkey->pk_eclass;
6865 :
6866 : /*
6867 : * is_foreign_expr would detect volatile expressions as well, but
6868 : * checking ec_has_volatile here saves some cycles.
6869 : */
6870 104 : if (pathkey_ec->ec_has_volatile)
6871 8 : return;
6872 :
6873 : /*
6874 : * Can't push down the sort if pathkey's opfamily is not shippable.
6875 : */
6876 96 : if (!is_shippable(pathkey->pk_opfamily, OperatorFamilyRelationId,
6877 : fpinfo))
6878 0 : return;
6879 :
6880 : /*
6881 : * The EC must contain a shippable EM that is computed in input_rel's
6882 : * reltarget, else we can't push down the sort.
6883 : */
6884 96 : if (find_em_for_rel_target(root,
6885 : pathkey_ec,
6886 : input_rel) == NULL)
6887 0 : return;
6888 : }
6889 :
6890 : /* Safe to push down */
6891 84 : fpinfo->pushdown_safe = true;
6892 :
6893 : /* Construct PgFdwPathExtraData */
6894 84 : fpextra = (PgFdwPathExtraData *) palloc0(sizeof(PgFdwPathExtraData));
6895 84 : fpextra->target = root->upper_targets[UPPERREL_ORDERED];
6896 84 : fpextra->has_final_sort = true;
6897 :
6898 : /* Estimate the costs of performing the final sort remotely */
6899 84 : estimate_path_cost_size(root, input_rel, NIL, root->sort_pathkeys, fpextra,
6900 : &rows, &width, &disabled_nodes,
6901 : &startup_cost, &total_cost);
6902 :
6903 : /*
6904 : * Build the fdw_private list that will be used by postgresGetForeignPlan.
6905 : * Items in the list must match order in enum FdwPathPrivateIndex.
6906 : */
6907 84 : fdw_private = list_make2(makeBoolean(true), makeBoolean(false));
6908 :
6909 : /* Create foreign ordering path */
6910 84 : ordered_path = create_foreign_upper_path(root,
6911 : input_rel,
6912 84 : root->upper_targets[UPPERREL_ORDERED],
6913 : rows,
6914 : disabled_nodes,
6915 : startup_cost,
6916 : total_cost,
6917 : root->sort_pathkeys,
6918 : NULL, /* no extra plan */
6919 : NIL, /* no fdw_restrictinfo
6920 : * list */
6921 : fdw_private);
6922 :
6923 : /* and add it to the ordered_rel */
6924 84 : add_path(ordered_rel, (Path *) ordered_path);
6925 : }
6926 :
6927 : /*
6928 : * add_foreign_final_paths
6929 : * Add foreign paths for performing the final processing remotely.
6930 : *
6931 : * Given input_rel contains the source-data Paths. The paths are added to the
6932 : * given final_rel.
6933 : */
6934 : static void
6935 1028 : add_foreign_final_paths(PlannerInfo *root, RelOptInfo *input_rel,
6936 : RelOptInfo *final_rel,
6937 : FinalPathExtraData *extra)
6938 : {
6939 1028 : Query *parse = root->parse;
6940 1028 : PgFdwRelationInfo *ifpinfo = (PgFdwRelationInfo *) input_rel->fdw_private;
6941 1028 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) final_rel->fdw_private;
6942 1028 : bool has_final_sort = false;
6943 1028 : List *pathkeys = NIL;
6944 : PgFdwPathExtraData *fpextra;
6945 1028 : bool save_use_remote_estimate = false;
6946 : double rows;
6947 : int width;
6948 : int disabled_nodes;
6949 : Cost startup_cost;
6950 : Cost total_cost;
6951 : List *fdw_private;
6952 : ForeignPath *final_path;
6953 :
6954 : /*
6955 : * Currently, we only support this for SELECT commands
6956 : */
6957 1028 : if (parse->commandType != CMD_SELECT)
6958 786 : return;
6959 :
6960 : /*
6961 : * No work if there is no FOR UPDATE/SHARE clause and if there is no need
6962 : * to add a LIMIT node
6963 : */
6964 802 : if (!parse->rowMarks && !extra->limit_needed)
6965 532 : return;
6966 :
6967 : /* We don't support cases where there are any SRFs in the targetlist */
6968 270 : if (parse->hasTargetSRFs)
6969 0 : return;
6970 :
6971 : /* Save the input_rel as outerrel in fpinfo */
6972 270 : fpinfo->outerrel = input_rel;
6973 :
6974 : /*
6975 : * Copy foreign table, foreign server, user mapping, FDW options etc.
6976 : * details from the input relation's fpinfo.
6977 : */
6978 270 : fpinfo->table = ifpinfo->table;
6979 270 : fpinfo->server = ifpinfo->server;
6980 270 : fpinfo->user = ifpinfo->user;
6981 270 : merge_fdw_options(fpinfo, ifpinfo, NULL);
6982 :
6983 : /*
6984 : * If there is no need to add a LIMIT node, there might be a ForeignPath
6985 : * in the input_rel's pathlist that implements all behavior of the query.
6986 : * Note: we would already have accounted for the query's FOR UPDATE/SHARE
6987 : * (if any) before we get here.
6988 : */
6989 270 : if (!extra->limit_needed)
6990 : {
6991 : ListCell *lc;
6992 :
6993 : Assert(parse->rowMarks);
6994 :
6995 : /*
6996 : * Grouping and aggregation are not supported with FOR UPDATE/SHARE,
6997 : * so the input_rel should be a base, join, or ordered relation; and
6998 : * if it's an ordered relation, its input relation should be a base or
6999 : * join relation.
7000 : */
7001 : Assert(input_rel->reloptkind == RELOPT_BASEREL ||
7002 : input_rel->reloptkind == RELOPT_JOINREL ||
7003 : (input_rel->reloptkind == RELOPT_UPPER_REL &&
7004 : ifpinfo->stage == UPPERREL_ORDERED &&
7005 : (ifpinfo->outerrel->reloptkind == RELOPT_BASEREL ||
7006 : ifpinfo->outerrel->reloptkind == RELOPT_JOINREL)));
7007 :
7008 8 : foreach(lc, input_rel->pathlist)
7009 : {
7010 8 : Path *path = (Path *) lfirst(lc);
7011 :
7012 : /*
7013 : * apply_scanjoin_target_to_paths() uses create_projection_path()
7014 : * to adjust each of its input paths if needed, whereas
7015 : * create_ordered_paths() uses apply_projection_to_path() to do
7016 : * that. So the former might have put a ProjectionPath on top of
7017 : * the ForeignPath; look through ProjectionPath and see if the
7018 : * path underneath it is ForeignPath.
7019 : */
7020 8 : if (IsA(path, ForeignPath) ||
7021 0 : (IsA(path, ProjectionPath) &&
7022 0 : IsA(((ProjectionPath *) path)->subpath, ForeignPath)))
7023 : {
7024 : /*
7025 : * Create foreign final path; this gets rid of a
7026 : * no-longer-needed outer plan (if any), which makes the
7027 : * EXPLAIN output look cleaner
7028 : */
7029 8 : final_path = create_foreign_upper_path(root,
7030 : path->parent,
7031 : path->pathtarget,
7032 : path->rows,
7033 : path->disabled_nodes,
7034 : path->startup_cost,
7035 : path->total_cost,
7036 : path->pathkeys,
7037 : NULL, /* no extra plan */
7038 : NIL, /* no fdw_restrictinfo
7039 : * list */
7040 : NIL); /* no fdw_private */
7041 :
7042 : /* and add it to the final_rel */
7043 8 : add_path(final_rel, (Path *) final_path);
7044 :
7045 : /* Safe to push down */
7046 8 : fpinfo->pushdown_safe = true;
7047 :
7048 8 : return;
7049 : }
7050 : }
7051 :
7052 : /*
7053 : * If we get here it means no ForeignPaths; since we would already
7054 : * have considered pushing down all operations for the query to the
7055 : * remote server, give up on it.
7056 : */
7057 0 : return;
7058 : }
7059 :
7060 : Assert(extra->limit_needed);
7061 :
7062 : /*
7063 : * If the input_rel is an ordered relation, replace the input_rel with its
7064 : * input relation
7065 : */
7066 262 : if (input_rel->reloptkind == RELOPT_UPPER_REL &&
7067 148 : ifpinfo->stage == UPPERREL_ORDERED)
7068 : {
7069 148 : input_rel = ifpinfo->outerrel;
7070 148 : ifpinfo = (PgFdwRelationInfo *) input_rel->fdw_private;
7071 148 : has_final_sort = true;
7072 148 : pathkeys = root->sort_pathkeys;
7073 : }
7074 :
7075 : /* The input_rel should be a base, join, or grouping relation */
7076 : Assert(input_rel->reloptkind == RELOPT_BASEREL ||
7077 : input_rel->reloptkind == RELOPT_JOINREL ||
7078 : (input_rel->reloptkind == RELOPT_UPPER_REL &&
7079 : ifpinfo->stage == UPPERREL_GROUP_AGG));
7080 :
7081 : /*
7082 : * We try to create a path below by extending a simple foreign path for
7083 : * the underlying base, join, or grouping relation to perform the final
7084 : * sort (if has_final_sort) and the LIMIT restriction remotely, which is
7085 : * stored into the fdw_private list of the resulting path. (We
7086 : * re-estimate the costs of sorting the underlying relation, if
7087 : * has_final_sort.)
7088 : */
7089 :
7090 : /*
7091 : * Assess if it is safe to push down the LIMIT and OFFSET to the remote
7092 : * server
7093 : */
7094 :
7095 : /*
7096 : * If the underlying relation has any local conditions, the LIMIT/OFFSET
7097 : * cannot be pushed down.
7098 : */
7099 262 : if (ifpinfo->local_conds)
7100 16 : return;
7101 :
7102 : /*
7103 : * If the query has FETCH FIRST .. WITH TIES, 1) it must have ORDER BY as
7104 : * well, which is used to determine which additional rows tie for the last
7105 : * place in the result set, and 2) ORDER BY must already have been
7106 : * determined to be safe to push down before we get here. So in that case
7107 : * the FETCH clause is safe to push down with ORDER BY if the remote
7108 : * server is v13 or later, but if not, the remote query will fail entirely
7109 : * for lack of support for it. Since we do not currently have a way to do
7110 : * a remote-version check (without accessing the remote server), disable
7111 : * pushing the FETCH clause for now.
7112 : */
7113 246 : if (parse->limitOption == LIMIT_OPTION_WITH_TIES)
7114 4 : return;
7115 :
7116 : /*
7117 : * Also, the LIMIT/OFFSET cannot be pushed down, if their expressions are
7118 : * not safe to remote.
7119 : */
7120 242 : if (!is_foreign_expr(root, input_rel, (Expr *) parse->limitOffset) ||
7121 242 : !is_foreign_expr(root, input_rel, (Expr *) parse->limitCount))
7122 0 : return;
7123 :
7124 : /* Safe to push down */
7125 242 : fpinfo->pushdown_safe = true;
7126 :
7127 : /* Construct PgFdwPathExtraData */
7128 242 : fpextra = (PgFdwPathExtraData *) palloc0(sizeof(PgFdwPathExtraData));
7129 242 : fpextra->target = root->upper_targets[UPPERREL_FINAL];
7130 242 : fpextra->has_final_sort = has_final_sort;
7131 242 : fpextra->has_limit = extra->limit_needed;
7132 242 : fpextra->limit_tuples = extra->limit_tuples;
7133 242 : fpextra->count_est = extra->count_est;
7134 242 : fpextra->offset_est = extra->offset_est;
7135 :
7136 : /*
7137 : * Estimate the costs of performing the final sort and the LIMIT
7138 : * restriction remotely. If has_final_sort is false, we wouldn't need to
7139 : * execute EXPLAIN anymore if use_remote_estimate, since the costs can be
7140 : * roughly estimated using the costs we already have for the underlying
7141 : * relation, in the same way as when use_remote_estimate is false. Since
7142 : * it's pretty expensive to execute EXPLAIN, force use_remote_estimate to
7143 : * false in that case.
7144 : */
7145 242 : if (!fpextra->has_final_sort)
7146 : {
7147 108 : save_use_remote_estimate = ifpinfo->use_remote_estimate;
7148 108 : ifpinfo->use_remote_estimate = false;
7149 : }
7150 242 : estimate_path_cost_size(root, input_rel, NIL, pathkeys, fpextra,
7151 : &rows, &width, &disabled_nodes,
7152 : &startup_cost, &total_cost);
7153 242 : if (!fpextra->has_final_sort)
7154 108 : ifpinfo->use_remote_estimate = save_use_remote_estimate;
7155 :
7156 : /*
7157 : * Build the fdw_private list that will be used by postgresGetForeignPlan.
7158 : * Items in the list must match order in enum FdwPathPrivateIndex.
7159 : */
7160 242 : fdw_private = list_make2(makeBoolean(has_final_sort),
7161 : makeBoolean(extra->limit_needed));
7162 :
7163 : /*
7164 : * Create foreign final path; this gets rid of a no-longer-needed outer
7165 : * plan (if any), which makes the EXPLAIN output look cleaner
7166 : */
7167 242 : final_path = create_foreign_upper_path(root,
7168 : input_rel,
7169 242 : root->upper_targets[UPPERREL_FINAL],
7170 : rows,
7171 : disabled_nodes,
7172 : startup_cost,
7173 : total_cost,
7174 : pathkeys,
7175 : NULL, /* no extra plan */
7176 : NIL, /* no fdw_restrictinfo list */
7177 : fdw_private);
7178 :
7179 : /* and add it to the final_rel */
7180 242 : add_path(final_rel, (Path *) final_path);
7181 : }
7182 :
7183 : /*
7184 : * postgresIsForeignPathAsyncCapable
7185 : * Check whether a given ForeignPath node is async-capable.
7186 : */
7187 : static bool
7188 474 : postgresIsForeignPathAsyncCapable(ForeignPath *path)
7189 : {
7190 474 : RelOptInfo *rel = ((Path *) path)->parent;
7191 474 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) rel->fdw_private;
7192 :
7193 474 : return fpinfo->async_capable;
7194 : }
7195 :
7196 : /*
7197 : * postgresForeignAsyncRequest
7198 : * Asynchronously request next tuple from a foreign PostgreSQL table.
7199 : */
7200 : static void
7201 12350 : postgresForeignAsyncRequest(AsyncRequest *areq)
7202 : {
7203 12350 : produce_tuple_asynchronously(areq, true);
7204 12350 : }
7205 :
7206 : /*
7207 : * postgresForeignAsyncConfigureWait
7208 : * Configure a file descriptor event for which we wish to wait.
7209 : */
7210 : static void
7211 356 : postgresForeignAsyncConfigureWait(AsyncRequest *areq)
7212 : {
7213 356 : ForeignScanState *node = (ForeignScanState *) areq->requestee;
7214 356 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
7215 356 : AsyncRequest *pendingAreq = fsstate->conn_state->pendingAreq;
7216 356 : AppendState *requestor = (AppendState *) areq->requestor;
7217 356 : WaitEventSet *set = requestor->as_eventset;
7218 :
7219 : /* This should not be called unless callback_pending */
7220 : Assert(areq->callback_pending);
7221 :
7222 : /*
7223 : * If process_pending_request() has been invoked on the given request
7224 : * before we get here, we might have some tuples already; in which case
7225 : * complete the request
7226 : */
7227 356 : if (fsstate->next_tuple < fsstate->num_tuples)
7228 : {
7229 10 : complete_pending_request(areq);
7230 10 : if (areq->request_complete)
7231 6 : return;
7232 : Assert(areq->callback_pending);
7233 : }
7234 :
7235 : /* We must have run out of tuples */
7236 : Assert(fsstate->next_tuple >= fsstate->num_tuples);
7237 :
7238 : /* The core code would have registered postmaster death event */
7239 : Assert(GetNumRegisteredWaitEvents(set) >= 1);
7240 :
7241 : /* Begin an asynchronous data fetch if not already done */
7242 350 : if (!pendingAreq)
7243 8 : fetch_more_data_begin(areq);
7244 342 : else if (pendingAreq->requestor != areq->requestor)
7245 : {
7246 : /*
7247 : * This is the case when the in-process request was made by another
7248 : * Append. Note that it might be useless to process the request made
7249 : * by that Append, because the query might not need tuples from that
7250 : * Append anymore; so we avoid processing it to begin a fetch for the
7251 : * given request if possible. If there are any child subplans of the
7252 : * same parent that are ready for new requests, skip the given
7253 : * request. Likewise, if there are any configured events other than
7254 : * the postmaster death event, skip it. Otherwise, process the
7255 : * in-process request, then begin a fetch to configure the event
7256 : * below, because we might otherwise end up with no configured events
7257 : * other than the postmaster death event.
7258 : */
7259 16 : if (!bms_is_empty(requestor->as_needrequest))
7260 0 : return;
7261 16 : if (GetNumRegisteredWaitEvents(set) > 1)
7262 12 : return;
7263 4 : process_pending_request(pendingAreq);
7264 4 : fetch_more_data_begin(areq);
7265 : }
7266 326 : else if (pendingAreq->requestee != areq->requestee)
7267 : {
7268 : /*
7269 : * This is the case when the in-process request was made by the same
7270 : * parent but for a different child. Since we configure only the
7271 : * event for the request made for that child, skip the given request.
7272 : */
7273 14 : return;
7274 : }
7275 : else
7276 : Assert(pendingAreq == areq);
7277 :
7278 322 : AddWaitEventToSet(set, WL_SOCKET_READABLE, PQsocket(fsstate->conn),
7279 : NULL, areq);
7280 : }
7281 :
7282 : /*
7283 : * postgresForeignAsyncNotify
7284 : * Fetch some more tuples from a file descriptor that becomes ready,
7285 : * requesting next tuple.
7286 : */
7287 : static void
7288 298 : postgresForeignAsyncNotify(AsyncRequest *areq)
7289 : {
7290 298 : ForeignScanState *node = (ForeignScanState *) areq->requestee;
7291 298 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
7292 :
7293 : /* The core code would have initialized the callback_pending flag */
7294 : Assert(!areq->callback_pending);
7295 :
7296 : /*
7297 : * If process_pending_request() has been invoked on the given request
7298 : * before we get here, we might have some tuples already; in which case
7299 : * produce the next tuple
7300 : */
7301 298 : if (fsstate->next_tuple < fsstate->num_tuples)
7302 : {
7303 0 : produce_tuple_asynchronously(areq, true);
7304 0 : return;
7305 : }
7306 :
7307 : /* We must have run out of tuples */
7308 : Assert(fsstate->next_tuple >= fsstate->num_tuples);
7309 :
7310 : /* The request should be currently in-process */
7311 : Assert(fsstate->conn_state->pendingAreq == areq);
7312 :
7313 : /* On error, report the original query, not the FETCH. */
7314 298 : if (!PQconsumeInput(fsstate->conn))
7315 0 : pgfdw_report_error(NULL, fsstate->conn, fsstate->query);
7316 :
7317 298 : fetch_more_data(node);
7318 :
7319 298 : produce_tuple_asynchronously(areq, true);
7320 : }
7321 :
7322 : /*
7323 : * Asynchronously produce next tuple from a foreign PostgreSQL table.
7324 : */
7325 : static void
7326 12658 : produce_tuple_asynchronously(AsyncRequest *areq, bool fetch)
7327 : {
7328 12658 : ForeignScanState *node = (ForeignScanState *) areq->requestee;
7329 12658 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
7330 12658 : AsyncRequest *pendingAreq = fsstate->conn_state->pendingAreq;
7331 : TupleTableSlot *result;
7332 :
7333 : /* This should not be called if the request is currently in-process */
7334 : Assert(areq != pendingAreq);
7335 :
7336 : /* Fetch some more tuples, if we've run out */
7337 12658 : if (fsstate->next_tuple >= fsstate->num_tuples)
7338 : {
7339 : /* No point in another fetch if we already detected EOF, though */
7340 380 : if (!fsstate->eof_reached)
7341 : {
7342 : /* Mark the request as pending for a callback */
7343 258 : ExecAsyncRequestPending(areq);
7344 : /* Begin another fetch if requested and if no pending request */
7345 258 : if (fetch && !pendingAreq)
7346 248 : fetch_more_data_begin(areq);
7347 : }
7348 : else
7349 : {
7350 : /* There's nothing more to do; just return a NULL pointer */
7351 122 : result = NULL;
7352 : /* Mark the request as complete */
7353 122 : ExecAsyncRequestDone(areq, result);
7354 : }
7355 380 : return;
7356 : }
7357 :
7358 : /* Get a tuple from the ForeignScan node */
7359 12278 : result = areq->requestee->ExecProcNodeReal(areq->requestee);
7360 12278 : if (!TupIsNull(result))
7361 : {
7362 : /* Mark the request as complete */
7363 12214 : ExecAsyncRequestDone(areq, result);
7364 12214 : return;
7365 : }
7366 :
7367 : /* We must have run out of tuples */
7368 : Assert(fsstate->next_tuple >= fsstate->num_tuples);
7369 :
7370 : /* Fetch some more tuples, if we've not detected EOF yet */
7371 64 : if (!fsstate->eof_reached)
7372 : {
7373 : /* Mark the request as pending for a callback */
7374 64 : ExecAsyncRequestPending(areq);
7375 : /* Begin another fetch if requested and if no pending request */
7376 64 : if (fetch && !pendingAreq)
7377 60 : fetch_more_data_begin(areq);
7378 : }
7379 : else
7380 : {
7381 : /* There's nothing more to do; just return a NULL pointer */
7382 0 : result = NULL;
7383 : /* Mark the request as complete */
7384 0 : ExecAsyncRequestDone(areq, result);
7385 : }
7386 : }
7387 :
7388 : /*
7389 : * Begin an asynchronous data fetch.
7390 : *
7391 : * Note: this function assumes there is no currently-in-progress asynchronous
7392 : * data fetch.
7393 : *
7394 : * Note: fetch_more_data must be called to fetch the result.
7395 : */
7396 : static void
7397 320 : fetch_more_data_begin(AsyncRequest *areq)
7398 : {
7399 320 : ForeignScanState *node = (ForeignScanState *) areq->requestee;
7400 320 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
7401 : char sql[64];
7402 :
7403 : Assert(!fsstate->conn_state->pendingAreq);
7404 :
7405 : /* Create the cursor synchronously. */
7406 320 : if (!fsstate->cursor_exists)
7407 136 : create_cursor(node);
7408 :
7409 : /* We will send this query, but not wait for the response. */
7410 318 : snprintf(sql, sizeof(sql), "FETCH %d FROM c%u",
7411 : fsstate->fetch_size, fsstate->cursor_number);
7412 :
7413 318 : if (!PQsendQuery(fsstate->conn, sql))
7414 0 : pgfdw_report_error(NULL, fsstate->conn, fsstate->query);
7415 :
7416 : /* Remember that the request is in process */
7417 318 : fsstate->conn_state->pendingAreq = areq;
7418 318 : }
7419 :
7420 : /*
7421 : * Process a pending asynchronous request.
7422 : */
7423 : void
7424 18 : process_pending_request(AsyncRequest *areq)
7425 : {
7426 18 : ForeignScanState *node = (ForeignScanState *) areq->requestee;
7427 18 : PgFdwScanState *fsstate = (PgFdwScanState *) node->fdw_state;
7428 :
7429 : /* The request would have been pending for a callback */
7430 : Assert(areq->callback_pending);
7431 :
7432 : /* The request should be currently in-process */
7433 : Assert(fsstate->conn_state->pendingAreq == areq);
7434 :
7435 18 : fetch_more_data(node);
7436 :
7437 : /*
7438 : * If we didn't get any tuples, must be end of data; complete the request
7439 : * now. Otherwise, we postpone completing the request until we are called
7440 : * from postgresForeignAsyncConfigureWait()/postgresForeignAsyncNotify().
7441 : */
7442 18 : if (fsstate->next_tuple >= fsstate->num_tuples)
7443 : {
7444 : /* Unlike AsyncNotify, we unset callback_pending ourselves */
7445 0 : areq->callback_pending = false;
7446 : /* Mark the request as complete */
7447 0 : ExecAsyncRequestDone(areq, NULL);
7448 : /* Unlike AsyncNotify, we call ExecAsyncResponse ourselves */
7449 0 : ExecAsyncResponse(areq);
7450 : }
7451 18 : }
7452 :
7453 : /*
7454 : * Complete a pending asynchronous request.
7455 : */
7456 : static void
7457 10 : complete_pending_request(AsyncRequest *areq)
7458 : {
7459 : /* The request would have been pending for a callback */
7460 : Assert(areq->callback_pending);
7461 :
7462 : /* Unlike AsyncNotify, we unset callback_pending ourselves */
7463 10 : areq->callback_pending = false;
7464 :
7465 : /* We begin a fetch afterwards if necessary; don't fetch */
7466 10 : produce_tuple_asynchronously(areq, false);
7467 :
7468 : /* Unlike AsyncNotify, we call ExecAsyncResponse ourselves */
7469 10 : ExecAsyncResponse(areq);
7470 :
7471 : /* Also, we do instrumentation ourselves, if required */
7472 10 : if (areq->requestee->instrument)
7473 2 : InstrUpdateTupleCount(areq->requestee->instrument,
7474 2 : TupIsNull(areq->result) ? 0.0 : 1.0);
7475 10 : }
7476 :
7477 : /*
7478 : * Create a tuple from the specified row of the PGresult.
7479 : *
7480 : * rel is the local representation of the foreign table, attinmeta is
7481 : * conversion data for the rel's tupdesc, and retrieved_attrs is an
7482 : * integer list of the table column numbers present in the PGresult.
7483 : * fsstate is the ForeignScan plan node's execution state.
7484 : * temp_context is a working context that can be reset after each tuple.
7485 : *
7486 : * Note: either rel or fsstate, but not both, can be NULL. rel is NULL
7487 : * if we're processing a remote join, while fsstate is NULL in a non-query
7488 : * context such as ANALYZE, or if we're processing a non-scan query node.
7489 : */
7490 : static HeapTuple
7491 178796 : make_tuple_from_result_row(PGresult *res,
7492 : int row,
7493 : Relation rel,
7494 : AttInMetadata *attinmeta,
7495 : List *retrieved_attrs,
7496 : ForeignScanState *fsstate,
7497 : MemoryContext temp_context)
7498 : {
7499 : HeapTuple tuple;
7500 : TupleDesc tupdesc;
7501 : Datum *values;
7502 : bool *nulls;
7503 178796 : ItemPointer ctid = NULL;
7504 : ConversionLocation errpos;
7505 : ErrorContextCallback errcallback;
7506 : MemoryContext oldcontext;
7507 : ListCell *lc;
7508 : int j;
7509 :
7510 : Assert(row < PQntuples(res));
7511 :
7512 : /*
7513 : * Do the following work in a temp context that we reset after each tuple.
7514 : * This cleans up not only the data we have direct access to, but any
7515 : * cruft the I/O functions might leak.
7516 : */
7517 178796 : oldcontext = MemoryContextSwitchTo(temp_context);
7518 :
7519 : /*
7520 : * Get the tuple descriptor for the row. Use the rel's tupdesc if rel is
7521 : * provided, otherwise look to the scan node's ScanTupleSlot.
7522 : */
7523 178796 : if (rel)
7524 106898 : tupdesc = RelationGetDescr(rel);
7525 : else
7526 : {
7527 : Assert(fsstate);
7528 71898 : tupdesc = fsstate->ss.ss_ScanTupleSlot->tts_tupleDescriptor;
7529 : }
7530 :
7531 178796 : values = (Datum *) palloc0(tupdesc->natts * sizeof(Datum));
7532 178796 : nulls = (bool *) palloc(tupdesc->natts * sizeof(bool));
7533 : /* Initialize to nulls for any columns not present in result */
7534 178796 : memset(nulls, true, tupdesc->natts * sizeof(bool));
7535 :
7536 : /*
7537 : * Set up and install callback to report where conversion error occurs.
7538 : */
7539 178796 : errpos.cur_attno = 0;
7540 178796 : errpos.rel = rel;
7541 178796 : errpos.fsstate = fsstate;
7542 178796 : errcallback.callback = conversion_error_callback;
7543 178796 : errcallback.arg = &errpos;
7544 178796 : errcallback.previous = error_context_stack;
7545 178796 : error_context_stack = &errcallback;
7546 :
7547 : /*
7548 : * i indexes columns in the relation, j indexes columns in the PGresult.
7549 : */
7550 178796 : j = 0;
7551 667978 : foreach(lc, retrieved_attrs)
7552 : {
7553 489192 : int i = lfirst_int(lc);
7554 : char *valstr;
7555 :
7556 : /* fetch next column's textual value */
7557 489192 : if (PQgetisnull(res, row, j))
7558 1506 : valstr = NULL;
7559 : else
7560 487686 : valstr = PQgetvalue(res, row, j);
7561 :
7562 : /*
7563 : * convert value to internal representation
7564 : *
7565 : * Note: we ignore system columns other than ctid and oid in result
7566 : */
7567 489192 : errpos.cur_attno = i;
7568 489192 : if (i > 0)
7569 : {
7570 : /* ordinary column */
7571 : Assert(i <= tupdesc->natts);
7572 482958 : nulls[i - 1] = (valstr == NULL);
7573 : /* Apply the input function even to nulls, to support domains */
7574 482948 : values[i - 1] = InputFunctionCall(&attinmeta->attinfuncs[i - 1],
7575 : valstr,
7576 482958 : attinmeta->attioparams[i - 1],
7577 482958 : attinmeta->atttypmods[i - 1]);
7578 : }
7579 6234 : else if (i == SelfItemPointerAttributeNumber)
7580 : {
7581 : /* ctid */
7582 6234 : if (valstr != NULL)
7583 : {
7584 : Datum datum;
7585 :
7586 6234 : datum = DirectFunctionCall1(tidin, CStringGetDatum(valstr));
7587 6234 : ctid = (ItemPointer) DatumGetPointer(datum);
7588 : }
7589 : }
7590 489182 : errpos.cur_attno = 0;
7591 :
7592 489182 : j++;
7593 : }
7594 :
7595 : /* Uninstall error context callback. */
7596 178786 : error_context_stack = errcallback.previous;
7597 :
7598 : /*
7599 : * Check we got the expected number of columns. Note: j == 0 and
7600 : * PQnfields == 1 is expected, since deparse emits a NULL if no columns.
7601 : */
7602 178786 : if (j > 0 && j != PQnfields(res))
7603 0 : elog(ERROR, "remote query result does not match the foreign table");
7604 :
7605 : /*
7606 : * Build the result tuple in caller's memory context.
7607 : */
7608 178786 : MemoryContextSwitchTo(oldcontext);
7609 :
7610 178786 : tuple = heap_form_tuple(tupdesc, values, nulls);
7611 :
7612 : /*
7613 : * If we have a CTID to return, install it in both t_self and t_ctid.
7614 : * t_self is the normal place, but if the tuple is converted to a
7615 : * composite Datum, t_self will be lost; setting t_ctid allows CTID to be
7616 : * preserved during EvalPlanQual re-evaluations (see ROW_MARK_COPY code).
7617 : */
7618 178786 : if (ctid)
7619 6234 : tuple->t_self = tuple->t_data->t_ctid = *ctid;
7620 :
7621 : /*
7622 : * Stomp on the xmin, xmax, and cmin fields from the tuple created by
7623 : * heap_form_tuple. heap_form_tuple actually creates the tuple with
7624 : * DatumTupleFields, not HeapTupleFields, but the executor expects
7625 : * HeapTupleFields and will happily extract system columns on that
7626 : * assumption. If we don't do this then, for example, the tuple length
7627 : * ends up in the xmin field, which isn't what we want.
7628 : */
7629 178786 : HeapTupleHeaderSetXmax(tuple->t_data, InvalidTransactionId);
7630 178786 : HeapTupleHeaderSetXmin(tuple->t_data, InvalidTransactionId);
7631 178786 : HeapTupleHeaderSetCmin(tuple->t_data, InvalidTransactionId);
7632 :
7633 : /* Clean up */
7634 178786 : MemoryContextReset(temp_context);
7635 :
7636 178786 : return tuple;
7637 : }
7638 :
7639 : /*
7640 : * Callback function which is called when error occurs during column value
7641 : * conversion. Print names of column and relation.
7642 : *
7643 : * Note that this function mustn't do any catalog lookups, since we are in
7644 : * an already-failed transaction. Fortunately, we can get the needed info
7645 : * from the relation or the query's rangetable instead.
7646 : */
7647 : static void
7648 10 : conversion_error_callback(void *arg)
7649 : {
7650 10 : ConversionLocation *errpos = (ConversionLocation *) arg;
7651 10 : Relation rel = errpos->rel;
7652 10 : ForeignScanState *fsstate = errpos->fsstate;
7653 10 : const char *attname = NULL;
7654 10 : const char *relname = NULL;
7655 10 : bool is_wholerow = false;
7656 :
7657 : /*
7658 : * If we're in a scan node, always use aliases from the rangetable, for
7659 : * consistency between the simple-relation and remote-join cases. Look at
7660 : * the relation's tupdesc only if we're not in a scan node.
7661 : */
7662 10 : if (fsstate)
7663 : {
7664 : /* ForeignScan case */
7665 8 : ForeignScan *fsplan = castNode(ForeignScan, fsstate->ss.ps.plan);
7666 8 : int varno = 0;
7667 8 : AttrNumber colno = 0;
7668 :
7669 8 : if (fsplan->scan.scanrelid > 0)
7670 : {
7671 : /* error occurred in a scan against a foreign table */
7672 2 : varno = fsplan->scan.scanrelid;
7673 2 : colno = errpos->cur_attno;
7674 : }
7675 : else
7676 : {
7677 : /* error occurred in a scan against a foreign join */
7678 : TargetEntry *tle;
7679 :
7680 6 : tle = list_nth_node(TargetEntry, fsplan->fdw_scan_tlist,
7681 : errpos->cur_attno - 1);
7682 :
7683 : /*
7684 : * Target list can have Vars and expressions. For Vars, we can
7685 : * get some information, however for expressions we can't. Thus
7686 : * for expressions, just show generic context message.
7687 : */
7688 6 : if (IsA(tle->expr, Var))
7689 : {
7690 4 : Var *var = (Var *) tle->expr;
7691 :
7692 4 : varno = var->varno;
7693 4 : colno = var->varattno;
7694 : }
7695 : }
7696 :
7697 8 : if (varno > 0)
7698 : {
7699 6 : EState *estate = fsstate->ss.ps.state;
7700 6 : RangeTblEntry *rte = exec_rt_fetch(varno, estate);
7701 :
7702 6 : relname = rte->eref->aliasname;
7703 :
7704 6 : if (colno == 0)
7705 2 : is_wholerow = true;
7706 4 : else if (colno > 0 && colno <= list_length(rte->eref->colnames))
7707 4 : attname = strVal(list_nth(rte->eref->colnames, colno - 1));
7708 0 : else if (colno == SelfItemPointerAttributeNumber)
7709 0 : attname = "ctid";
7710 : }
7711 : }
7712 2 : else if (rel)
7713 : {
7714 : /* Non-ForeignScan case (we should always have a rel here) */
7715 2 : TupleDesc tupdesc = RelationGetDescr(rel);
7716 :
7717 2 : relname = RelationGetRelationName(rel);
7718 2 : if (errpos->cur_attno > 0 && errpos->cur_attno <= tupdesc->natts)
7719 2 : {
7720 2 : Form_pg_attribute attr = TupleDescAttr(tupdesc,
7721 2 : errpos->cur_attno - 1);
7722 :
7723 2 : attname = NameStr(attr->attname);
7724 : }
7725 0 : else if (errpos->cur_attno == SelfItemPointerAttributeNumber)
7726 0 : attname = "ctid";
7727 : }
7728 :
7729 10 : if (relname && is_wholerow)
7730 2 : errcontext("whole-row reference to foreign table \"%s\"", relname);
7731 8 : else if (relname && attname)
7732 6 : errcontext("column \"%s\" of foreign table \"%s\"", attname, relname);
7733 : else
7734 2 : errcontext("processing expression at position %d in select list",
7735 2 : errpos->cur_attno);
7736 10 : }
7737 :
7738 : /*
7739 : * Given an EquivalenceClass and a foreign relation, find an EC member
7740 : * that can be used to sort the relation remotely according to a pathkey
7741 : * using this EC.
7742 : *
7743 : * If there is more than one suitable candidate, return an arbitrary
7744 : * one of them. If there is none, return NULL.
7745 : *
7746 : * This checks that the EC member expression uses only Vars from the given
7747 : * rel and is shippable. Caller must separately verify that the pathkey's
7748 : * ordering operator is shippable.
7749 : */
7750 : EquivalenceMember *
7751 3600 : find_em_for_rel(PlannerInfo *root, EquivalenceClass *ec, RelOptInfo *rel)
7752 : {
7753 3600 : PgFdwRelationInfo *fpinfo = (PgFdwRelationInfo *) rel->fdw_private;
7754 : EquivalenceMemberIterator it;
7755 : EquivalenceMember *em;
7756 :
7757 3600 : setup_eclass_member_iterator(&it, ec, rel->relids);
7758 6010 : while ((em = eclass_member_iterator_next(&it)) != NULL)
7759 : {
7760 : /*
7761 : * Note we require !bms_is_empty, else we'd accept constant
7762 : * expressions which are not suitable for the purpose.
7763 : */
7764 5450 : if (bms_is_subset(em->em_relids, rel->relids) &&
7765 6194 : !bms_is_empty(em->em_relids) &&
7766 6168 : bms_is_empty(bms_intersect(em->em_relids, fpinfo->hidden_subquery_rels)) &&
7767 3072 : is_foreign_expr(root, rel, em->em_expr))
7768 3040 : return em;
7769 : }
7770 :
7771 560 : return NULL;
7772 : }
7773 :
7774 : /*
7775 : * Find an EquivalenceClass member that is to be computed as a sort column
7776 : * in the given rel's reltarget, and is shippable.
7777 : *
7778 : * If there is more than one suitable candidate, return an arbitrary
7779 : * one of them. If there is none, return NULL.
7780 : *
7781 : * This checks that the EC member expression uses only Vars from the given
7782 : * rel and is shippable. Caller must separately verify that the pathkey's
7783 : * ordering operator is shippable.
7784 : */
7785 : EquivalenceMember *
7786 510 : find_em_for_rel_target(PlannerInfo *root, EquivalenceClass *ec,
7787 : RelOptInfo *rel)
7788 : {
7789 510 : PathTarget *target = rel->reltarget;
7790 : ListCell *lc1;
7791 : int i;
7792 :
7793 510 : i = 0;
7794 850 : foreach(lc1, target->exprs)
7795 : {
7796 850 : Expr *expr = (Expr *) lfirst(lc1);
7797 850 : Index sgref = get_pathtarget_sortgroupref(target, i);
7798 : ListCell *lc2;
7799 :
7800 : /* Ignore non-sort expressions */
7801 1530 : if (sgref == 0 ||
7802 680 : get_sortgroupref_clause_noerr(sgref,
7803 680 : root->parse->sortClause) == NULL)
7804 : {
7805 186 : i++;
7806 186 : continue;
7807 : }
7808 :
7809 : /* We ignore binary-compatible relabeling on both ends */
7810 664 : while (expr && IsA(expr, RelabelType))
7811 0 : expr = ((RelabelType *) expr)->arg;
7812 :
7813 : /*
7814 : * Locate an EquivalenceClass member matching this expr, if any.
7815 : * Ignore child members.
7816 : */
7817 826 : foreach(lc2, ec->ec_members)
7818 : {
7819 672 : EquivalenceMember *em = (EquivalenceMember *) lfirst(lc2);
7820 : Expr *em_expr;
7821 :
7822 : /* Don't match constants */
7823 672 : if (em->em_is_const)
7824 0 : continue;
7825 :
7826 : /* Child members should not exist in ec_members */
7827 : Assert(!em->em_is_child);
7828 :
7829 : /* Match if same expression (after stripping relabel) */
7830 672 : em_expr = em->em_expr;
7831 696 : while (em_expr && IsA(em_expr, RelabelType))
7832 24 : em_expr = ((RelabelType *) em_expr)->arg;
7833 :
7834 672 : if (!equal(em_expr, expr))
7835 162 : continue;
7836 :
7837 : /* Check that expression (including relabels!) is shippable */
7838 510 : if (is_foreign_expr(root, rel, em->em_expr))
7839 510 : return em;
7840 : }
7841 :
7842 154 : i++;
7843 : }
7844 :
7845 0 : return NULL;
7846 : }
7847 :
7848 : /*
7849 : * Determine batch size for a given foreign table. The option specified for
7850 : * a table has precedence.
7851 : */
7852 : static int
7853 292 : get_batch_size_option(Relation rel)
7854 : {
7855 292 : Oid foreigntableid = RelationGetRelid(rel);
7856 : ForeignTable *table;
7857 : ForeignServer *server;
7858 : List *options;
7859 : ListCell *lc;
7860 :
7861 : /* we use 1 by default, which means "no batching" */
7862 292 : int batch_size = 1;
7863 :
7864 : /*
7865 : * Load options for table and server. We append server options after table
7866 : * options, because table options take precedence.
7867 : */
7868 292 : table = GetForeignTable(foreigntableid);
7869 292 : server = GetForeignServer(table->serverid);
7870 :
7871 292 : options = NIL;
7872 292 : options = list_concat(options, table->options);
7873 292 : options = list_concat(options, server->options);
7874 :
7875 : /* See if either table or server specifies batch_size. */
7876 1532 : foreach(lc, options)
7877 : {
7878 1310 : DefElem *def = (DefElem *) lfirst(lc);
7879 :
7880 1310 : if (strcmp(def->defname, "batch_size") == 0)
7881 : {
7882 70 : (void) parse_int(defGetString(def), &batch_size, 0, NULL);
7883 70 : break;
7884 : }
7885 : }
7886 :
7887 292 : return batch_size;
7888 : }
|
