Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * repack.c
4 : : * REPACK a table; formerly known as CLUSTER. VACUUM FULL also uses
5 : : * parts of this code.
6 : : *
7 : : * There are two somewhat different ways to rewrite a table. In non-
8 : : * concurrent mode, it's easy: take AccessExclusiveLock, create a new
9 : : * transient relation, copy the tuples over to the relfilenode of the new
10 : : * relation, swap the relfilenodes, then drop the old relation.
11 : : *
12 : : * In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
13 : : * then do an initial copy as above. However, while the tuples are being
14 : : * copied, concurrent transactions could modify the table. To cope with those
15 : : * changes, we rely on logical decoding to obtain them from WAL. A bgworker
16 : : * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
17 : : * from being reserved), and accumulates the changes in a file. Once the
18 : : * initial copy is complete, we read the changes from the file and re-apply
19 : : * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
20 : : * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
21 : : * a strong lock on the table is much reduced, and the bloat is eliminated.
22 : : *
23 : : *
24 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
25 : : * Portions Copyright (c) 1994-5, Regents of the University of California
26 : : *
27 : : *
28 : : * IDENTIFICATION
29 : : * src/backend/commands/repack.c
30 : : *
31 : : *-------------------------------------------------------------------------
32 : : */
33 : : #include "postgres.h"
34 : :
35 : : #include "access/amapi.h"
36 : : #include "access/heapam.h"
37 : : #include "access/multixact.h"
38 : : #include "access/relscan.h"
39 : : #include "access/tableam.h"
40 : : #include "access/toast_internals.h"
41 : : #include "access/transam.h"
42 : : #include "access/xact.h"
43 : : #include "access/xlog.h"
44 : : #include "catalog/catalog.h"
45 : : #include "catalog/dependency.h"
46 : : #include "catalog/heap.h"
47 : : #include "catalog/index.h"
48 : : #include "catalog/namespace.h"
49 : : #include "catalog/objectaccess.h"
50 : : #include "catalog/pg_am.h"
51 : : #include "catalog/pg_attrdef.h"
52 : : #include "catalog/pg_constraint.h"
53 : : #include "catalog/pg_inherits.h"
54 : : #include "catalog/toasting.h"
55 : : #include "commands/defrem.h"
56 : : #include "commands/progress.h"
57 : : #include "commands/repack.h"
58 : : #include "commands/repack_internal.h"
59 : : #include "commands/tablecmds.h"
60 : : #include "commands/vacuum.h"
61 : : #include "executor/executor.h"
62 : : #include "libpq/pqformat.h"
63 : : #include "libpq/pqmq.h"
64 : : #include "miscadmin.h"
65 : : #include "optimizer/optimizer.h"
66 : : #include "parser/parse_relation.h"
67 : : #include "pgstat.h"
68 : : #include "replication/logicalrelation.h"
69 : : #include "storage/bufmgr.h"
70 : : #include "storage/ipc.h"
71 : : #include "storage/lmgr.h"
72 : : #include "storage/predicate.h"
73 : : #include "storage/proc.h"
74 : : #include "utils/acl.h"
75 : : #include "utils/fmgroids.h"
76 : : #include "utils/guc.h"
77 : : #include "utils/injection_point.h"
78 : : #include "utils/inval.h"
79 : : #include "utils/lsyscache.h"
80 : : #include "utils/memutils.h"
81 : : #include "utils/pg_rusage.h"
82 : : #include "utils/relmapper.h"
83 : : #include "utils/snapmgr.h"
84 : : #include "utils/syscache.h"
85 : : #include "utils/wait_event_types.h"
86 : :
87 : : /*
88 : : * This struct is used to pass around the information on tables to be
89 : : * clustered. We need this so we can make a list of them when invoked without
90 : : * a specific table/index pair.
91 : : */
92 : : typedef struct
93 : : {
94 : : Oid tableOid;
95 : : Oid indexOid;
96 : : } RelToCluster;
97 : :
98 : : /*
99 : : * The first file exported by the decoding worker must contain a snapshot, the
100 : : * following ones contain the data changes.
101 : : */
102 : : #define WORKER_FILE_SNAPSHOT 0
103 : :
104 : : /*
105 : : * Information needed to apply concurrent data changes.
106 : : */
107 : : typedef struct ChangeContext
108 : : {
109 : : /* The relation the changes are applied to. */
110 : : Relation cc_rel;
111 : :
112 : : /* Needed to update indexes of cc_rel. */
113 : : ResultRelInfo *cc_rri;
114 : : EState *cc_estate;
115 : :
116 : : /*
117 : : * Existing tuples to UPDATE and DELETE are located via this index. We
118 : : * keep the scankey in partially initialized state to avoid repeated work.
119 : : * sk_argument is completed on the fly.
120 : : */
121 : : Relation cc_ident_index;
122 : : ScanKey cc_ident_key;
123 : : int cc_ident_key_nentries;
124 : :
125 : : /* The latest column we need to deform to have the tuple identity */
126 : : AttrNumber cc_last_key_attno;
127 : :
128 : : /* Sequential number of the file containing the changes. */
129 : : int cc_file_seq;
130 : : } ChangeContext;
131 : :
132 : : /*
133 : : * Backend-local information to control the decoding worker.
134 : : */
135 : : typedef struct DecodingWorker
136 : : {
137 : : /* The worker. */
138 : : BackgroundWorkerHandle *handle;
139 : :
140 : : /* DecodingWorkerShared is in this segment. */
141 : : dsm_segment *seg;
142 : :
143 : : /* Handle of the error queue. */
144 : : shm_mq_handle *error_mqh;
145 : : } DecodingWorker;
146 : :
147 : : /* Pointer to currently running decoding worker. */
148 : : static DecodingWorker *decoding_worker = NULL;
149 : :
150 : : /*
151 : : * Is there a message sent by a repack worker that the backend needs to
152 : : * receive?
153 : : */
154 : : volatile sig_atomic_t RepackMessagePending = false;
155 : :
156 : : static LOCKMODE RepackLockLevel(bool concurrent);
157 : : static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
158 : : Oid indexOid, Oid userid, LOCKMODE lmode,
159 : : int options);
160 : : static void check_concurrent_repack_requirements(Relation rel,
161 : : Oid *ident_idx_p);
162 : : static void check_index_requirements(Relation rel, RepackCommand cmd);
163 : : static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
164 : : Oid ident_idx);
165 : : static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
166 : : Snapshot snapshot,
167 : : bool verbose,
168 : : bool *pSwapToastByContent,
169 : : TransactionId *pFreezeXid,
170 : : MultiXactId *pCutoffMulti);
171 : : static List *get_tables_to_repack(RepackCommand cmd, bool usingindex,
172 : : MemoryContext permcxt);
173 : : static List *get_tables_to_repack_partitioned(RepackStmt *stmt,
174 : : Relation rel,
175 : : MemoryContext permcxt);
176 : : static bool repack_is_permitted_for_relation(RepackCommand cmd,
177 : : Oid relid, Oid userid,
178 : : bool already_locked);
179 : :
180 : : static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt);
181 : : static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot,
182 : : ChangeContext *chgcxt);
183 : : static void apply_concurrent_update(Relation rel, TupleTableSlot *spilled_tuple,
184 : : TupleTableSlot *ondisk_tuple,
185 : : ChangeContext *chgcxt);
186 : : static void apply_concurrent_delete(Relation rel, TupleTableSlot *slot);
187 : : static void restore_tuple(BufFile *file, Relation relation,
188 : : TupleTableSlot *slot);
189 : : static void adjust_toast_pointers(Relation relation, TupleTableSlot *dest,
190 : : TupleTableSlot *src);
191 : : static bool find_target_tuple(Relation rel, ChangeContext *chgcxt,
192 : : TupleTableSlot *locator,
193 : : TupleTableSlot *retrieved);
194 : : static bool identity_key_equal(ChangeContext *chgcxt,
195 : : TupleTableSlot *locator,
196 : : TupleTableSlot *candidate);
197 : : static void process_concurrent_changes(XLogRecPtr end_of_wal,
198 : : ChangeContext *chgcxt,
199 : : bool done);
200 : : static void initialize_change_context(ChangeContext *chgcxt,
201 : : Relation relation,
202 : : Oid ident_index_id);
203 : : static void release_change_context(ChangeContext *chgcxt);
204 : : static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
205 : : Oid identIdx,
206 : : TransactionId frozenXid,
207 : : MultiXactId cutoffMulti);
208 : : static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
209 : : static void copy_index_constraints(Relation old_index, Oid new_index_id,
210 : : Oid new_heap_id);
211 : : static void copy_attribute_defaults(Oid old_heap_oid, Oid new_heap_oid);
212 : : static Relation process_single_relation(RepackStmt *stmt,
213 : : LOCKMODE lockmode,
214 : : bool isTopLevel,
215 : : ClusterParams *params);
216 : : static Oid determine_clustered_index(Relation rel, bool usingindex,
217 : : const char *indexname);
218 : :
219 : : static void start_repack_decoding_worker(Oid relid);
220 : : static void wait_for_repack_decoding_worker(void);
221 : : static void stop_repack_decoding_worker(void);
222 : : static void stop_repack_decoding_worker_cb(int code, Datum arg);
223 : : static Snapshot get_initial_snapshot(DecodingWorker *worker);
224 : :
225 : : static void ProcessRepackMessage(StringInfo msg);
226 : : static const char *RepackCommandAsString(RepackCommand cmd);
227 : :
228 : :
229 : : /*
230 : : * The repack code allows for processing multiple tables at once. Because
231 : : * of this, we cannot just run everything on a single transaction, or we
232 : : * would be forced to acquire exclusive locks on all the tables being
233 : : * clustered, simultaneously --- very likely leading to deadlock.
234 : : *
235 : : * To solve this we follow a similar strategy to VACUUM code, processing each
236 : : * relation in a separate transaction. For this to work, we need to:
237 : : *
238 : : * - provide a separate memory context so that we can pass information in
239 : : * a way that survives across transactions
240 : : * - start a new transaction every time a new relation is clustered
241 : : * - check for validity of the information on to-be-clustered relations,
242 : : * as someone might have deleted a relation behind our back, or
243 : : * clustered one on a different index
244 : : * - end the transaction
245 : : *
246 : : * The single-relation case does not have any such overhead.
247 : : *
248 : : * We also allow a relation to be repacked following an index, but without
249 : : * naming a specific one. In that case, the indisclustered bit will be
250 : : * looked up, and an ERROR will be thrown if no so-marked index is found.
251 : : */
252 : : void
194 alvherre@kurilemu.de 253 :CBC 265 : ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
254 : : {
2071 michael@paquier.xyz 255 : 265 : ClusterParams params = {0};
1632 alvherre@alvh.no-ip. 256 : 265 : Relation rel = NULL;
257 : : MemoryContext repack_context;
258 : : LOCKMODE lockmode;
259 : : List *rtcs;
125 fujii@postgresql.org 260 : 265 : bool verbose = false;
261 : 265 : bool analyze = false;
262 : 265 : bool concurrently = false;
263 : :
264 : : /* Parse option list */
194 alvherre@kurilemu.de 265 [ + + + + : 576 : foreach_node(DefElem, opt, stmt->params)
+ + ]
266 : : {
2117 michael@paquier.xyz 267 [ + + ]: 46 : if (strcmp(opt->defname, "verbose") == 0)
125 fujii@postgresql.org 268 : 9 : verbose = defGetBoolean(opt);
194 alvherre@kurilemu.de 269 [ + + ]: 37 : else if (strcmp(opt->defname, "analyze") == 0 ||
270 [ - + ]: 21 : strcmp(opt->defname, "analyse") == 0)
271 : : {
38 nathan@postgresql.or 272 [ - + ]: 16 : if (stmt->command != REPACK_COMMAND_REPACK)
38 nathan@postgresql.or 273 [ # # ]:UBC 0 : ereport(ERROR,
274 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
275 : : errmsg("ANALYZE option not supported for %s",
276 : : RepackCommandAsString(stmt->command)));
125 fujii@postgresql.org 277 :CBC 16 : analyze = defGetBoolean(opt);
278 : : }
279 [ + - ]: 21 : else if (strcmp(opt->defname, "concurrently") == 0)
280 : : {
167 alvherre@kurilemu.de 281 [ - + ]: 21 : if (stmt->command != REPACK_COMMAND_REPACK)
167 alvherre@kurilemu.de 282 [ # # ]:UBC 0 : ereport(ERROR,
283 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
284 : : errmsg("CONCURRENTLY option not supported for %s",
285 : : RepackCommandAsString(stmt->command)));
125 fujii@postgresql.org 286 :CBC 21 : concurrently = defGetBoolean(opt);
287 : : }
288 : : else
2117 michael@paquier.xyz 289 [ # # ]:UBC 0 : ereport(ERROR,
290 : : errcode(ERRCODE_SYNTAX_ERROR),
291 : : errmsg("unrecognized %s option \"%s\"",
292 : : RepackCommandAsString(stmt->command),
293 : : opt->defname),
294 : : parser_errposition(pstate, opt->location));
295 : : }
296 : :
125 fujii@postgresql.org 297 :CBC 530 : params.options |=
298 : 530 : (verbose ? CLUOPT_VERBOSE : 0) |
299 [ + + ]: 265 : (analyze ? CLUOPT_ANALYZE : 0) |
300 [ + + ]: 265 : (concurrently ? CLUOPT_CONCURRENT : 0);
301 : :
302 : : /* Determine the lock mode to use. */
167 alvherre@kurilemu.de 303 : 265 : lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
304 : :
305 [ + + ]: 265 : if ((params.options & CLUOPT_CONCURRENT) != 0)
306 : : {
307 : : /*
308 : : * In concurrent mode, make sure we're not in a transaction block.
309 : : *
310 : : * The reason is that repack_setup_logical_decoding() could wait
311 : : * indefinitely for our XID to complete. (The deadlock detector would
312 : : * not recognize it because we'd be waiting for ourselves, i.e. no
313 : : * real lock conflict.) It would be possible to run in a transaction
314 : : * block if we had no XID, but this restriction is simpler for users
315 : : * to understand and we don't lose any functionality.
316 : : */
317 : 21 : PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
318 : : }
12 319 [ + + ]: 244 : else if ((params.options & CLUOPT_ANALYZE) != 0)
320 : : {
321 : : /*
322 : : * With ANALYZE, process_single_relation() would commit the current
323 : : * transaction and start a new one, which would break our state if
324 : : * we're in a transaction block or PL-execution environment. Reject
325 : : * the option in that case. It may be possible to remove this
326 : : * restriction in the future.
327 : : */
328 : 16 : PreventInTransactionBlock(isTopLevel, "REPACK (ANALYZE)");
329 : : }
330 : :
331 : : /*
332 : : * If a single relation is specified, process it and we're done ... unless
333 : : * the relation is a partitioned table, in which case we fall through.
334 : : */
8665 tgl@sss.pgh.pa.us 335 [ + + ]: 257 : if (stmt->relation != NULL)
336 : : {
167 alvherre@kurilemu.de 337 : 239 : rel = process_single_relation(stmt, lockmode, isTopLevel, ¶ms);
194 338 [ + + ]: 190 : if (rel == NULL)
339 : 157 : return; /* all done */
340 : : }
341 : :
342 : : /*
343 : : * Don't allow ANALYZE in the multiple-relation case for now. Maybe we
344 : : * can add support for this later.
345 : : */
346 [ - + ]: 51 : if (params.options & CLUOPT_ANALYZE)
194 alvherre@kurilemu.de 347 [ # # ]:UBC 0 : ereport(ERROR,
348 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
349 : : errmsg("cannot execute %s on multiple tables",
350 : : "REPACK (ANALYZE)"));
351 : :
352 : : /*
353 : : * By here, we know we are in a multi-table situation.
354 : : *
355 : : * Concurrent processing is currently considered rather special (e.g. in
356 : : * terms of resources consumed) so it is not performed in bulk.
357 : : */
167 alvherre@kurilemu.de 358 [ + + ]:CBC 51 : if (params.options & CLUOPT_CONCURRENT)
359 : : {
360 [ + - ]: 1 : if (rel != NULL)
361 : : {
362 [ - + ]: 1 : Assert(rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE);
363 [ + - ]: 1 : ereport(ERROR,
364 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
365 : : errmsg("%s is not supported for partitioned tables",
366 : : "REPACK (CONCURRENTLY)"),
367 : : errhint("Consider running the command on individual partitions."));
368 : : }
369 : : else
167 alvherre@kurilemu.de 370 [ # # ]:UBC 0 : ereport(ERROR,
371 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
372 : : errmsg("%s requires an explicit table name",
373 : : "REPACK (CONCURRENTLY)"));
374 : : }
375 : :
376 : : /*
377 : : * In order to avoid holding locks for too long, we want to process each
378 : : * table in its own transaction. This forces us to disallow running
379 : : * inside a user transaction block.
380 : : */
194 alvherre@kurilemu.de 381 :CBC 50 : PreventInTransactionBlock(isTopLevel, RepackCommandAsString(stmt->command));
382 : :
383 : : /* Also, we need a memory context to hold our list of relations */
384 : 50 : repack_context = AllocSetContextCreate(PortalContext,
385 : : "Repack",
386 : : ALLOCSET_DEFAULT_SIZES);
387 : :
388 : : /*
389 : : * Since we open a new transaction for each relation, we have to check
390 : : * that the relation still is what we think it is.
391 : : *
392 : : * In single-transaction CLUSTER, we don't need the overhead.
393 : : */
394 : 50 : params.options |= CLUOPT_RECHECK;
395 : :
396 : : /*
397 : : * If we don't have a relation yet, determine a relation list. If we do,
398 : : * then it must be a partitioned table, and we want to process its
399 : : * partitions. Note that we don't acquire any locks on these tables, so
400 : : * the returned list must be treated with suspicion.
401 : : */
402 [ + + ]: 50 : if (rel == NULL)
403 : : {
404 [ - + ]: 18 : Assert(stmt->indexname == NULL);
405 : 18 : rtcs = get_tables_to_repack(stmt->command, stmt->usingindex,
406 : : repack_context);
407 : 18 : params.options |= CLUOPT_RECHECK_ISCLUSTERED;
408 : : }
409 : : else
410 : : {
62 411 : 32 : rtcs = get_tables_to_repack_partitioned(stmt, rel, repack_context);
412 : 20 : rel = NULL; /* clobber no longer valid pointer */
413 : : }
414 : :
415 : : /* Commit to get out of starting transaction */
1632 alvherre@alvh.no-ip. 416 : 38 : PopActiveSnapshot();
417 : 38 : CommitTransactionCommand();
418 : :
419 : : /* Cluster the tables, each in a separate transaction */
194 alvherre@kurilemu.de 420 [ - + ]: 38 : Assert(rel == NULL);
421 [ + + + + : 128 : foreach_ptr(RelToCluster, rtc, rtcs)
+ + ]
422 : : {
423 : : /* Start a new transaction for each relation. */
8530 tgl@sss.pgh.pa.us 424 : 52 : StartTransactionCommand();
425 : :
426 : : /*
427 : : * Open the target table. It may have been dropped or replaced with
428 : : * something different, in which case silently skip it.
429 : : */
72 alvherre@kurilemu.de 430 : 52 : rel = try_relation_open(rtc->tableOid, lockmode);
194 431 [ - + ]: 52 : if (rel == NULL)
432 : : {
194 alvherre@kurilemu.de 433 :UBC 0 : CommitTransactionCommand();
434 : 0 : continue;
435 : : }
72 alvherre@kurilemu.de 436 [ - + ]:CBC 52 : if (rel->rd_rel->relkind != RELKIND_RELATION &&
72 alvherre@kurilemu.de 437 [ # # ]:UBC 0 : rel->rd_rel->relkind != RELKIND_MATVIEW)
438 : : {
439 : 0 : relation_close(rel, lockmode);
440 : 0 : CommitTransactionCommand();
441 : 0 : continue;
442 : : }
443 : :
444 : : /* functions in indexes may want a snapshot set */
1632 alvherre@alvh.no-ip. 445 :CBC 52 : PushActiveSnapshot(GetTransactionSnapshot());
446 : :
447 : : /* Process this table */
167 alvherre@kurilemu.de 448 : 52 : cluster_rel(stmt->command, rel, rtc->indexOid, ¶ms, isTopLevel);
449 : : /* cluster_rel closes the relation, but keeps lock */
450 : :
1632 alvherre@alvh.no-ip. 451 : 52 : PopActiveSnapshot();
452 : 52 : CommitTransactionCommand();
453 : : }
454 : :
455 : : /* Start a new transaction for the cleanup work. */
194 alvherre@kurilemu.de 456 : 38 : StartTransactionCommand();
457 : :
458 : : /* Clean up working storage */
459 : 38 : MemoryContextDelete(repack_context);
460 : : }
461 : :
462 : : /*
463 : : * In the non-concurrent case, we obtain AccessExclusiveLock throughout the
464 : : * operation to avoid any lock-upgrade hazards. In the concurrent case, we
465 : : * grab ShareUpdateExclusiveLock (just like VACUUM) for most of the
466 : : * processing and only acquire AccessExclusiveLock at the end, to swap the
467 : : * relation -- supposedly for a short time.
468 : : */
469 : : static LOCKMODE
167 470 : 1603 : RepackLockLevel(bool concurrent)
471 : : {
472 [ + + ]: 1603 : if (concurrent)
473 : 59 : return ShareUpdateExclusiveLock;
474 : : else
475 : 1544 : return AccessExclusiveLock;
476 : : }
477 : :
478 : : /*
479 : : * cluster_rel
480 : : *
481 : : * This clusters the table by creating a new, clustered table and
482 : : * swapping the relfilenumbers of the new table and the old table, so
483 : : * the OID of the original table is preserved. Thus we do not lose
484 : : * GRANT, inheritance nor references to this table.
485 : : *
486 : : * Indexes are rebuilt too, via REINDEX. Since we are effectively bulk-loading
487 : : * the new table, it's better to create the indexes afterwards than to fill
488 : : * them incrementally while we load the table.
489 : : *
490 : : * If indexOid is InvalidOid, the table will be rewritten in physical order
491 : : * instead of index order.
492 : : *
493 : : * Note that, in the concurrent case, the function releases the lock at some
494 : : * point, in order to get AccessExclusiveLock for the final steps (i.e. to
495 : : * swap the relation files). To make things simpler, the caller should expect
496 : : * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
497 : : * AccessExclusiveLock is kept till the end of the transaction.)
498 : : *
499 : : * 'cmd' indicates which command is being executed, to be used for error
500 : : * messages.
501 : : */
502 : : void
194 503 : 462 : cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
504 : : ClusterParams *params, bool isTopLevel)
505 : : {
618 alvherre@alvh.no-ip. 506 : 462 : Oid tableOid = RelationGetRelid(OldHeap);
507 : : Relation index;
508 : : LOCKMODE lmode;
509 : : Oid save_userid;
510 : : int save_sec_context;
511 : : int save_nestlevel;
2071 michael@paquier.xyz 512 : 462 : bool verbose = ((params->options & CLUOPT_VERBOSE) != 0);
513 : 462 : bool recheck = ((params->options & CLUOPT_RECHECK) != 0);
167 alvherre@kurilemu.de 514 : 462 : bool concurrent = ((params->options & CLUOPT_CONCURRENT) != 0);
515 : 462 : Oid ident_idx = InvalidOid;
10 fujii@postgresql.org 516 : 462 : const int progress_index[] = {
517 : : PROGRESS_REPACK_COMMAND,
518 : : PROGRESS_REPACK_INDEX_RELID
519 : : };
520 : 462 : const int64 progress_values[] = {cmd, indexOid};
521 : :
522 : : /* Determine the lock mode to use. */
167 alvherre@kurilemu.de 523 : 462 : lmode = RepackLockLevel(concurrent);
524 : :
525 : : /*
526 : : * Check some preconditions in the concurrent case. This also obtains the
527 : : * replica index OID.
528 : : */
529 [ + + ]: 462 : if (concurrent)
530 : 20 : check_concurrent_repack_requirements(OldHeap, &ident_idx);
531 : :
532 : : /*
533 : : * Also check the state of indexes; this can abort the command for REPACK.
534 : : * Historically this hasn't affected CLUSTER or VACUUM FULL, so don't do
535 : : * it for those commands.
536 : : */
7 537 [ + + ]: 452 : if (cmd == REPACK_COMMAND_REPACK)
538 : 83 : check_index_requirements(OldHeap, cmd);
539 : :
540 : : /* Check for user-requested abort. */
8710 bruce@momjian.us 541 [ - + ]: 450 : CHECK_FOR_INTERRUPTS();
542 : :
194 alvherre@kurilemu.de 543 : 450 : pgstat_progress_start_command(PROGRESS_COMMAND_REPACK, tableOid);
544 : : /* Report the ordering index even when using a sequential scan and sort. */
10 fujii@postgresql.org 545 : 450 : pgstat_progress_update_multi_param(2, progress_index, progress_values);
546 : :
547 : : /*
548 : : * Switch to the table owner's userid, so that any index functions are run
549 : : * as that user. Also lock down security-restricted operations and
550 : : * arrange to make GUC variable changes local to this command.
551 : : */
1595 noah@leadboat.com 552 : 450 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
553 : 450 : SetUserIdAndSecContext(OldHeap->rd_rel->relowner,
554 : : save_sec_context | SECURITY_RESTRICTED_OPERATION);
555 : 450 : save_nestlevel = NewGUCNestLevel();
930 jdavis@postgresql.or 556 : 450 : RestrictSearchPath();
557 : :
558 : : /*
559 : : * Recheck that the relation is still what it was when we started.
560 : : *
561 : : * Note that it's critical to skip this in single-relation CLUSTER;
562 : : * otherwise, we would reject an attempt to cluster using a
563 : : * not-previously-clustered index.
564 : : */
194 alvherre@kurilemu.de 565 [ + + ]: 450 : if (recheck &&
566 [ - + ]: 52 : !cluster_rel_recheck(cmd, OldHeap, indexOid, save_userid,
167 567 : 52 : lmode, params->options))
194 alvherre@kurilemu.de 568 :UBC 0 : goto out;
569 : :
570 : : /*
571 : : * We allow repacking shared catalogs only when not using an index. It
572 : : * would work to use an index in most respects, but the index would only
573 : : * get marked as indisclustered in the current database, leading to
574 : : * unexpected behavior if CLUSTER were later invoked in another database.
575 : : */
6069 tgl@sss.pgh.pa.us 576 [ + + - + ]:CBC 450 : if (OidIsValid(indexOid) && OldHeap->rd_rel->relisshared)
6069 tgl@sss.pgh.pa.us 577 [ # # ]:UBC 0 : ereport(ERROR,
578 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
579 : : /*- translator: first %s is name of a SQL command, eg. REPACK */
580 : : errmsg("cannot execute %s on a shared catalog",
581 : : RepackCommandAsString(cmd)));
582 : :
583 : : /*
584 : : * The CONCURRENTLY case should have been rejected earlier because it does
585 : : * not support system catalogs.
586 : : */
167 alvherre@kurilemu.de 587 [ + + - + ]:CBC 450 : Assert(!(OldHeap->rd_rel->relisshared && concurrent));
588 : :
589 : : /*
590 : : * Don't process temp tables of other backends ... their local buffer
591 : : * manager is not going to cope.
592 : : */
6069 tgl@sss.pgh.pa.us 593 [ + + - + ]: 450 : if (RELATION_IS_OTHER_TEMP(OldHeap))
194 alvherre@kurilemu.de 594 [ # # ]:UBC 0 : ereport(ERROR,
595 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
596 : : /*- translator: first %s is name of a SQL command, eg. REPACK */
597 : : errmsg("cannot execute %s on temporary tables of other sessions",
598 : : RepackCommandAsString(cmd)));
599 : :
600 : : /*
601 : : * Also check for active uses of the relation in the current transaction,
602 : : * including open scans and pending AFTER trigger events.
603 : : */
194 alvherre@kurilemu.de 604 :CBC 450 : CheckTableNotInUse(OldHeap, RepackCommandAsString(cmd));
605 : :
606 : : /* Check heap and index are valid to cluster on */
6069 tgl@sss.pgh.pa.us 607 [ + + ]: 450 : if (OidIsValid(indexOid))
608 : : {
609 : : /* verify the index is good and lock it */
167 alvherre@kurilemu.de 610 : 159 : check_index_is_clusterable(OldHeap, indexOid, lmode);
611 : : /* also open it */
618 alvherre@alvh.no-ip. 612 : 159 : index = index_open(indexOid, NoLock);
613 : : }
614 : : else
615 : 291 : index = NULL;
616 : :
617 : : /*
618 : : * When allow_system_table_mods is turned off, we disallow repacking a
619 : : * catalog on a particular index unless that's already the clustered index
620 : : * for that catalog.
621 : : *
622 : : * XXX We don't check for this in CLUSTER, because it's historically been
623 : : * allowed.
624 : : */
194 alvherre@kurilemu.de 625 [ + + ]: 450 : if (cmd != REPACK_COMMAND_CLUSTER &&
626 [ + - + + ]: 318 : !allowSystemTableMods && OidIsValid(indexOid) &&
627 [ + + + - ]: 27 : IsCatalogRelation(OldHeap) && !index->rd_index->indisclustered)
628 [ + - ]: 4 : ereport(ERROR,
629 : : errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
630 : : errmsg("permission denied: \"%s\" is a system catalog",
631 : : RelationGetRelationName(OldHeap)),
632 : : errdetail("System catalogs can only be clustered by the index they're already clustered on, if any, unless \"%s\" is enabled.",
633 : : "allow_system_table_mods"));
634 : :
635 : : /*
636 : : * Quietly ignore the request if this is a materialized view which has not
637 : : * been populated from its query. No harm is done because there is no data
638 : : * to deal with, and we don't want to throw an error if this is part of a
639 : : * multi-relation request -- for example, CLUSTER was run on the entire
640 : : * database.
641 : : */
4949 kgrittn@postgresql.o 642 [ + + ]: 446 : if (OldHeap->rd_rel->relkind == RELKIND_MATVIEW &&
4885 tgl@sss.pgh.pa.us 643 [ + - ]: 8 : !RelationIsPopulated(OldHeap))
644 : : {
167 alvherre@kurilemu.de 645 [ + - ]: 8 : if (index)
646 : 8 : index_close(index, lmode);
647 : 8 : relation_close(OldHeap, lmode);
1595 noah@leadboat.com 648 : 8 : goto out;
649 : : }
650 : :
1632 alvherre@alvh.no-ip. 651 [ + + + - : 438 : Assert(OldHeap->rd_rel->relkind == RELKIND_RELATION ||
- + ]
652 : : OldHeap->rd_rel->relkind == RELKIND_MATVIEW ||
653 : : OldHeap->rd_rel->relkind == RELKIND_TOASTVALUE);
654 : :
655 : : /*
656 : : * All predicate locks on the tuples or pages are about to be made
657 : : * invalid, because we move tuples around. Promote them to relation
658 : : * locks. Predicate locks on indexes will be promoted when they are
659 : : * reindexed.
660 : : *
661 : : * During concurrent processing, the heap as well as its indexes stay in
662 : : * operation, so we postpone this step until they are locked using
663 : : * AccessExclusiveLock near the end of the processing.
664 : : */
167 alvherre@kurilemu.de 665 [ + + ]: 438 : if (!concurrent)
666 : 429 : TransferPredicateLocksToHeapRelation(OldHeap);
667 : :
668 : : /*
669 : : * rebuild_relation does all the dirty work, and closes OldHeap and index,
670 : : * if valid.
671 : : *
672 : : * In concurrent mode, make sure the worker terminates; normally it does
673 : : * so by itself, but a PG_ENSURE_ERROR_CLEANUP callback ensures that this
674 : : * happens even in case this backend dies early on a FATAL exit. Normal
675 : : * mode doesn't need that overhead.
676 : : */
124 677 [ + + ]: 438 : if (concurrent)
678 : : {
679 [ + - ]: 9 : PG_ENSURE_ERROR_CLEANUP(stop_repack_decoding_worker_cb, 0);
680 : : {
681 : 9 : rebuild_relation(OldHeap, index, verbose, ident_idx);
682 : : }
683 [ - + ]: 9 : PG_END_ENSURE_ERROR_CLEANUP(stop_repack_decoding_worker_cb, 0);
684 : 9 : stop_repack_decoding_worker();
685 : : }
686 : : else
687 : 429 : rebuild_relation(OldHeap, index, verbose, ident_idx);
688 : :
1595 noah@leadboat.com 689 : 442 : out:
690 : : /* Roll back any GUC changes executed by index functions */
691 : 442 : AtEOXact_GUC(false, save_nestlevel);
692 : :
693 : : /* Restore userid and security context */
694 : 442 : SetUserIdAndSecContext(save_userid, save_sec_context);
695 : :
2736 rhaas@postgresql.org 696 : 442 : pgstat_progress_end_command();
8172 tgl@sss.pgh.pa.us 697 : 442 : }
698 : :
699 : : /*
700 : : * Check if the table (and its index) still meets the requirements of
701 : : * cluster_rel().
702 : : */
703 : : static bool
194 alvherre@kurilemu.de 704 : 52 : cluster_rel_recheck(RepackCommand cmd, Relation OldHeap, Oid indexOid,
705 : : Oid userid, LOCKMODE lmode, int options)
706 : : {
707 : 52 : Oid tableOid = RelationGetRelid(OldHeap);
708 : :
72 709 [ - + ]: 52 : Assert(CheckRelationLockedByMe(OldHeap, lmode, false));
710 : :
711 : : /* Check that the user still has privileges for the relation */
32 712 [ - + ]: 52 : if (!repack_is_permitted_for_relation(cmd, tableOid, userid, true))
713 : : {
167 alvherre@kurilemu.de 714 :UBC 0 : relation_close(OldHeap, lmode);
194 715 : 0 : return false;
716 : : }
717 : :
718 : : /*
719 : : * Silently skip a temp table for a remote session. Only doing this check
720 : : * in the "recheck" case is appropriate (which currently means somebody is
721 : : * executing a database-wide CLUSTER or on a partitioned table), because
722 : : * there is another check in cluster() which will stop any attempt to
723 : : * cluster remote temp tables by name. There is another check in
724 : : * cluster_rel which is redundant, but we leave it for extra safety.
725 : : */
194 alvherre@kurilemu.de 726 [ - + - - ]:CBC 52 : if (RELATION_IS_OTHER_TEMP(OldHeap))
727 : : {
167 alvherre@kurilemu.de 728 :UBC 0 : relation_close(OldHeap, lmode);
194 729 : 0 : return false;
730 : : }
731 : :
194 alvherre@kurilemu.de 732 [ + + ]:CBC 52 : if (OidIsValid(indexOid))
733 : : {
734 : : /*
735 : : * Check that the index still exists
736 : : */
737 [ - + ]: 32 : if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(indexOid)))
738 : : {
167 alvherre@kurilemu.de 739 :UBC 0 : relation_close(OldHeap, lmode);
194 740 : 0 : return false;
741 : : }
742 : :
743 : : /*
744 : : * Check that the index is still the one with indisclustered set, if
745 : : * needed.
746 : : */
194 alvherre@kurilemu.de 747 [ + + ]:CBC 32 : if ((options & CLUOPT_RECHECK_ISCLUSTERED) != 0 &&
748 [ - + ]: 4 : !get_index_isclustered(indexOid))
749 : : {
167 alvherre@kurilemu.de 750 :UBC 0 : relation_close(OldHeap, lmode);
194 751 : 0 : return false;
752 : : }
753 : : }
754 : :
194 alvherre@kurilemu.de 755 :CBC 52 : return true;
756 : : }
757 : :
758 : : /*
759 : : * Verify that the specified heap and index are valid to cluster on
760 : : *
761 : : * Side effect: obtains lock on the index. The caller may
762 : : * in some cases already have a lock of the same strength on the table, but
763 : : * not in all cases so we can't rely on the table-level lock for
764 : : * protection here.
765 : : */
766 : : void
1621 michael@paquier.xyz 767 : 357 : check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode)
768 : : {
769 : : Relation OldIndex;
770 : :
5897 simon@2ndQuadrant.co 771 : 357 : OldIndex = index_open(indexOid, lockmode);
772 : :
773 : : /*
774 : : * Check that index is in fact an index on the given relation
775 : : */
8806 tgl@sss.pgh.pa.us 776 [ + - ]: 357 : if (OldIndex->rd_index == NULL ||
8172 777 [ + + ]: 357 : OldIndex->rd_index->indrelid != RelationGetRelid(OldHeap))
8463 778 [ + - ]: 4 : ereport(ERROR,
779 : : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
780 : : errmsg("\"%s\" is not an index for table \"%s\"",
781 : : RelationGetRelationName(OldIndex),
782 : : RelationGetRelationName(OldHeap))));
783 : :
784 : : /* Index AM must allow clustering */
2799 andres@anarazel.de 785 [ + + ]: 353 : if (!OldIndex->rd_indam->amclusterable)
6069 tgl@sss.pgh.pa.us 786 [ + - ]: 4 : ereport(ERROR,
787 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
788 : : errmsg("cannot cluster on index \"%s\" because access method does not support clustering",
789 : : RelationGetRelationName(OldIndex))));
790 : :
791 : : /*
792 : : * Disallow clustering on incomplete indexes (those that might not index
793 : : * every row of the relation). We could relax this by making a separate
794 : : * seqscan pass over the table to copy the missing rows, but that seems
795 : : * expensive and tedious.
796 : : */
3098 andrew@dunslane.net 797 [ + + ]: 349 : if (!heap_attisnull(OldIndex->rd_indextuple, Anum_pg_index_indpred, NULL))
8463 tgl@sss.pgh.pa.us 798 [ + - ]: 4 : ereport(ERROR,
799 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
800 : : errmsg("cannot cluster on partial index \"%s\"",
801 : : RelationGetRelationName(OldIndex))));
802 : :
803 : : /*
804 : : * Disallow if index is left over from a failed CREATE INDEX CONCURRENTLY;
805 : : * it might well not contain entries for every heap row, or might not even
806 : : * be internally consistent. (But note that we don't check indcheckxmin;
807 : : * the worst consequence of following broken HOT chains would be that we
808 : : * might put recently-dead tuples out-of-order in the new table, and there
809 : : * is little harm in that.)
810 : : */
2824 peter_e@gmx.net 811 [ + + ]: 345 : if (!OldIndex->rd_index->indisvalid)
6931 tgl@sss.pgh.pa.us 812 [ + - ]: 4 : ereport(ERROR,
813 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
814 : : errmsg("cannot cluster on invalid index \"%s\"",
815 : : RelationGetRelationName(OldIndex))));
816 : :
817 : : /* Drop relcache refcnt on OldIndex, but keep lock */
7356 818 : 341 : index_close(OldIndex, NoLock);
8702 bruce@momjian.us 819 : 341 : }
820 : :
821 : : /*
822 : : * mark_index_clustered: mark the specified index as the one clustered on
823 : : *
824 : : * With indexOid == InvalidOid, will mark all indexes of rel not-clustered.
825 : : */
826 : : void
4935 rhaas@postgresql.org 827 : 194 : mark_index_clustered(Relation rel, Oid indexOid, bool is_internal)
828 : : {
829 : : HeapTuple indexTuple;
830 : : Form_pg_index indexForm;
831 : : Relation pg_index;
832 : : ListCell *index;
833 : :
188 michael@paquier.xyz 834 [ - + ]: 194 : Assert(rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE);
835 : :
836 : : /*
837 : : * If the index is already marked clustered, no need to do anything.
838 : : */
8170 tgl@sss.pgh.pa.us 839 [ + + ]: 194 : if (OidIsValid(indexOid))
840 : : {
2358 michael@paquier.xyz 841 [ + + ]: 186 : if (get_index_isclustered(indexOid))
8170 tgl@sss.pgh.pa.us 842 : 39 : return;
843 : : }
844 : :
845 : : /*
846 : : * Check each index of the relation and set/clear the bit as needed.
847 : : */
2799 andres@anarazel.de 848 : 155 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
849 : :
8170 tgl@sss.pgh.pa.us 850 [ + - + + : 464 : foreach(index, RelationGetIndexList(rel))
+ + ]
851 : : {
8057 bruce@momjian.us 852 : 309 : Oid thisIndexOid = lfirst_oid(index);
853 : :
6062 rhaas@postgresql.org 854 : 309 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
855 : : ObjectIdGetDatum(thisIndexOid));
8170 tgl@sss.pgh.pa.us 856 [ - + ]: 309 : if (!HeapTupleIsValid(indexTuple))
8170 tgl@sss.pgh.pa.us 857 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for index %u", thisIndexOid);
8170 tgl@sss.pgh.pa.us 858 :CBC 309 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
859 : :
860 : : /*
861 : : * Unset the bit if set. We know it's wrong because we checked this
862 : : * earlier.
863 : : */
864 [ + + ]: 309 : if (indexForm->indisclustered)
865 : : {
866 : 20 : indexForm->indisclustered = false;
3519 alvherre@alvh.no-ip. 867 : 20 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
868 : : }
8170 tgl@sss.pgh.pa.us 869 [ + + ]: 289 : else if (thisIndexOid == indexOid)
870 : : {
871 : : /* this was checked earlier, but let's be real sure */
2824 peter_e@gmx.net 872 [ - + ]: 147 : if (!indexForm->indisvalid)
5044 tgl@sss.pgh.pa.us 873 [ # # ]:UBC 0 : elog(ERROR, "cannot cluster on invalid index %u", indexOid);
8170 tgl@sss.pgh.pa.us 874 :CBC 147 : indexForm->indisclustered = true;
3519 alvherre@alvh.no-ip. 875 : 147 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
876 : : }
877 : :
4935 rhaas@postgresql.org 878 [ - + ]: 309 : InvokeObjectPostAlterHookArg(IndexRelationId, thisIndexOid, 0,
879 : : InvalidOid, is_internal);
880 : :
8170 tgl@sss.pgh.pa.us 881 : 309 : heap_freetuple(indexTuple);
882 : : }
883 : :
2799 andres@anarazel.de 884 : 155 : table_close(pg_index, RowExclusiveLock);
885 : : }
886 : :
887 : : /*
888 : : * check_index_requirements: verify index state on relation being processed
889 : : *
890 : : * Throw an error if any incompletely-built indexes are found.
891 : : *
892 : : * Indexes that are not ready for inserts, such as ones left behind by failed
893 : : * CREATE INDEX CONCURRENTLY, are not maintained by DML. Indexes that aren't
894 : : * marked valid could have been in the middle of validation when their build
895 : : * failed, and thus it's not certain that they could be built. In both cases,
896 : : * attempting to rebuild may fail altogether. Throwing an error here forces
897 : : * the user to take action on these indexes separately from the table
898 : : * reconstruction, which prevents perpetuating them for no reason.
899 : : */
900 : : static void
7 alvherre@kurilemu.de 901 : 83 : check_index_requirements(Relation rel, RepackCommand cmd)
902 : : {
903 : : Relation indrel;
904 : : SysScanDesc indscan;
905 : : ScanKeyData skey;
906 : : HeapTuple htup;
907 : 83 : int num_invalid_idxs = 0;
908 : : StringInfoData dest;
909 : :
910 : 83 : initStringInfo(&dest);
911 : :
912 : : /* Prepare to scan pg_index for entries having indrelid = this rel. */
913 : 83 : ScanKeyInit(&skey,
914 : : Anum_pg_index_indrelid,
915 : : BTEqualStrategyNumber, F_OIDEQ,
916 : : ObjectIdGetDatum(RelationGetRelid(rel)));
917 : :
918 : 83 : indrel = table_open(IndexRelationId, AccessShareLock);
919 : 83 : indscan = systable_beginscan(indrel, IndexIndrelidIndexId, true,
920 : : NULL, 1, &skey);
921 : :
922 [ + + ]: 334 : while (HeapTupleIsValid(htup = systable_getnext(indscan)))
923 : : {
924 : 168 : Form_pg_index index = (Form_pg_index) GETSTRUCT(htup);
925 : :
5 926 [ + + ]: 168 : if (!index->indisvalid)
927 : : {
7 928 [ + + ]: 4 : if (num_invalid_idxs == 0)
929 : 2 : appendStringInfo(&dest, _("\"%s\""), get_rel_name(index->indexrelid));
930 : : else
931 : 2 : appendStringInfo(&dest, _(", \"%s\""), get_rel_name(index->indexrelid));
932 : 4 : num_invalid_idxs++;
933 : : }
934 : : }
935 : 83 : systable_endscan(indscan);
936 : 83 : table_close(indrel, AccessShareLock);
937 : :
938 [ + + ]: 83 : if (num_invalid_idxs > 0)
939 [ + - ]: 2 : ereport(ERROR,
940 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
941 : : errmsg("cannot execute %s on relation \"%s\"",
942 : : RepackCommandAsString(cmd), RelationGetRelationName(rel)),
943 : : errdetail_plural("An invalid index cannot be processed correctly: %s.",
944 : : "Some invalid indexes cannot be processed correctly: %s.",
945 : : num_invalid_idxs,
946 : : dest.data),
947 : : errhint("Use DROP INDEX or REINDEX."));
948 : 81 : }
949 : :
950 : : /*
951 : : * Check if the CONCURRENTLY option is legal for the relation.
952 : : *
953 : : * *Ident_idx_p receives OID of the identity index.
954 : : */
955 : : static void
167 956 : 20 : check_concurrent_repack_requirements(Relation rel, Oid *ident_idx_p)
957 : : {
958 : : char relpersistence,
959 : : replident;
960 : : Oid ident_idx;
961 : :
115 962 [ - + ]: 20 : if (wal_level < WAL_LEVEL_REPLICA)
115 alvherre@kurilemu.de 963 [ # # ]:UBC 0 : ereport(ERROR,
964 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
965 : : errmsg("cannot execute %s in this configuration",
966 : : "REPACK (CONCURRENTLY)"),
967 : : errdetail("This operation requires \"wal_level\" to be set to \"replica\" or higher."));
968 : :
969 : : /*
970 : : * A table AM that doesn't support logical decoding would cause REPACK
971 : : * (CONCURRENTLY) to silently lose the changes made during the rewrite.
972 : : * Nothing in TableAmRoutine tells us whether it does, so for now restrict
973 : : * to heap. Check the routine rather than the AM OID, so that an AM
974 : : * reusing the heap handler still works.
975 : : */
12 msawada@postgresql.o 976 [ - + ]:CBC 20 : if (rel->rd_tableam != GetHeapamTableAmRoutine())
12 msawada@postgresql.o 977 [ # # ]:UBC 0 : ereport(ERROR,
978 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
979 : : errmsg("cannot execute %s on relation \"%s\"",
980 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
981 : : errdetail("This operation is only supported for the \"heap\" access method."));
982 : :
983 : : /* Data changes in system relations are not logically decoded. */
167 alvherre@kurilemu.de 984 [ + + ]:CBC 20 : if (IsCatalogRelation(rel))
985 [ + - ]: 1 : ereport(ERROR,
986 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
987 : : errmsg("cannot execute %s on relation \"%s\"",
988 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
989 : : errdetail("This operation is not supported for system catalogs."));
990 : :
991 : : /*
992 : : * REPACK (CONCURRENTLY) is not MVCC-safe; it doesn't preserve visibility
993 : : * information, which logical decoding needs because it reads user catalog
994 : : * tables under a historic snapshot. Removing this check requires making
995 : : * it MVCC-safe and logical rewrite mappings.
996 : : */
12 msawada@postgresql.o 997 [ + + - + : 19 : if (RelationIsUsedAsCatalogTable(rel))
- - + + ]
998 [ + - ]: 1 : ereport(ERROR,
999 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1000 : : errmsg("cannot execute %s on relation \"%s\"",
1001 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
1002 : : errdetail("This operation is not supported for user catalog tables."));
1003 : :
1004 : : /*
1005 : : * reorderbuffer.c does not seem to handle processing of TOAST relation
1006 : : * alone.
1007 : : */
167 alvherre@kurilemu.de 1008 [ + + ]: 18 : if (IsToastRelation(rel))
1009 [ + - ]: 1 : ereport(ERROR,
1010 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1011 : : errmsg("cannot execute %s on relation \"%s\"",
1012 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
1013 : : errdetail("This operation is not supported for TOAST tables."));
1014 : :
1015 : 17 : relpersistence = rel->rd_rel->relpersistence;
1016 [ + + ]: 17 : if (relpersistence != RELPERSISTENCE_PERMANENT)
1017 [ + - ]: 2 : ereport(ERROR,
1018 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1019 : : errmsg("cannot execute %s on relation \"%s\"",
1020 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
1021 : : errdetail("This operation is only supported for permanent relations."));
1022 : :
1023 : : /* A materialized view produces no logically decoded changes. */
11 1024 [ + + ]: 15 : if (rel->rd_rel->relkind == RELKIND_MATVIEW)
1025 [ + - ]: 1 : ereport(ERROR,
1026 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1027 : : errmsg("cannot execute %s on relation \"%s\"",
1028 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
1029 : : errdetail_relkind_not_supported(rel->rd_rel->relkind));
1030 : :
1031 : : /*
1032 : : * With NOTHING, WAL does not contain the old tuple; FULL is not yet
1033 : : * supported.
1034 : : */
167 1035 : 14 : replident = rel->rd_rel->relreplident;
115 1036 [ + + - + ]: 14 : if (replident == REPLICA_IDENTITY_NOTHING ||
1037 : : replident == REPLICA_IDENTITY_FULL)
167 1038 [ + - + - ]: 1 : ereport(ERROR,
1039 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1040 : : errmsg("cannot execute %s on relation \"%s\"",
1041 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
1042 : : errdetail("This operation does not support tables with %s.",
1043 : : replident == REPLICA_IDENTITY_NOTHING ?
1044 : : "REPLICA IDENTITY NOTHING" : "REPLICA IDENTITY FULL"));
1045 : :
1046 : : /*
1047 : : * Obtain the replica identity index to use. If there isn't one, the
1048 : : * table cannot be repacked concurrently. (Replica identity FULL is not
1049 : : * supported yet.)
1050 : : */
9 1051 : 13 : ident_idx = RelationGetReplicaIndex(rel);
167 1052 [ + + ]: 13 : if (!OidIsValid(ident_idx))
1053 : : {
1054 : : /* This special case warrants its own error message */
115 1055 [ + + + + ]: 3 : if (OidIsValid(rel->rd_pkindex) && rel->rd_ispkdeferrable)
1056 [ + - ]: 1 : ereport(ERROR,
1057 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1058 : : errmsg("cannot execute %s on relation \"%s\"",
1059 : : "REPACK (CONCURRENTLY)",
1060 : : RelationGetRelationName(rel)),
1061 : : errdetail("This operation does not support deferrable primary keys."),
1062 : : errhint("Use ALTER TABLE ... REPLICA IDENTITY USING INDEX to designate another index as replica identity."));
1063 : :
167 1064 [ + - ]: 2 : ereport(ERROR,
1065 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1066 : : errmsg("cannot execute %s on relation \"%s\"",
1067 : : "REPACK (CONCURRENTLY)", RelationGetRelationName(rel)),
1068 : : errdetail("Relation \"%s\" has no identity index.",
1069 : : RelationGetRelationName(rel)));
1070 : : }
1071 : :
1072 : 10 : *ident_idx_p = ident_idx;
1073 : 10 : }
1074 : :
1075 : :
1076 : : /*
1077 : : * rebuild_relation: rebuild an existing relation in index or physical order
1078 : : *
1079 : : * OldHeap: table to rebuild. See cluster_rel() for comments on the required
1080 : : * lock strength.
1081 : : *
1082 : : * index: index to cluster by, or NULL to rewrite in physical order.
1083 : : *
1084 : : * ident_idx: identity index, to handle replaying of concurrent data changes
1085 : : * to the new heap. InvalidOid if there's no CONCURRENTLY option.
1086 : : *
1087 : : * On entry, heap and index (if one is given) must be open, and the
1088 : : * appropriate lock held on them -- AccessExclusiveLock for exclusive
1089 : : * processing and ShareUpdateExclusiveLock for concurrent processing.
1090 : : *
1091 : : * On exit, they are closed, but still locked with AccessExclusiveLock.
1092 : : * (The function handles the lock upgrade if 'concurrent' is true.)
1093 : : */
1094 : : static void
1095 : 438 : rebuild_relation(Relation OldHeap, Relation index, bool verbose,
1096 : : Oid ident_idx)
1097 : : {
8665 tgl@sss.pgh.pa.us 1098 : 438 : Oid tableOid = RelationGetRelid(OldHeap);
1880 michael@paquier.xyz 1099 : 438 : Oid accessMethod = OldHeap->rd_rel->relam;
8106 tgl@sss.pgh.pa.us 1100 : 438 : Oid tableSpace = OldHeap->rd_rel->reltablespace;
1101 : : Oid OIDNewHeap;
1102 : : Relation NewHeap;
1103 : : char relpersistence;
1104 : : bool swap_toast_by_content;
1105 : : TransactionId frozenXid;
1106 : : MultiXactId cutoffMulti;
167 alvherre@kurilemu.de 1107 : 438 : bool concurrent = OidIsValid(ident_idx);
1108 : 438 : Snapshot snapshot = NULL;
1109 : : #if USE_ASSERT_CHECKING
1110 : : LOCKMODE lmode;
1111 : :
1112 : 438 : lmode = RepackLockLevel(concurrent);
1113 : :
1114 [ - + ]: 438 : Assert(CheckRelationLockedByMe(OldHeap, lmode, false));
1115 [ + + - + ]: 438 : Assert(index == NULL || CheckRelationLockedByMe(index, lmode, false));
1116 : : #endif
1117 : :
1118 [ + + ]: 438 : if (concurrent)
1119 : : {
1120 : : /*
1121 : : * The worker needs to be member of the locking group we're the leader
1122 : : * of. We ought to become the leader before the worker starts. The
1123 : : * worker will join the group as soon as it starts.
1124 : : *
1125 : : * This is to make sure that the deadlock described below is
1126 : : * detectable by deadlock.c: if the worker waits for a transaction to
1127 : : * complete and we are waiting for the worker output, then effectively
1128 : : * we (i.e. this backend) are waiting for that transaction.
1129 : : */
1130 : 9 : BecomeLockGroupLeader();
1131 : :
1132 : : /*
1133 : : * Start the worker that decodes data changes applied while we're
1134 : : * copying the table contents.
1135 : : *
1136 : : * Note that the worker has to wait for all transactions with XID
1137 : : * already assigned to finish. If some of those transactions is
1138 : : * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
1139 : : * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
1140 : : * Not sure this risk is worth unlocking/locking the table (and its
1141 : : * clustering index) and checking again if it's still eligible for
1142 : : * REPACK CONCURRENTLY.
1143 : : */
1144 : 9 : start_repack_decoding_worker(tableOid);
1145 : :
1146 : : /*
1147 : : * Wait until the worker has the initial snapshot and retrieve it.
1148 : : */
1149 : 9 : snapshot = get_initial_snapshot(decoding_worker);
1150 : :
1151 : 9 : PushActiveSnapshot(snapshot);
1152 : : }
1153 : :
1154 : : /* for CLUSTER or REPACK USING INDEX, mark the index as the one to use */
194 1155 [ + + ]: 438 : if (index != NULL)
618 alvherre@alvh.no-ip. 1156 : 147 : mark_index_clustered(OldHeap, RelationGetRelid(index), true);
1157 : :
1158 : : /* Remember info about rel before closing OldHeap */
3487 tgl@sss.pgh.pa.us 1159 : 438 : relpersistence = OldHeap->rd_rel->relpersistence;
1160 : :
1161 : : /*
1162 : : * Create the transient table that will receive the re-ordered data.
1163 : : *
1164 : : * OldHeap is already locked, so no need to lock it again. make_new_heap
1165 : : * obtains AccessExclusiveLock on the new heap and its toast table.
1166 : : */
4412 alvherre@alvh.no-ip. 1167 : 438 : OIDNewHeap = make_new_heap(tableOid, tableSpace,
1168 : : accessMethod,
1169 : : relpersistence,
1170 : : NoLock);
618 1171 [ - + ]: 438 : Assert(CheckRelationOidLockedByMe(OIDNewHeap, AccessExclusiveLock, false));
1172 : 438 : NewHeap = table_open(OIDNewHeap, NoLock);
1173 : :
1174 : : /*
1175 : : * In concurrent mode, create a copy of the attribute defaults on the temp
1176 : : * table, which the executor needs when replaying concurrent data changes.
1177 : : */
79 alvherre@kurilemu.de 1178 [ + + ]: 438 : if (concurrent)
1179 : 9 : copy_attribute_defaults(tableOid, OIDNewHeap);
1180 : :
1181 : : /* Copy the heap data into the new table in the desired order */
167 1182 : 438 : copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
1183 : : &swap_toast_by_content, &frozenXid, &cutoffMulti);
1184 : :
1185 : : /* The historic snapshot won't be needed anymore. */
1186 [ + + ]: 438 : if (snapshot)
1187 : : {
1188 : 9 : PopActiveSnapshot();
1189 : 9 : UpdateActiveSnapshotCommandId();
1190 : : }
1191 : :
1192 [ + + ]: 438 : if (concurrent)
1193 : : {
1194 [ - + ]: 9 : Assert(!swap_toast_by_content);
1195 : :
1196 : : /*
1197 : : * Close the index, but keep the lock. Both heaps will be closed by
1198 : : * the following call.
1199 : : */
1200 [ + + ]: 9 : if (index)
1201 : 3 : index_close(index, NoLock);
1202 : :
1203 : 9 : rebuild_relation_finish_concurrent(NewHeap, OldHeap, ident_idx,
1204 : : frozenXid, cutoffMulti);
1205 : :
1206 : 9 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
1207 : : PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
1208 : : }
1209 : : else
1210 : : {
1211 : 429 : bool is_system_catalog = IsSystemRelation(OldHeap);
1212 : :
1213 : : /* Close relcache entries, but keep lock until transaction commit */
1214 : 429 : table_close(OldHeap, NoLock);
1215 [ + + ]: 429 : if (index)
1216 : 144 : index_close(index, NoLock);
1217 : :
1218 : : /*
1219 : : * Close the new relation so it can be dropped as soon as the storage
1220 : : * is swapped. The relation is not visible to others, so no need to
1221 : : * unlock it explicitly.
1222 : : */
1223 : 429 : table_close(NewHeap, NoLock);
1224 : :
1225 : : /*
1226 : : * Swap the physical files of the target and transient tables, then
1227 : : * rebuild the target's indexes and throw away the transient table.
1228 : : */
1229 : 429 : finish_heap_swap(tableOid, OIDNewHeap, is_system_catalog,
1230 : : swap_toast_by_content, false, true,
1231 : : true, /* reindex */
1232 : : frozenXid, cutoffMulti,
1233 : : relpersistence);
1234 : : }
11030 scrappy@hub.org 1235 : 434 : }
1236 : :
1237 : :
1238 : : /*
1239 : : * Create the transient table that will be filled with new data during
1240 : : * CLUSTER, ALTER TABLE, and similar operations. The transient table
1241 : : * duplicates the logical structure of the OldHeap; but will have the
1242 : : * specified physical storage properties NewTableSpace, NewAccessMethod, and
1243 : : * relpersistence.
1244 : : *
1245 : : * After this, the caller should load the new heap with transferred/modified
1246 : : * data, then call finish_heap_swap to complete the operation.
1247 : : */
1248 : : Oid
1880 michael@paquier.xyz 1249 : 1598 : make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
1250 : : char relpersistence, LOCKMODE lockmode)
1251 : : {
1252 : : TupleDesc OldHeapDesc;
1253 : : char NewHeapName[NAMEDATALEN];
1254 : : Oid OIDNewHeap;
1255 : : Oid toastid;
1256 : : Relation OldHeap;
1257 : : HeapTuple tuple;
1258 : : Datum reloptions;
1259 : : bool isNull;
1260 : : Oid namespaceid;
1261 : :
2799 andres@anarazel.de 1262 : 1598 : OldHeap = table_open(OIDOldHeap, lockmode);
10246 bruce@momjian.us 1263 : 1598 : OldHeapDesc = RelationGetDescr(OldHeap);
1264 : :
1265 : : /*
1266 : : * Note that the NewHeap will not receive any of the defaults or
1267 : : * constraints associated with the OldHeap; we don't need 'em, and there's
1268 : : * no reason to spend cycles inserting them into the catalogs only to
1269 : : * delete them.
1270 : : */
1271 : :
1272 : : /*
1273 : : * But we do want to use reloptions of the old heap for new heap.
1274 : : */
6062 rhaas@postgresql.org 1275 : 1598 : tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(OIDOldHeap));
7384 tgl@sss.pgh.pa.us 1276 [ - + ]: 1598 : if (!HeapTupleIsValid(tuple))
7384 tgl@sss.pgh.pa.us 1277 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap);
7384 tgl@sss.pgh.pa.us 1278 :CBC 1598 : reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions,
1279 : : &isNull);
1280 [ + + ]: 1598 : if (isNull)
1281 : 1481 : reloptions = (Datum) 0;
1282 : :
4412 alvherre@alvh.no-ip. 1283 [ + + ]: 1598 : if (relpersistence == RELPERSISTENCE_TEMP)
4814 kgrittn@postgresql.o 1284 : 98 : namespaceid = LookupCreationNamespace("pg_temp");
1285 : : else
1286 : 1500 : namespaceid = RelationGetNamespace(OldHeap);
1287 : :
1288 : : /*
1289 : : * Create the new heap, using a temporary name in the same namespace as
1290 : : * the existing table. NOTE: there is some risk of collision with user
1291 : : * relnames. Working around this seems more trouble than it's worth; in
1292 : : * particular, we can't create the new heap in a different namespace from
1293 : : * the old, or we will have problems with the TEMP status of temp tables.
1294 : : *
1295 : : * Note: the new heap is not a shared relation, even if we are rebuilding
1296 : : * a shared rel. However, we do make the new heap mapped if the source is
1297 : : * mapped. This simplifies swap_relation_files, and is absolutely
1298 : : * necessary for rebuilding pg_class, for reasons explained there.
1299 : : */
6072 tgl@sss.pgh.pa.us 1300 : 1598 : snprintf(NewHeapName, sizeof(NewHeapName), "pg_temp_%u", OIDOldHeap);
1301 : :
1302 : 1598 : OIDNewHeap = heap_create_with_catalog(NewHeapName,
1303 : : namespaceid,
1304 : : NewTableSpace,
1305 : : InvalidOid,
1306 : : InvalidOid,
1307 : : InvalidOid,
7695 1308 : 1598 : OldHeap->rd_rel->relowner,
1309 : : NewAccessMethod,
1310 : : OldHeapDesc,
1311 : : NIL,
1312 : : RELKIND_RELATION,
1313 : : relpersistence,
1314 : : false,
6069 1315 [ + + + - : 1598 : RelationIsMapped(OldHeap),
+ - + + +
- + + ]
1316 : : ONCOMMIT_NOOP,
1317 : : reloptions,
1318 : : false,
1319 : : true,
1320 : : true,
1321 : : OIDOldHeap,
4219 alvherre@alvh.no-ip. 1322 : 1598 : NULL);
5901 rhaas@postgresql.org 1323 [ - + ]: 1598 : Assert(OIDNewHeap != InvalidOid);
1324 : :
7385 bruce@momjian.us 1325 : 1598 : ReleaseSysCache(tuple);
1326 : :
1327 : : /*
1328 : : * Advance command counter so that the newly-created relation's catalog
1329 : : * tuples will be visible to table_open.
1330 : : */
9393 tgl@sss.pgh.pa.us 1331 : 1598 : CommandCounterIncrement();
1332 : :
1333 : : /*
1334 : : * If necessary, create a TOAST table for the new relation.
1335 : : *
1336 : : * If the relation doesn't have a TOAST table already, we can't need one
1337 : : * for the new relation. The other way around is possible though: if some
1338 : : * wide columns have been dropped, NewHeapCreateToastTable can decide that
1339 : : * no TOAST table is needed for the new table.
1340 : : *
1341 : : * Note that NewHeapCreateToastTable ends with CommandCounterIncrement, so
1342 : : * that the TOAST table will be visible for insertion.
1343 : : */
6439 alvherre@alvh.no-ip. 1344 : 1598 : toastid = OldHeap->rd_rel->reltoastrelid;
1345 [ + + ]: 1598 : if (OidIsValid(toastid))
1346 : : {
1347 : : /* keep the existing toast table's reloptions, if any */
6062 rhaas@postgresql.org 1348 : 595 : tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(toastid));
6439 alvherre@alvh.no-ip. 1349 [ - + ]: 595 : if (!HeapTupleIsValid(tuple))
6439 alvherre@alvh.no-ip. 1350 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for relation %u", toastid);
6439 alvherre@alvh.no-ip. 1351 :CBC 595 : reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions,
1352 : : &isNull);
1353 [ + - ]: 595 : if (isNull)
1354 : 595 : reloptions = (Datum) 0;
1355 : :
1852 akapila@postgresql.o 1356 : 595 : NewHeapCreateToastTable(OIDNewHeap, reloptions, lockmode, toastid);
1357 : :
6439 alvherre@alvh.no-ip. 1358 : 595 : ReleaseSysCache(tuple);
1359 : : }
1360 : :
2799 andres@anarazel.de 1361 : 1598 : table_close(OldHeap, NoLock);
1362 : :
9447 tgl@sss.pgh.pa.us 1363 : 1598 : return OIDNewHeap;
1364 : : }
1365 : :
1366 : : /*
1367 : : * Do the physical copying of table data.
1368 : : *
1369 : : * 'snapshot' and 'decoding_ctx': see table_relation_copy_for_cluster(). Pass
1370 : : * iff concurrent processing is required.
1371 : : *
1372 : : * There are three output parameters:
1373 : : * *pSwapToastByContent is set true if toast tables must be swapped by content.
1374 : : * *pFreezeXid receives the TransactionId used as freeze cutoff point.
1375 : : * *pCutoffMulti receives the MultiXactId used as a cutoff point.
1376 : : */
1377 : : static void
167 alvherre@kurilemu.de 1378 : 438 : copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
1379 : : Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
1380 : : TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
1381 : : {
1382 : : Relation relRelation;
1383 : : HeapTuple reltup;
1384 : : Form_pg_class relform;
1385 : : TupleDesc oldTupDesc PG_USED_FOR_ASSERTS_ONLY;
1386 : : TupleDesc newTupDesc PG_USED_FOR_ASSERTS_ONLY;
1387 : : VacuumParams params;
1388 : : struct VacuumCutoffs cutoffs;
1389 : : bool use_sort;
5827 tgl@sss.pgh.pa.us 1390 : 438 : double num_tuples = 0,
1391 : 438 : tups_vacuumed = 0,
1392 : 438 : tups_recently_dead = 0;
1393 : : BlockNumber num_pages;
1394 [ + + ]: 438 : int elevel = verbose ? INFO : DEBUG2;
1395 : : PGRUsage ru0;
1396 : : char *nspname;
167 alvherre@kurilemu.de 1397 : 438 : bool concurrent = snapshot != NULL;
1398 : : LOCKMODE lmode;
1399 : :
1400 : 438 : lmode = RepackLockLevel(concurrent);
1401 : :
5827 tgl@sss.pgh.pa.us 1402 : 438 : pg_rusage_init(&ru0);
1403 : :
1404 : : /* Store a copy of the namespace name for logging purposes */
1861 dgustafsson@postgres 1405 : 438 : nspname = get_namespace_name(RelationGetNamespace(OldHeap));
1406 : :
1407 : : /*
1408 : : * Their tuple descriptors should be exactly alike, but here we only need
1409 : : * assume that they have the same number of columns.
1410 : : */
7896 tgl@sss.pgh.pa.us 1411 : 438 : oldTupDesc = RelationGetDescr(OldHeap);
1412 : 438 : newTupDesc = RelationGetDescr(NewHeap);
1413 [ - + ]: 438 : Assert(newTupDesc->natts == oldTupDesc->natts);
1414 : :
1415 : : /*
1416 : : * If the OldHeap has a toast table, get lock on the toast table to keep
1417 : : * it from being vacuumed. This is needed because autovacuum processes
1418 : : * toast tables independently of their main tables, with no lock on the
1419 : : * latter. If an autovacuum were to start on the toast table after we
1420 : : * compute our OldestXmin below, it would use a later OldestXmin, and then
1421 : : * possibly remove as DEAD toast tuples belonging to main tuples we think
1422 : : * are only RECENTLY_DEAD. Then we'd fail while trying to copy those
1423 : : * tuples.
1424 : : *
1425 : : * We don't need to open the toast relation here, just lock it. The lock
1426 : : * will be held till end of transaction.
1427 : : */
5621 1428 [ + + ]: 438 : if (OldHeap->rd_rel->reltoastrelid)
167 alvherre@kurilemu.de 1429 : 167 : LockRelationOid(OldHeap->rd_rel->reltoastrelid, lmode);
1430 : :
1431 : : /*
1432 : : * If both tables have TOAST tables, perform toast swap by content. It is
1433 : : * possible that the old table has a toast table but the new one doesn't,
1434 : : * if toastable columns have been dropped. In that case we have to do
1435 : : * swap by links. This is okay because swap by content is only essential
1436 : : * for system catalogs, and we don't support schema changes for them.
1437 : : */
1438 [ + + + - ]: 438 : if (OldHeap->rd_rel->reltoastrelid && NewHeap->rd_rel->reltoastrelid &&
1439 [ + + ]: 167 : !concurrent)
1440 : : {
6072 tgl@sss.pgh.pa.us 1441 : 162 : *pSwapToastByContent = true;
1442 : :
1443 : : /*
1444 : : * When doing swap by content, any toast pointers written into NewHeap
1445 : : * must use the old toast table's OID, because that's where the toast
1446 : : * data will eventually be found. Set this up by setting rd_toastoid.
1447 : : * This also tells toast_save_datum() to preserve the toast value
1448 : : * OIDs, which we want so as not to invalidate toast pointers in
1449 : : * system catalog caches, and to avoid making multiple copies of a
1450 : : * single toast value.
1451 : : *
1452 : : * Note that we must hold NewHeap open until we are done writing data,
1453 : : * since the relcache will not guarantee to remember this setting once
1454 : : * the relation is closed. Also, this technique depends on the fact
1455 : : * that no one will try to read from the NewHeap until after we've
1456 : : * finished writing it and swapping the rels --- otherwise they could
1457 : : * follow the toast pointers to the wrong place. (It would actually
1458 : : * work for values copied over from the old toast table, but not for
1459 : : * any values that we toast which were previously not toasted.)
1460 : : *
1461 : : * This would not work with CONCURRENTLY because we may need to delete
1462 : : * TOASTed tuples from the new heap. With this hack, we'd delete them
1463 : : * from the old heap.
1464 : : */
1465 : 162 : NewHeap->rd_toastoid = OldHeap->rd_rel->reltoastrelid;
1466 : : }
1467 : : else
1468 : 276 : *pSwapToastByContent = false;
1469 : :
1470 : : /*
1471 : : * Compute xids used to freeze and weed out dead tuples and multixacts.
1472 : : * Since we're going to rewrite the whole table anyway, there's no reason
1473 : : * not to be aggressive about this.
1474 : : */
1397 pg@bowt.ie 1475 : 438 : memset(¶ms, 0, sizeof(VacuumParams));
173 nathan@postgresql.or 1476 : 438 : vacuum_get_cutoffs(OldHeap, ¶ms, &cutoffs);
1477 : :
1478 : : /*
1479 : : * FreezeXid will become the table's new relfrozenxid, and that mustn't go
1480 : : * backwards, so take the max.
1481 : : */
1482 : : {
874 noah@leadboat.com 1483 : 438 : TransactionId relfrozenxid = OldHeap->rd_rel->relfrozenxid;
1484 : :
1485 [ + - + + ]: 876 : if (TransactionIdIsValid(relfrozenxid) &&
1486 : 438 : TransactionIdPrecedes(cutoffs.FreezeLimit, relfrozenxid))
1487 : 41 : cutoffs.FreezeLimit = relfrozenxid;
1488 : : }
1489 : :
1490 : : /*
1491 : : * MultiXactCutoff, similarly, shouldn't go backwards either.
1492 : : */
1493 : : {
1494 : 438 : MultiXactId relminmxid = OldHeap->rd_rel->relminmxid;
1495 : :
1496 [ + - - + ]: 876 : if (MultiXactIdIsValid(relminmxid) &&
1497 : 438 : MultiXactIdPrecedes(cutoffs.MultiXactCutoff, relminmxid))
874 noah@leadboat.com 1498 :UBC 0 : cutoffs.MultiXactCutoff = relminmxid;
1499 : : }
1500 : :
1501 : : /*
1502 : : * Decide whether to use an indexscan or seqscan-and-optional-sort to scan
1503 : : * the OldHeap. If the index is a btree, ask the planner to choose via
1504 : : * normal path cost comparison.
1505 : : *
1506 : : * The underlying tuplesort.c code doesn't support AMs other than btree,
1507 : : * so we must always use a normal indexscan if a non-btree index is
1508 : : * specified -- or an unsorted seqscan if no index is given.
1509 : : */
5827 tgl@sss.pgh.pa.us 1510 [ + + + + ]:CBC 438 : if (OldIndex != NULL && OldIndex->rd_rel->relam == BTREE_AM_OID)
618 alvherre@alvh.no-ip. 1511 : 145 : use_sort = plan_cluster_use_sort(RelationGetRelid(OldHeap),
1512 : : RelationGetRelid(OldIndex));
1513 : : else
5827 tgl@sss.pgh.pa.us 1514 : 293 : use_sort = false;
1515 : :
1516 : : /* Log what we're doing */
2733 andres@anarazel.de 1517 [ + + + + ]: 438 : if (OldIndex != NULL && !use_sort)
5827 tgl@sss.pgh.pa.us 1518 [ - + ]: 62 : ereport(elevel,
1519 : : errmsg("repacking \"%s.%s\" using index scan on \"%s\"",
1520 : : nspname,
1521 : : RelationGetRelationName(OldHeap),
1522 : : RelationGetRelationName(OldIndex)));
2733 andres@anarazel.de 1523 [ + + ]: 376 : else if (use_sort)
5827 tgl@sss.pgh.pa.us 1524 [ - + ]: 85 : ereport(elevel,
1525 : : errmsg("repacking \"%s.%s\" using sequential scan and sort",
1526 : : nspname,
1527 : : RelationGetRelationName(OldHeap)));
1528 : : else
1529 [ + + ]: 291 : ereport(elevel,
1530 : : errmsg("repacking \"%s.%s\" in physical order",
1531 : : nspname,
1532 : : RelationGetRelationName(OldHeap)));
1533 : :
1534 : : /*
1535 : : * Hand off the actual copying to AM specific function, the generic code
1536 : : * cannot know how to deal with visibility across AMs. Note that this
1537 : : * routine is allowed to set FreezeXid / MultiXactCutoff to different
1538 : : * values (e.g. because the AM doesn't use freezing).
1539 : : */
2733 andres@anarazel.de 1540 : 438 : table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
1541 : : cutoffs.OldestXmin, snapshot,
1542 : : &cutoffs.FreezeLimit,
1543 : : &cutoffs.MultiXactCutoff,
1544 : : &num_tuples, &tups_vacuumed,
1545 : : &tups_recently_dead);
1546 : :
1547 : : /* return selected values to caller, get set as relfrozenxid/minmxid */
1368 pg@bowt.ie 1548 : 438 : *pFreezeXid = cutoffs.FreezeLimit;
1549 : 438 : *pCutoffMulti = cutoffs.MultiXactCutoff;
1550 : :
1551 : : /*
1552 : : * Reset rd_toastoid just to be tidy --- it shouldn't be looked at again.
1553 : : * In the CONCURRENTLY case, we need to set it again before applying the
1554 : : * concurrent changes.
1555 : : */
6072 tgl@sss.pgh.pa.us 1556 : 438 : NewHeap->rd_toastoid = InvalidOid;
1557 : :
3189 teodor@sigaev.ru 1558 : 438 : num_pages = RelationGetNumberOfBlocks(NewHeap);
1559 : :
1560 : : /* Log what we did */
5827 tgl@sss.pgh.pa.us 1561 [ + + ]: 438 : ereport(elevel,
1562 : : (errmsg("\"%s.%s\": found %.0f removable, %.0f nonremovable row versions in %u pages",
1563 : : nspname,
1564 : : RelationGetRelationName(OldHeap),
1565 : : tups_vacuumed, num_tuples,
1566 : : RelationGetNumberOfBlocks(OldHeap)),
1567 : : errdetail("%.0f dead row versions cannot be removed yet.\n"
1568 : : "%s.",
1569 : : tups_recently_dead,
1570 : : pg_rusage_show(&ru0))));
1571 : :
1572 : : /* Update pg_class to reflect the correct values of pages and tuples. */
2799 andres@anarazel.de 1573 : 438 : relRelation = table_open(RelationRelationId, RowExclusiveLock);
1574 : :
618 alvherre@alvh.no-ip. 1575 : 438 : reltup = SearchSysCacheCopy1(RELOID,
1576 : : ObjectIdGetDatum(RelationGetRelid(NewHeap)));
3189 teodor@sigaev.ru 1577 [ - + ]: 438 : if (!HeapTupleIsValid(reltup))
618 alvherre@alvh.no-ip. 1578 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for relation %u",
1579 : : RelationGetRelid(NewHeap));
3189 teodor@sigaev.ru 1580 :CBC 438 : relform = (Form_pg_class) GETSTRUCT(reltup);
1581 : :
1582 : 438 : relform->relpages = num_pages;
1583 : 438 : relform->reltuples = num_tuples;
1584 : :
1585 : : /* Don't update the stats for pg_class. See swap_relation_files. */
618 alvherre@alvh.no-ip. 1586 [ + + ]: 438 : if (RelationGetRelid(OldHeap) != RelationRelationId)
3189 teodor@sigaev.ru 1587 : 415 : CatalogTupleUpdate(relRelation, &reltup->t_self, reltup);
1588 : : else
1589 : 23 : CacheInvalidateRelcacheByTuple(reltup);
1590 : :
1591 : : /* Clean up. */
1592 : 438 : heap_freetuple(reltup);
2799 andres@anarazel.de 1593 : 438 : table_close(relRelation, RowExclusiveLock);
1594 : :
1595 : : /* Make the update visible */
3189 teodor@sigaev.ru 1596 : 438 : CommandCounterIncrement();
11030 scrappy@hub.org 1597 : 438 : }
1598 : :
1599 : : /*
1600 : : * Swap the physical files of two given relations.
1601 : : *
1602 : : * We swap the physical identity (reltablespace, relfilenumber) while keeping
1603 : : * the same logical identities of the two relations. relpersistence is also
1604 : : * swapped, which is critical since it determines where buffers live for each
1605 : : * relation.
1606 : : *
1607 : : * We can swap associated TOAST data in either of two ways: recursively swap
1608 : : * the physical content of the toast tables (and their indexes), or swap the
1609 : : * TOAST links in the given relations' pg_class entries. The former is needed
1610 : : * to manage rewrites of shared catalogs (where we cannot change the pg_class
1611 : : * links) while the latter is the only way to handle cases in which a toast
1612 : : * table is added or removed altogether.
1613 : : *
1614 : : * Additionally, the first relation is marked with relfrozenxid set to
1615 : : * frozenXid. It seems a bit ugly to have this here, but the caller would
1616 : : * have to do it anyway, so having it here saves a heap_update. Note: in
1617 : : * the swap-toast-links case, we assume we don't need to change the toast
1618 : : * table's relfrozenxid: the new version of the toast table should already
1619 : : * have relfrozenxid set to RecentXmin, which is good enough.
1620 : : *
1621 : : * Lastly, if r2 and its toast table and toast index (if any) are mapped,
1622 : : * their OIDs are emitted into mapped_tables[]. This is hacky but beats
1623 : : * having to look the information up again later in finish_heap_swap.
1624 : : */
1625 : : static void
6069 tgl@sss.pgh.pa.us 1626 : 1807 : swap_relation_files(Oid r1, Oid r2, bool target_is_pg_class,
1627 : : bool swap_toast_by_content,
1628 : : bool is_internal,
1629 : : TransactionId frozenXid,
1630 : : MultiXactId cutoffMulti,
1631 : : Oid *mapped_tables)
1632 : : {
1633 : : Relation relRelation;
1634 : : HeapTuple reltup1,
1635 : : reltup2;
1636 : : Form_pg_class relform1,
1637 : : relform2;
1638 : : RelFileNumber relfilenumber1,
1639 : : relfilenumber2;
1640 : : RelFileNumber swaptemp;
1641 : : char swptmpchr;
1642 : : Oid relam1,
1643 : : relam2;
1644 : :
1645 : : /* We need writable copies of both pg_class tuples. */
2799 andres@anarazel.de 1646 : 1807 : relRelation = table_open(RelationRelationId, RowExclusiveLock);
1647 : :
6062 rhaas@postgresql.org 1648 : 1807 : reltup1 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r1));
8806 tgl@sss.pgh.pa.us 1649 [ - + ]: 1807 : if (!HeapTupleIsValid(reltup1))
8463 tgl@sss.pgh.pa.us 1650 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for relation %u", r1);
8806 tgl@sss.pgh.pa.us 1651 :CBC 1807 : relform1 = (Form_pg_class) GETSTRUCT(reltup1);
1652 : :
6062 rhaas@postgresql.org 1653 : 1807 : reltup2 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r2));
8806 tgl@sss.pgh.pa.us 1654 [ - + ]: 1807 : if (!HeapTupleIsValid(reltup2))
8463 tgl@sss.pgh.pa.us 1655 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for relation %u", r2);
8806 tgl@sss.pgh.pa.us 1656 :CBC 1807 : relform2 = (Form_pg_class) GETSTRUCT(reltup2);
1657 : :
1537 rhaas@postgresql.org 1658 : 1807 : relfilenumber1 = relform1->relfilenode;
1659 : 1807 : relfilenumber2 = relform2->relfilenode;
1178 michael@paquier.xyz 1660 : 1807 : relam1 = relform1->relam;
1661 : 1807 : relam2 = relform2->relam;
1662 : :
1537 rhaas@postgresql.org 1663 [ + + + - ]: 1807 : if (RelFileNumberIsValid(relfilenumber1) &&
1664 : : RelFileNumberIsValid(relfilenumber2))
1665 : : {
1666 : : /*
1667 : : * Normal non-mapped relations: swap relfilenumbers, reltablespaces,
1668 : : * relpersistence
1669 : : */
6069 tgl@sss.pgh.pa.us 1670 [ - + ]: 1718 : Assert(!target_is_pg_class);
1671 : :
1672 : 1718 : swaptemp = relform1->relfilenode;
1673 : 1718 : relform1->relfilenode = relform2->relfilenode;
1674 : 1718 : relform2->relfilenode = swaptemp;
1675 : :
1676 : 1718 : swaptemp = relform1->reltablespace;
1677 : 1718 : relform1->reltablespace = relform2->reltablespace;
1678 : 1718 : relform2->reltablespace = swaptemp;
1679 : :
1880 michael@paquier.xyz 1680 : 1718 : swaptemp = relform1->relam;
1681 : 1718 : relform1->relam = relform2->relam;
1682 : 1718 : relform2->relam = swaptemp;
1683 : :
4412 alvherre@alvh.no-ip. 1684 : 1718 : swptmpchr = relform1->relpersistence;
1685 : 1718 : relform1->relpersistence = relform2->relpersistence;
1686 : 1718 : relform2->relpersistence = swptmpchr;
1687 : :
1688 : : /* Also swap toast links, if we're swapping by links */
6069 tgl@sss.pgh.pa.us 1689 [ + + ]: 1718 : if (!swap_toast_by_content)
1690 : : {
1691 : 1292 : swaptemp = relform1->reltoastrelid;
1692 : 1292 : relform1->reltoastrelid = relform2->reltoastrelid;
1693 : 1292 : relform2->reltoastrelid = swaptemp;
1694 : : }
1695 : : }
1696 : : else
1697 : : {
1698 : : /*
1699 : : * Mapped-relation case. Here we have to swap the relation mappings
1700 : : * instead of modifying the pg_class columns. Both must be mapped.
1701 : : */
1537 rhaas@postgresql.org 1702 [ + - - + ]: 89 : if (RelFileNumberIsValid(relfilenumber1) ||
1703 : : RelFileNumberIsValid(relfilenumber2))
6069 tgl@sss.pgh.pa.us 1704 [ # # ]:UBC 0 : elog(ERROR, "cannot swap mapped relation \"%s\" with non-mapped relation",
1705 : : NameStr(relform1->relname));
1706 : :
1707 : : /*
1708 : : * We can't change the tablespace nor persistence of a mapped rel, and
1709 : : * we can't handle toast link swapping for one either, because we must
1710 : : * not apply any critical changes to its pg_class row. These cases
1711 : : * should be prevented by upstream permissions tests, so these checks
1712 : : * are non-user-facing emergency backstop.
1713 : : */
6069 tgl@sss.pgh.pa.us 1714 [ - + ]:CBC 89 : if (relform1->reltablespace != relform2->reltablespace)
6069 tgl@sss.pgh.pa.us 1715 [ # # ]:UBC 0 : elog(ERROR, "cannot change tablespace of mapped relation \"%s\"",
1716 : : NameStr(relform1->relname));
4412 alvherre@alvh.no-ip. 1717 [ - + ]:CBC 89 : if (relform1->relpersistence != relform2->relpersistence)
4412 alvherre@alvh.no-ip. 1718 [ # # ]:UBC 0 : elog(ERROR, "cannot change persistence of mapped relation \"%s\"",
1719 : : NameStr(relform1->relname));
1880 michael@paquier.xyz 1720 [ - + ]:CBC 89 : if (relform1->relam != relform2->relam)
1880 michael@paquier.xyz 1721 [ # # ]:UBC 0 : elog(ERROR, "cannot change access method of mapped relation \"%s\"",
1722 : : NameStr(relform1->relname));
6069 tgl@sss.pgh.pa.us 1723 [ + + ]:CBC 89 : if (!swap_toast_by_content &&
1724 [ + - - + ]: 29 : (relform1->reltoastrelid || relform2->reltoastrelid))
6069 tgl@sss.pgh.pa.us 1725 [ # # ]:UBC 0 : elog(ERROR, "cannot swap toast by links for mapped relation \"%s\"",
1726 : : NameStr(relform1->relname));
1727 : :
1728 : : /*
1729 : : * Fetch the mappings --- shouldn't fail, but be paranoid
1730 : : */
1537 rhaas@postgresql.org 1731 :CBC 89 : relfilenumber1 = RelationMapOidToFilenumber(r1, relform1->relisshared);
1732 [ - + ]: 89 : if (!RelFileNumberIsValid(relfilenumber1))
6069 tgl@sss.pgh.pa.us 1733 [ # # ]:UBC 0 : elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
1734 : : NameStr(relform1->relname), r1);
1537 rhaas@postgresql.org 1735 :CBC 89 : relfilenumber2 = RelationMapOidToFilenumber(r2, relform2->relisshared);
1736 [ - + ]: 89 : if (!RelFileNumberIsValid(relfilenumber2))
6069 tgl@sss.pgh.pa.us 1737 [ # # ]:UBC 0 : elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
1738 : : NameStr(relform2->relname), r2);
1739 : :
1740 : : /*
1741 : : * Send replacement mappings to relmapper. Note these won't actually
1742 : : * take effect until CommandCounterIncrement.
1743 : : */
1537 rhaas@postgresql.org 1744 :CBC 89 : RelationMapUpdateMap(r1, relfilenumber2, relform1->relisshared, false);
1745 : 89 : RelationMapUpdateMap(r2, relfilenumber1, relform2->relisshared, false);
1746 : :
1747 : : /* Pass OIDs of mapped r2 tables back to caller */
6069 tgl@sss.pgh.pa.us 1748 : 89 : *mapped_tables++ = r2;
1749 : : }
1750 : :
1751 : : /*
1752 : : * Recognize that rel1's relfilenumber (swapped from rel2) is new in this
1753 : : * subtransaction. The rel2 storage (swapped from rel1) may or may not be
1754 : : * new.
1755 : : */
1756 : : {
1757 : : Relation rel1,
1758 : : rel2;
1759 : :
2360 noah@leadboat.com 1760 : 1807 : rel1 = relation_open(r1, NoLock);
1761 : 1807 : rel2 = relation_open(r2, NoLock);
1762 : 1807 : rel2->rd_createSubid = rel1->rd_createSubid;
1537 rhaas@postgresql.org 1763 : 1807 : rel2->rd_newRelfilelocatorSubid = rel1->rd_newRelfilelocatorSubid;
1764 : 1807 : rel2->rd_firstRelfilelocatorSubid = rel1->rd_firstRelfilelocatorSubid;
1765 : 1807 : RelationAssumeNewRelfilelocator(rel1);
2360 noah@leadboat.com 1766 : 1807 : relation_close(rel1, NoLock);
1767 : 1807 : relation_close(rel2, NoLock);
1768 : : }
1769 : :
1770 : : /*
1771 : : * In the case of a shared catalog, these next few steps will only affect
1772 : : * our own database's pg_class row; but that's okay, because they are all
1773 : : * noncritical updates. That's also an important fact for the case of a
1774 : : * mapped catalog, because it's possible that we'll commit the map change
1775 : : * and then fail to commit the pg_class update.
1776 : : */
1777 : :
1778 : : /* set rel1's frozen Xid and minimum MultiXid */
6072 tgl@sss.pgh.pa.us 1779 [ + + ]: 1807 : if (relform1->relkind != RELKIND_INDEX)
1780 : : {
2707 andres@anarazel.de 1781 [ + - - + ]: 1635 : Assert(!TransactionIdIsValid(frozenXid) ||
1782 : : TransactionIdIsNormal(frozenXid));
6072 tgl@sss.pgh.pa.us 1783 : 1635 : relform1->relfrozenxid = frozenXid;
4752 alvherre@alvh.no-ip. 1784 : 1635 : relform1->relminmxid = cutoffMulti;
1785 : : }
1786 : :
1787 : : /* swap size statistics too, since new rel has freshly-updated stats */
1788 : : {
1789 : : int32 swap_pages;
1790 : : float4 swap_tuples;
1791 : : int32 swap_allvisible;
1792 : : int32 swap_allfrozen;
1793 : :
8723 tgl@sss.pgh.pa.us 1794 : 1807 : swap_pages = relform1->relpages;
1795 : 1807 : relform1->relpages = relform2->relpages;
1796 : 1807 : relform2->relpages = swap_pages;
1797 : :
1798 : 1807 : swap_tuples = relform1->reltuples;
1799 : 1807 : relform1->reltuples = relform2->reltuples;
1800 : 1807 : relform2->reltuples = swap_tuples;
1801 : :
5455 1802 : 1807 : swap_allvisible = relform1->relallvisible;
1803 : 1807 : relform1->relallvisible = relform2->relallvisible;
1804 : 1807 : relform2->relallvisible = swap_allvisible;
1805 : :
566 melanieplageman@gmai 1806 : 1807 : swap_allfrozen = relform1->relallfrozen;
1807 : 1807 : relform1->relallfrozen = relform2->relallfrozen;
1808 : 1807 : relform2->relallfrozen = swap_allfrozen;
1809 : : }
1810 : :
1811 : : /*
1812 : : * Update the tuples in pg_class --- unless the target relation of the
1813 : : * swap is pg_class itself. In that case, there is zero point in making
1814 : : * changes because we'd be updating the old data that we're about to throw
1815 : : * away. Because the real work being done here for a mapped relation is
1816 : : * just to change the relation map settings, it's all right to not update
1817 : : * the pg_class rows in this case. The most important changes will instead
1818 : : * performed later, in finish_heap_swap() itself.
1819 : : */
6069 tgl@sss.pgh.pa.us 1820 [ + + ]: 1807 : if (!target_is_pg_class)
1821 : : {
1822 : : CatalogIndexState indstate;
1823 : :
1824 : 1784 : indstate = CatalogOpenIndexes(relRelation);
3518 1825 : 1784 : CatalogTupleUpdateWithInfo(relRelation, &reltup1->t_self, reltup1,
1826 : : indstate);
1827 : 1784 : CatalogTupleUpdateWithInfo(relRelation, &reltup2->t_self, reltup2,
1828 : : indstate);
6069 1829 : 1784 : CatalogCloseIndexes(indstate);
1830 : : }
1831 : : else
1832 : : {
1833 : : /* no update ... but we do still need relcache inval */
1834 : 23 : CacheInvalidateRelcacheByTuple(reltup1);
1835 : 23 : CacheInvalidateRelcacheByTuple(reltup2);
1836 : : }
1837 : :
1838 : : /*
1839 : : * Now that pg_class has been updated with its relevant information for
1840 : : * the swap, update the dependency of the relations to point to their new
1841 : : * table AM, if it has changed.
1842 : : */
1178 michael@paquier.xyz 1843 [ + + ]: 1807 : if (relam1 != relam2)
1844 : : {
1845 [ - + ]: 24 : if (changeDependencyFor(RelationRelationId,
1846 : : r1,
1847 : : AccessMethodRelationId,
1848 : : relam1,
1849 : : relam2) != 1)
1168 michael@paquier.xyz 1850 [ # # ]:UBC 0 : elog(ERROR, "could not change access method dependency for relation \"%s.%s\"",
1851 : : get_namespace_name(get_rel_namespace(r1)),
1852 : : get_rel_name(r1));
1178 michael@paquier.xyz 1853 [ - + ]:CBC 24 : if (changeDependencyFor(RelationRelationId,
1854 : : r2,
1855 : : AccessMethodRelationId,
1856 : : relam2,
1857 : : relam1) != 1)
1168 michael@paquier.xyz 1858 [ # # ]:UBC 0 : elog(ERROR, "could not change access method dependency for relation \"%s.%s\"",
1859 : : get_namespace_name(get_rel_namespace(r2)),
1860 : : get_rel_name(r2));
1861 : : }
1862 : :
1863 : : /*
1864 : : * Post alter hook for modified relations. The change to r2 is always
1865 : : * internal, but r1 depends on the invocation context.
1866 : : */
4935 rhaas@postgresql.org 1867 [ - + ]:CBC 1807 : InvokeObjectPostAlterHookArg(RelationRelationId, r1, 0,
1868 : : InvalidOid, is_internal);
1869 [ - + ]: 1807 : InvokeObjectPostAlterHookArg(RelationRelationId, r2, 0,
1870 : : InvalidOid, true);
1871 : :
1872 : : /*
1873 : : * If we have toast tables associated with the relations being swapped,
1874 : : * deal with them too.
1875 : : */
8806 tgl@sss.pgh.pa.us 1876 [ + + + + ]: 1807 : if (relform1->reltoastrelid || relform2->reltoastrelid)
1877 : : {
6072 1878 [ + + ]: 566 : if (swap_toast_by_content)
1879 : : {
1880 [ + - + - ]: 162 : if (relform1->reltoastrelid && relform2->reltoastrelid)
1881 : : {
1882 : : /* Recursively swap the contents of the toast tables */
1883 : 162 : swap_relation_files(relform1->reltoastrelid,
1884 : : relform2->reltoastrelid,
1885 : : target_is_pg_class,
1886 : : swap_toast_by_content,
1887 : : is_internal,
1888 : : frozenXid,
1889 : : cutoffMulti,
1890 : : mapped_tables);
1891 : : }
1892 : : else
1893 : : {
1894 : : /* caller messed up */
6072 tgl@sss.pgh.pa.us 1895 [ # # ]:UBC 0 : elog(ERROR, "cannot swap toast files by content when there's only one");
1896 : : }
1897 : : }
1898 : : else
1899 : : {
1900 : : /*
1901 : : * We swapped the ownership links, so we need to change dependency
1902 : : * data to match.
1903 : : *
1904 : : * NOTE: it is possible that only one table has a toast table.
1905 : : *
1906 : : * NOTE: at present, a TOAST table's only dependency is the one on
1907 : : * its owning table. If more are ever created, we'd need to use
1908 : : * something more selective than deleteDependencyRecordsFor() to
1909 : : * get rid of just the link we want.
1910 : : */
1911 : : ObjectAddress baseobject,
1912 : : toastobject;
1913 : : long count;
1914 : :
1915 : : /*
1916 : : * We disallow this case for system catalogs, to avoid the
1917 : : * possibility that the catalog we're rebuilding is one of the
1918 : : * ones the dependency changes would change. It's too late to be
1919 : : * making any data changes to the target catalog.
1920 : : */
4679 rhaas@postgresql.org 1921 [ - + ]:CBC 404 : if (IsSystemClass(r1, relform1))
6069 tgl@sss.pgh.pa.us 1922 [ # # ]:UBC 0 : elog(ERROR, "cannot swap toast files by links for system catalogs");
1923 : :
1924 : : /* Delete old dependencies */
6072 tgl@sss.pgh.pa.us 1925 [ + + ]:CBC 404 : if (relform1->reltoastrelid)
1926 : : {
1927 : 383 : count = deleteDependencyRecordsFor(RelationRelationId,
1928 : : relform1->reltoastrelid,
1929 : : false);
1930 [ - + ]: 383 : if (count != 1)
6072 tgl@sss.pgh.pa.us 1931 [ # # ]:UBC 0 : elog(ERROR, "expected one dependency record for TOAST table, found %ld",
1932 : : count);
1933 : : }
6072 tgl@sss.pgh.pa.us 1934 [ + - ]:CBC 404 : if (relform2->reltoastrelid)
1935 : : {
1936 : 404 : count = deleteDependencyRecordsFor(RelationRelationId,
1937 : : relform2->reltoastrelid,
1938 : : false);
1939 [ - + ]: 404 : if (count != 1)
6072 tgl@sss.pgh.pa.us 1940 [ # # ]:UBC 0 : elog(ERROR, "expected one dependency record for TOAST table, found %ld",
1941 : : count);
1942 : : }
1943 : :
1944 : : /* Register new dependencies */
6072 tgl@sss.pgh.pa.us 1945 :CBC 404 : baseobject.classId = RelationRelationId;
1946 : 404 : baseobject.objectSubId = 0;
1947 : 404 : toastobject.classId = RelationRelationId;
1948 : 404 : toastobject.objectSubId = 0;
1949 : :
1950 [ + + ]: 404 : if (relform1->reltoastrelid)
1951 : : {
1952 : 383 : baseobject.objectId = r1;
1953 : 383 : toastobject.objectId = relform1->reltoastrelid;
1954 : 383 : recordDependencyOn(&toastobject, &baseobject,
1955 : : DEPENDENCY_INTERNAL);
1956 : : }
1957 : :
1958 [ + - ]: 404 : if (relform2->reltoastrelid)
1959 : : {
1960 : 404 : baseobject.objectId = r2;
1961 : 404 : toastobject.objectId = relform2->reltoastrelid;
1962 : 404 : recordDependencyOn(&toastobject, &baseobject,
1963 : : DEPENDENCY_INTERNAL);
1964 : : }
1965 : : }
1966 : : }
1967 : :
1968 : : /*
1969 : : * If we're swapping two toast tables by content, do the same for their
1970 : : * valid index. The swap can actually be safely done only if the relations
1971 : : * have indexes.
1972 : : */
1973 [ + + ]: 1807 : if (swap_toast_by_content &&
4826 fujii@postgresql.org 1974 [ + + ]: 486 : relform1->relkind == RELKIND_TOASTVALUE &&
1975 [ + - ]: 162 : relform2->relkind == RELKIND_TOASTVALUE)
1976 : : {
1977 : : Oid toastIndex1,
1978 : : toastIndex2;
1979 : :
1980 : : /* Get valid index for each relation */
1981 : 162 : toastIndex1 = toast_get_valid_index(r1,
1982 : : AccessExclusiveLock);
1983 : 162 : toastIndex2 = toast_get_valid_index(r2,
1984 : : AccessExclusiveLock);
1985 : :
1986 : 162 : swap_relation_files(toastIndex1,
1987 : : toastIndex2,
1988 : : target_is_pg_class,
1989 : : swap_toast_by_content,
1990 : : is_internal,
1991 : : InvalidTransactionId,
1992 : : InvalidMultiXactId,
1993 : : mapped_tables);
1994 : : }
1995 : :
1996 : : /* Clean up. */
8806 tgl@sss.pgh.pa.us 1997 : 1807 : heap_freetuple(reltup1);
1998 : 1807 : heap_freetuple(reltup2);
1999 : :
2799 andres@anarazel.de 2000 : 1807 : table_close(relRelation, RowExclusiveLock);
8807 bruce@momjian.us 2001 : 1807 : }
2002 : :
2003 : : /*
2004 : : * Remove the transient table that was built by make_new_heap, and finish
2005 : : * cleaning up (including rebuilding all indexes on the old heap).
2006 : : */
2007 : : void
6069 tgl@sss.pgh.pa.us 2008 : 1473 : finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
2009 : : bool is_system_catalog,
2010 : : bool swap_toast_by_content,
2011 : : bool check_constraints,
2012 : : bool is_internal,
2013 : : bool reindex,
2014 : : TransactionId frozenXid,
2015 : : MultiXactId cutoffMulti,
2016 : : char newrelpersistence)
2017 : : {
2018 : : ObjectAddress object;
2019 : : Oid mapped_tables[4];
2020 : : int i;
2021 : :
2022 : : /* Report that we are now swapping relation files */
194 alvherre@kurilemu.de 2023 : 1473 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
2024 : : PROGRESS_REPACK_PHASE_SWAP_REL_FILES);
2025 : :
2026 : : /* Zero out possible results from swapped_relation_files */
6069 tgl@sss.pgh.pa.us 2027 : 1473 : memset(mapped_tables, 0, sizeof(mapped_tables));
2028 : :
2029 : : /*
2030 : : * Swap the contents of the heap relations (including any toast tables).
2031 : : * Also set old heap's relfrozenxid to frozenXid.
2032 : : */
2033 : 1473 : swap_relation_files(OIDOldHeap, OIDNewHeap,
2034 : : (OIDOldHeap == RelationRelationId),
2035 : : swap_toast_by_content, is_internal,
2036 : : frozenXid, cutoffMulti, mapped_tables);
2037 : :
2038 : : /*
2039 : : * If it's a system catalog, queue a sinval message to flush all catcaches
2040 : : * on the catalog when we reach CommandCounterIncrement.
2041 : : */
2042 [ + + ]: 1473 : if (is_system_catalog)
2043 : 116 : CacheInvalidateCatalog(OIDOldHeap);
2044 : :
167 alvherre@kurilemu.de 2045 [ + + ]: 1473 : if (reindex)
2046 : : {
2047 : : int reindex_flags;
2048 : 1464 : ReindexParams reindex_params = {0};
2049 : :
2050 : : /*
2051 : : * Rebuild each index on the relation (but not the toast table, which
2052 : : * is all-new at this point). It is important to do this before the
2053 : : * DROP step because if we are processing a system catalog that will
2054 : : * be used during DROP, we want to have its indexes available. There
2055 : : * is no advantage to the other order anyway because this is all
2056 : : * transactional, so no chance to reclaim disk space before commit. We
2057 : : * do not need a final CommandCounterIncrement() because
2058 : : * reindex_relation does it.
2059 : : *
2060 : : * Note: because index_build is called via reindex_relation, it will
2061 : : * never set indcheckxmin true for the indexes. This is OK even
2062 : : * though in some sense we are building new indexes rather than
2063 : : * rebuilding existing ones, because the new heap won't contain any
2064 : : * HOT chains at all, let alone broken ones, so it can't be necessary
2065 : : * to set indcheckxmin.
2066 : : */
2067 : 1464 : reindex_flags = REINDEX_REL_SUPPRESS_INDEX_USE;
2068 [ + + ]: 1464 : if (check_constraints)
2069 : 1035 : reindex_flags |= REINDEX_REL_CHECK_CONSTRAINTS;
2070 : :
2071 : : /*
2072 : : * Ensure that the indexes have the same persistence as the parent
2073 : : * relation.
2074 : : */
2075 [ + + ]: 1464 : if (newrelpersistence == RELPERSISTENCE_UNLOGGED)
2076 : 25 : reindex_flags |= REINDEX_REL_FORCE_INDEXES_UNLOGGED;
2077 [ + + ]: 1439 : else if (newrelpersistence == RELPERSISTENCE_PERMANENT)
2078 : 1386 : reindex_flags |= REINDEX_REL_FORCE_INDEXES_PERMANENT;
2079 : :
2080 : : /* Report that we are now reindexing relations */
2081 : 1464 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
2082 : : PROGRESS_REPACK_PHASE_REBUILD_INDEX);
2083 : :
2084 : 1464 : reindex_relation(NULL, OIDOldHeap, reindex_flags, &reindex_params);
2085 : : }
2086 : :
2087 : : /* Report that we are now doing clean up */
194 2088 : 1461 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
2089 : : PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
2090 : :
2091 : : /*
2092 : : * If the relation being rebuilt is pg_class, swap_relation_files()
2093 : : * couldn't update pg_class's own pg_class entry (check comments in
2094 : : * swap_relation_files()), thus relfrozenxid was not updated. That's
2095 : : * annoying because a potential reason for doing a VACUUM FULL is a
2096 : : * imminent or actual anti-wraparound shutdown. So, now that we can
2097 : : * access the new relation using its indices, update relfrozenxid.
2098 : : * pg_class doesn't have a toast relation, so we don't need to update the
2099 : : * corresponding toast relation. Not that there's little point moving all
2100 : : * relfrozenxid updates here since swap_relation_files() needs to write to
2101 : : * pg_class for non-mapped relations anyway.
2102 : : */
4583 rhaas@postgresql.org 2103 [ + + ]: 1461 : if (OIDOldHeap == RelationRelationId)
2104 : : {
2105 : : Relation relRelation;
2106 : : HeapTuple reltup;
2107 : : Form_pg_class relform;
2108 : :
2799 andres@anarazel.de 2109 : 23 : relRelation = table_open(RelationRelationId, RowExclusiveLock);
2110 : :
4583 rhaas@postgresql.org 2111 : 23 : reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(OIDOldHeap));
2112 [ - + ]: 23 : if (!HeapTupleIsValid(reltup))
4583 rhaas@postgresql.org 2113 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap);
4583 rhaas@postgresql.org 2114 :CBC 23 : relform = (Form_pg_class) GETSTRUCT(reltup);
2115 : :
2116 : 23 : relform->relfrozenxid = frozenXid;
2117 : 23 : relform->relminmxid = cutoffMulti;
2118 : :
3519 alvherre@alvh.no-ip. 2119 : 23 : CatalogTupleUpdate(relRelation, &reltup->t_self, reltup);
2120 : :
2799 andres@anarazel.de 2121 : 23 : table_close(relRelation, RowExclusiveLock);
2122 : : }
2123 : :
2124 : : /* Destroy new heap with old filenumber */
6072 tgl@sss.pgh.pa.us 2125 : 1461 : object.classId = RelationRelationId;
2126 : 1461 : object.objectId = OIDNewHeap;
2127 : 1461 : object.objectSubId = 0;
2128 : :
167 alvherre@kurilemu.de 2129 [ + + ]: 1461 : if (!reindex)
2130 : : {
2131 : : /*
2132 : : * Make sure the changes in pg_class are visible. This is especially
2133 : : * important if !swap_toast_by_content, so that the correct TOAST
2134 : : * relation is dropped. (reindex_relation() above did not help in this
2135 : : * case))
2136 : : */
2137 : 9 : CommandCounterIncrement();
2138 : : }
2139 : :
2140 : : /*
2141 : : * The new relation is local to our transaction and we know nothing
2142 : : * depends on it, so DROP_RESTRICT should be OK.
2143 : : */
5351 rhaas@postgresql.org 2144 : 1461 : performDeletion(&object, DROP_RESTRICT, PERFORM_DELETION_INTERNAL);
2145 : :
2146 : : /* performDeletion does CommandCounterIncrement at end */
2147 : :
2148 : : /*
2149 : : * Now we must remove any relation mapping entries that we set up for the
2150 : : * transient table, as well as its toast table and toast index if any. If
2151 : : * we fail to do this before commit, the relmapper will complain about new
2152 : : * permanent map entries being added post-bootstrap.
2153 : : */
6069 tgl@sss.pgh.pa.us 2154 [ + + ]: 1550 : for (i = 0; OidIsValid(mapped_tables[i]); i++)
2155 : 89 : RelationMapRemoveMapping(mapped_tables[i]);
2156 : :
2157 : : /*
2158 : : * At this point, everything is kosher except that, if we did toast swap
2159 : : * by links, the toast table's name corresponds to the transient table.
2160 : : * The name is irrelevant to the backend because it's referenced by OID,
2161 : : * but users looking at the catalogs could be confused. Rename it to
2162 : : * prevent this problem.
2163 : : *
2164 : : * Note no lock required on the relation, because we already hold an
2165 : : * exclusive lock on it.
2166 : : */
6072 2167 [ + + ]: 1461 : if (!swap_toast_by_content)
2168 : : {
2169 : : Relation newrel;
2170 : :
2799 andres@anarazel.de 2171 : 1299 : newrel = table_open(OIDOldHeap, NoLock);
6072 tgl@sss.pgh.pa.us 2172 [ + + ]: 1299 : if (OidIsValid(newrel->rd_rel->reltoastrelid))
2173 : : {
2174 : : Oid toastidx;
2175 : : char NewToastName[NAMEDATALEN];
2176 : :
2177 : : /* Get the associated valid index to be renamed */
4826 fujii@postgresql.org 2178 : 383 : toastidx = toast_get_valid_index(newrel->rd_rel->reltoastrelid,
2179 : : AccessExclusiveLock);
2180 : :
2181 : : /* rename the toast table ... */
6072 tgl@sss.pgh.pa.us 2182 : 383 : snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u",
2183 : : OIDOldHeap);
2184 : 383 : RenameRelationInternal(newrel->rd_rel->reltoastrelid,
2185 : : NewToastName, true, false);
2186 : :
2187 : : /* ... and its valid index too. */
2188 : 383 : snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u_index",
2189 : : OIDOldHeap);
2190 : :
2191 : 383 : RenameRelationInternal(toastidx,
2192 : : NewToastName, true, true);
2193 : :
2194 : : /*
2195 : : * Reset the relrewrite for the toast. The command-counter
2196 : : * increment is required here as we are about to update the tuple
2197 : : * that is updated as part of RenameRelationInternal.
2198 : : */
1852 akapila@postgresql.o 2199 : 383 : CommandCounterIncrement();
2200 : 383 : ResetRelRewrite(newrel->rd_rel->reltoastrelid);
2201 : : }
6072 tgl@sss.pgh.pa.us 2202 : 1299 : relation_close(newrel, NoLock);
2203 : : }
2204 : :
2205 : : /* if it's not a catalog table, clear any missing attribute settings */
3098 andrew@dunslane.net 2206 [ + + ]: 1461 : if (!is_system_catalog)
2207 : : {
2208 : : Relation newrel;
2209 : :
2799 andres@anarazel.de 2210 : 1345 : newrel = table_open(OIDOldHeap, NoLock);
3098 andrew@dunslane.net 2211 : 1345 : RelationClearMissing(newrel);
2212 : 1345 : relation_close(newrel, NoLock);
2213 : : }
6072 tgl@sss.pgh.pa.us 2214 : 1461 : }
2215 : :
2216 : : /*
2217 : : * Determine which relations to process, when REPACK/CLUSTER is called
2218 : : * without specifying a table name. The exact process depends on whether
2219 : : * USING INDEX was given or not, and in any case we only return tables and
2220 : : * materialized views that the current user has privileges to repack/cluster.
2221 : : *
2222 : : * If USING INDEX was given, we scan pg_index to find those that have
2223 : : * indisclustered set; if it was not given, scan pg_class and return all
2224 : : * tables.
2225 : : *
2226 : : * Return it as a list of RelToCluster in the given memory context.
2227 : : */
2228 : : static List *
194 alvherre@kurilemu.de 2229 : 18 : get_tables_to_repack(RepackCommand cmd, bool usingindex, MemoryContext permcxt)
2230 : : {
2231 : : Relation catalog;
2232 : : TableScanDesc scan;
2233 : : HeapTuple tuple;
1632 alvherre@alvh.no-ip. 2234 : 18 : List *rtcs = NIL;
2235 : :
194 alvherre@kurilemu.de 2236 [ + + ]: 18 : if (usingindex)
2237 : : {
2238 : : ScanKeyData entry;
2239 : :
2240 : : /*
2241 : : * For USING INDEX, scan pg_index to find those with indisclustered.
2242 : : *
2243 : : * Note we don't obtain lock of any kind on the index, which means the
2244 : : * index or its owning table could be gone or change at any point. We
2245 : : * have to be extra careful when examining catalog state for them.
2246 : : */
2247 : 14 : catalog = table_open(IndexRelationId, AccessShareLock);
2248 : 14 : ScanKeyInit(&entry,
2249 : : Anum_pg_index_indisclustered,
2250 : : BTEqualStrategyNumber, F_BOOLEQ,
2251 : : BoolGetDatum(true));
2252 : 14 : scan = table_beginscan_catalog(catalog, 1, &entry);
2253 [ + + ]: 26 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
2254 : : {
2255 : : RelToCluster *rtc;
2256 : : Form_pg_index index;
2257 : : HeapTuple classtup;
2258 : : Oid relnamespace;
2259 : : char relpersistence;
2260 : : MemoryContext oldcxt;
2261 : :
2262 : 12 : index = (Form_pg_index) GETSTRUCT(tuple);
2263 : :
138 2264 : 12 : classtup = SearchSysCache1(RELOID, ObjectIdGetDatum(index->indrelid));
2265 [ - + ]: 12 : if (!HeapTupleIsValid(classtup))
194 alvherre@kurilemu.de 2266 :UBC 0 : continue;
72 alvherre@kurilemu.de 2267 :CBC 12 : relnamespace = ((Form_pg_class) GETSTRUCT(classtup))->relnamespace;
2268 : 12 : relpersistence = ((Form_pg_class) GETSTRUCT(classtup))->relpersistence;
2269 : 12 : ReleaseSysCache(classtup);
2270 : :
2271 : : /* Skip temp relations belonging to other sessions */
2272 [ - + ]: 12 : if (relpersistence == RELPERSISTENCE_TEMP &&
72 alvherre@kurilemu.de 2273 [ # # ]:UBC 0 : !isTempOrTempToastNamespace(relnamespace))
138 2274 : 0 : continue;
2275 : :
2276 : : /* noisily skip rels which the user can't process */
194 alvherre@kurilemu.de 2277 [ + + ]:CBC 12 : if (!repack_is_permitted_for_relation(cmd, index->indrelid,
2278 : : GetUserId(), false))
2279 : 8 : continue;
2280 : :
2281 : : /* Use a permanent memory context for the result list */
2282 : 4 : oldcxt = MemoryContextSwitchTo(permcxt);
2283 : 4 : rtc = palloc_object(RelToCluster);
2284 : 4 : rtc->tableOid = index->indrelid;
2285 : 4 : rtc->indexOid = index->indexrelid;
2286 : 4 : rtcs = lappend(rtcs, rtc);
2287 : 4 : MemoryContextSwitchTo(oldcxt);
2288 : : }
2289 : : }
2290 : : else
2291 : : {
2292 : 4 : catalog = table_open(RelationRelationId, AccessShareLock);
2293 : 4 : scan = table_beginscan_catalog(catalog, 0, NULL);
2294 : :
2295 [ + + ]: 8692 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
2296 : : {
2297 : : RelToCluster *rtc;
2298 : : Form_pg_class class;
2299 : : MemoryContext oldcxt;
2300 : :
2301 : 8688 : class = (Form_pg_class) GETSTRUCT(tuple);
2302 : :
2303 : : /* Can only process plain tables and matviews */
2304 [ + + ]: 8688 : if (class->relkind != RELKIND_RELATION &&
2305 [ + + ]: 5772 : class->relkind != RELKIND_MATVIEW)
2306 : 5740 : continue;
2307 : :
2308 : : /* Skip temp relations belonging to other sessions */
138 2309 [ + + ]: 2948 : if (class->relpersistence == RELPERSISTENCE_TEMP &&
2310 [ - + ]: 16 : !isTempOrTempToastNamespace(class->relnamespace))
138 alvherre@kurilemu.de 2311 :UBC 0 : continue;
2312 : :
2313 : : /* noisily skip rels which the user can't process */
194 alvherre@kurilemu.de 2314 [ + + ]:CBC 2948 : if (!repack_is_permitted_for_relation(cmd, class->oid,
2315 : : GetUserId(), false))
2316 : 2940 : continue;
2317 : :
2318 : : /* Use a permanent memory context for the result list */
2319 : 8 : oldcxt = MemoryContextSwitchTo(permcxt);
2320 : 8 : rtc = palloc_object(RelToCluster);
2321 : 8 : rtc->tableOid = class->oid;
2322 : 8 : rtc->indexOid = InvalidOid;
2323 : 8 : rtcs = lappend(rtcs, rtc);
2324 : 8 : MemoryContextSwitchTo(oldcxt);
2325 : : }
2326 : : }
2327 : :
2328 : 18 : table_endscan(scan);
72 2329 : 18 : table_close(catalog, AccessShareLock);
2330 : :
1632 alvherre@alvh.no-ip. 2331 : 18 : return rtcs;
2332 : : }
2333 : :
2334 : : /*
2335 : : * Determine relations to process, when REPACK/CLUSTER is called with a
2336 : : * partitioning table; that is, a list of its leaf partitions. That table has
2337 : : * already been opened by caller and is passed as 'rel'. It is closed and
2338 : : * unlocked here before return, so caller should clobber its pointer to avoid
2339 : : * confusion.
2340 : : *
2341 : : * Return it as a list of RelToCluster.
2342 : : *
2343 : : * XXX we don't support CONCURRENTLY for partitioned tables yet.
2344 : : */
2345 : : static List *
62 alvherre@kurilemu.de 2346 : 32 : get_tables_to_repack_partitioned(RepackStmt *stmt, Relation rel,
2347 : : MemoryContext permcxt)
2348 : : {
2349 : : Oid relid;
2350 : : bool rel_is_index;
2351 : : List *inhoids;
1632 alvherre@alvh.no-ip. 2352 : 32 : List *rtcs = NIL;
2353 : :
62 alvherre@kurilemu.de 2354 [ - + ]: 32 : Assert(rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE);
2355 [ - + ]: 32 : Assert(CheckRelationLockedByMe(rel, AccessExclusiveLock, false));
2356 : :
2357 : : /*
2358 : : * We find the list of tables by looking for inheritors. If USING INDEX
2359 : : * was given, look for inheritors of that index, whose name we resolve
2360 : : * now.
2361 : : *
2362 : : * Otherwise we look for inheritors of the table itself.
2363 : : */
2364 [ + + ]: 32 : if (stmt->usingindex)
2365 : : {
2366 : : /*
2367 : : * If no index name was specified when repacking a partitioned table,
2368 : : * punt for now. Maybe we can improve this later.
2369 : : */
2370 [ + + ]: 28 : if (!stmt->indexname)
2371 : : {
2372 [ + + ]: 8 : if (stmt->command == REPACK_COMMAND_CLUSTER)
2373 [ + - ]: 4 : ereport(ERROR,
2374 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2375 : : errmsg("there is no previously clustered index for table \"%s\"",
2376 : : RelationGetRelationName(rel)));
2377 : : else
2378 [ + - ]: 4 : ereport(ERROR,
2379 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2380 : : /*- translator: first %s is name of a SQL command, eg. REPACK */
2381 : : errmsg("cannot execute %s on partitioned table \"%s\" USING INDEX with no index name",
2382 : : RepackCommandAsString(stmt->command),
2383 : : RelationGetRelationName(rel)));
2384 : : }
2385 : :
2386 : 20 : relid = determine_clustered_index(rel, stmt->usingindex,
2387 : 20 : stmt->indexname);
2388 [ - + ]: 20 : if (!OidIsValid(relid))
62 alvherre@kurilemu.de 2389 [ # # ]:UBC 0 : elog(ERROR, "unable to determine index to cluster on");
62 alvherre@kurilemu.de 2390 :CBC 20 : check_index_is_clusterable(rel, relid, AccessExclusiveLock);
2391 : :
2392 : 16 : rel_is_index = true;
2393 : : }
2394 : : else
2395 : : {
2396 : 4 : relid = RelationGetRelid(rel);
2397 : 4 : rel_is_index = false;
2398 : : }
2399 : :
2400 : : /*
2401 : : * Do not lock the children until they're processed. Note that we do hold
2402 : : * a lock on the parent partitioned table.
2403 : : */
194 2404 : 20 : inhoids = find_all_inheritors(relid, NoLock, NULL);
2405 [ + - + + : 148 : foreach_oid(child_oid, inhoids)
+ + ]
2406 : : {
2407 : : Oid table_oid,
2408 : : index_oid;
2409 : : RelToCluster *rtc;
2410 : : MemoryContext oldcxt;
2411 : :
2412 [ + + ]: 108 : if (rel_is_index)
2413 : : {
2414 : : /* consider only leaf indexes */
2415 [ + + ]: 80 : if (get_rel_relkind(child_oid) != RELKIND_INDEX)
2416 : 40 : continue;
2417 : :
2418 : : /*
2419 : : * Although we do have a lock on some ancestor partitioned index,
2420 : : * we may not have one on the immediate parent, so this lookup may
2421 : : * still return invalid.
2422 : : */
62 2423 : 40 : table_oid = IndexGetRelation(child_oid, true);
2424 [ - + ]: 40 : if (!OidIsValid(table_oid))
62 alvherre@kurilemu.de 2425 :UBC 0 : continue;
194 alvherre@kurilemu.de 2426 :CBC 40 : index_oid = child_oid;
2427 : : }
2428 : : else
2429 : : {
2430 : : /* consider only leaf relations */
2431 [ + + ]: 28 : if (get_rel_relkind(child_oid) != RELKIND_RELATION)
2432 : 16 : continue;
2433 : :
2434 : 12 : table_oid = child_oid;
2435 : 12 : index_oid = InvalidOid;
2436 : : }
2437 : :
2438 : : /*
2439 : : * It's possible that the user does not have privileges to CLUSTER the
2440 : : * leaf partition despite having them on the partitioned table. Skip
2441 : : * if so.
2442 : : */
62 2443 [ + + ]: 52 : if (!repack_is_permitted_for_relation(stmt->command, table_oid,
2444 : : GetUserId(), false))
1186 nathan@postgresql.or 2445 : 12 : continue;
2446 : :
2447 : : /* Use a permanent memory context for the result list */
194 alvherre@kurilemu.de 2448 : 40 : oldcxt = MemoryContextSwitchTo(permcxt);
284 michael@paquier.xyz 2449 : 40 : rtc = palloc_object(RelToCluster);
194 alvherre@kurilemu.de 2450 : 40 : rtc->tableOid = table_oid;
2451 : 40 : rtc->indexOid = index_oid;
1632 alvherre@alvh.no-ip. 2452 : 40 : rtcs = lappend(rtcs, rtc);
194 alvherre@kurilemu.de 2453 : 40 : MemoryContextSwitchTo(oldcxt);
2454 : : }
2455 : :
2456 : : /* close parent relation, releasing lock on it */
62 2457 : 20 : table_close(rel, AccessExclusiveLock);
2458 : :
1632 alvherre@alvh.no-ip. 2459 : 20 : return rtcs;
2460 : : }
2461 : :
2462 : :
2463 : : /*
2464 : : * Return whether userid has privileges to execute REPACK/CLUSTER on relid.
2465 : : *
2466 : : * The relation may already be locked by caller, in which case it cannot
2467 : : * possibly go missing; otherwise it may have been removed recently. If
2468 : : * it's been removed, silently return false. If the relation exists but
2469 : : * the user doesn't have the required privs, emit a WARNING and return false.
2470 : : *
2471 : : * Otherwise the relation exists and user has required perms, so return true.
2472 : : */
2473 : : static bool
32 alvherre@kurilemu.de 2474 : 3064 : repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid,
2475 : : bool already_locked)
2476 : : {
72 2477 : 3064 : bool is_missing = false;
2478 : : AclResult result;
2479 : : char *relname;
2480 : :
194 2481 [ + + - + ]: 3064 : Assert(cmd == REPACK_COMMAND_CLUSTER || cmd == REPACK_COMMAND_REPACK);
32 2482 [ + + - + ]: 3064 : Assert(!already_locked ||
2483 : : CheckRelationOidLockedByMe(relid, AccessShareLock, true));
2484 : :
72 2485 : 3064 : result = pg_class_aclcheck_ext(relid, userid, ACL_MAINTAIN, &is_missing);
2486 : :
2487 : : /*
2488 : : * If the relation was concurrently dropped, nothing to do. This is only
2489 : : * reachable when the caller doesn't already have a lock on the relation.
2490 : : */
2491 [ - + ]: 3064 : if (is_missing)
2492 : : {
32 alvherre@kurilemu.de 2493 [ # # ]:UBC 0 : Assert(!already_locked);
72 2494 : 0 : return false;
2495 : : }
2496 : :
72 alvherre@kurilemu.de 2497 [ + + ]:CBC 3064 : if (result == ACLCHECK_OK)
921 nathan@postgresql.or 2498 : 104 : return true;
2499 : :
2500 : : /*
2501 : : * The relation can also be dropped after we tested its ACL and before we
2502 : : * read its relname, so be careful here.
2503 : : */
72 alvherre@kurilemu.de 2504 : 2960 : relname = get_rel_name(relid);
2505 [ + - ]: 2960 : if (relname != NULL)
2506 : : {
2507 [ + - ]: 2960 : ereport(WARNING,
2508 : : errmsg("permission denied to execute %s on \"%s\", skipping it",
2509 : : RepackCommandAsString(cmd), relname));
2510 : 2960 : pfree(relname);
2511 : : }
2512 : :
921 nathan@postgresql.or 2513 : 2960 : return false;
2514 : : }
2515 : :
2516 : :
2517 : : /*
2518 : : * Given a RepackStmt with an indicated relation name, resolve the relation
2519 : : * name, obtain lock on it, then determine what to do based on the relation
2520 : : * type: if it's table and not partitioned, repack it as indicated (using an
2521 : : * existing clustered index, or following the given one), and return NULL.
2522 : : *
2523 : : * On the other hand, if the table is partitioned, do nothing further and
2524 : : * instead return the opened and locked relcache entry, so that caller can
2525 : : * process the partitions using the multiple-table handling code. In this
2526 : : * case, if an index name is given, it's up to the caller to resolve it.
2527 : : */
2528 : : static Relation
167 alvherre@kurilemu.de 2529 : 239 : process_single_relation(RepackStmt *stmt, LOCKMODE lockmode, bool isTopLevel,
2530 : : ClusterParams *params)
2531 : : {
2532 : : Relation rel;
2533 : : Oid tableOid;
2534 : :
194 2535 [ - + ]: 239 : Assert(stmt->relation != NULL);
2536 [ + + - + ]: 239 : Assert(stmt->command == REPACK_COMMAND_CLUSTER ||
2537 : : stmt->command == REPACK_COMMAND_REPACK);
2538 : :
2539 : : /*
2540 : : * Make sure ANALYZE is specified if a column list is present.
2541 : : */
2542 [ + + + + ]: 239 : if ((params->options & CLUOPT_ANALYZE) == 0 && stmt->relation->va_cols != NIL)
2543 [ + - ]: 4 : ereport(ERROR,
2544 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2545 : : errmsg("ANALYZE option must be specified when a column list is provided"));
2546 : :
2547 : : /* Find, lock, and check permissions on the table. */
2548 : 235 : tableOid = RangeVarGetRelidExtended(stmt->relation->relation,
2549 : : lockmode,
2550 : : 0,
2551 : : RangeVarCallbackMaintainsTable,
2552 : : NULL);
2553 : 227 : rel = table_open(tableOid, NoLock);
2554 : :
2555 : : /*
2556 : : * Reject clustering a remote temp table ... their local buffer manager is
2557 : : * not going to cope.
2558 : : */
2559 [ + + + + ]: 227 : if (RELATION_IS_OTHER_TEMP(rel))
2560 [ + - ]: 1 : ereport(ERROR,
2561 : : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2562 : : /*- translator: first %s is name of a SQL command, eg. REPACK */
2563 : : errmsg("cannot execute %s on temporary tables of other sessions",
2564 : : RepackCommandAsString(stmt->command)));
2565 : :
2566 : : /*
2567 : : * For partitioned tables, let caller handle this. Otherwise, process it
2568 : : * here and we're done.
2569 : : */
2570 [ + + ]: 226 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
2571 : 33 : return rel;
2572 : : else
2573 : : {
167 2574 : 193 : Oid indexOid = InvalidOid;
2575 : :
194 2576 : 193 : indexOid = determine_clustered_index(rel, stmt->usingindex,
2577 : 193 : stmt->indexname);
2578 [ + + ]: 185 : if (OidIsValid(indexOid))
167 2579 : 139 : check_index_is_clusterable(rel, indexOid, lockmode);
2580 : :
2581 : 173 : cluster_rel(stmt->command, rel, indexOid, params, isTopLevel);
2582 : :
2583 : : /*
2584 : : * Do an analyze, if requested. We close the transaction and start a
2585 : : * new one, so that we don't hold the stronger lock for longer than
2586 : : * needed.
2587 : : */
194 2588 [ + + ]: 157 : if (params->options & CLUOPT_ANALYZE)
2589 : : {
2590 : 8 : VacuumParams vac_params = {0};
2591 : :
2592 : 8 : PopActiveSnapshot();
2593 : 8 : CommitTransactionCommand();
2594 : :
2595 : 8 : StartTransactionCommand();
2596 : 8 : PushActiveSnapshot(GetTransactionSnapshot());
2597 : :
2598 : 8 : vac_params.options |= VACOPT_ANALYZE;
2599 [ - + ]: 8 : if (params->options & CLUOPT_VERBOSE)
194 alvherre@kurilemu.de 2600 :UBC 0 : vac_params.options |= VACOPT_VERBOSE;
173 nathan@postgresql.or 2601 :CBC 8 : analyze_rel(tableOid, NULL, &vac_params,
194 alvherre@kurilemu.de 2602 : 8 : stmt->relation->va_cols, true, NULL);
2603 : 8 : PopActiveSnapshot();
2604 : 8 : CommandCounterIncrement();
2605 : : }
2606 : :
2607 : 157 : return NULL;
2608 : : }
2609 : : }
2610 : :
2611 : : /*
2612 : : * Given a relation and the usingindex/indexname options in a
2613 : : * REPACK USING INDEX or CLUSTER command, return the OID of the
2614 : : * index to use for clustering the table.
2615 : : *
2616 : : * Caller must hold lock on the relation so that the set of indexes
2617 : : * doesn't change, and must call check_index_is_clusterable.
2618 : : */
2619 : : static Oid
2620 : 213 : determine_clustered_index(Relation rel, bool usingindex, const char *indexname)
2621 : : {
2622 : : Oid indexOid;
2623 : :
2624 [ + + + + ]: 213 : if (indexname == NULL && usingindex)
2625 : : {
2626 : : /*
2627 : : * If USING INDEX with no name is given, find a clustered index, or
2628 : : * error out if none.
2629 : : */
2630 : 20 : indexOid = InvalidOid;
2631 [ + - + + : 44 : foreach_oid(idxoid, RelationGetIndexList(rel))
+ + ]
2632 : : {
2633 [ + + ]: 20 : if (get_index_isclustered(idxoid))
2634 : : {
2635 : 16 : indexOid = idxoid;
2636 : 16 : break;
2637 : : }
2638 : : }
2639 : :
2640 [ + + ]: 20 : if (!OidIsValid(indexOid))
2641 [ + - ]: 4 : ereport(ERROR,
2642 : : errcode(ERRCODE_UNDEFINED_OBJECT),
2643 : : errmsg("there is no previously clustered index for table \"%s\"",
2644 : : RelationGetRelationName(rel)));
2645 : : }
2646 [ + + ]: 193 : else if (indexname != NULL)
2647 : : {
2648 : : /* An index was specified; obtain its OID. */
2649 : 147 : indexOid = get_relname_relid(indexname, rel->rd_rel->relnamespace);
2650 [ + + ]: 147 : if (!OidIsValid(indexOid))
2651 [ + - ]: 4 : ereport(ERROR,
2652 : : errcode(ERRCODE_UNDEFINED_OBJECT),
2653 : : errmsg("index \"%s\" for table \"%s\" does not exist",
2654 : : indexname, RelationGetRelationName(rel)));
2655 : : }
2656 : : else
2657 : 46 : indexOid = InvalidOid;
2658 : :
2659 : 205 : return indexOid;
2660 : : }
2661 : :
2662 : : static const char *
2663 : 3467 : RepackCommandAsString(RepackCommand cmd)
2664 : : {
2665 [ + + + - ]: 3467 : switch (cmd)
2666 : : {
2667 : 3043 : case REPACK_COMMAND_REPACK:
2668 : 3043 : return "REPACK";
2669 : 237 : case REPACK_COMMAND_VACUUMFULL:
2670 : 237 : return "VACUUM";
2671 : 187 : case REPACK_COMMAND_CLUSTER:
2672 : 187 : return "CLUSTER";
2673 : : }
194 alvherre@kurilemu.de 2674 :UBC 0 : return "???"; /* keep compiler quiet */
2675 : : }
2676 : :
2677 : : /*
2678 : : * Apply all the changes stored in 'file'.
2679 : : */
2680 : : static void
167 alvherre@kurilemu.de 2681 :CBC 18 : apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt)
2682 : : {
2683 : 18 : ConcurrentChangeKind kind = '\0';
2684 : 18 : Relation rel = chgcxt->cc_rel;
2685 : : TupleTableSlot *spilled_tuple;
2686 : : TupleTableSlot *old_update_tuple;
2687 : : TupleTableSlot *ondisk_tuple;
2688 : 18 : bool have_old_tuple = false;
2689 : : MemoryContext oldcxt;
2690 : :
2691 : 18 : spilled_tuple = MakeSingleTupleTableSlot(RelationGetDescr(rel),
2692 : : &TTSOpsVirtual);
2693 : 18 : ondisk_tuple = MakeSingleTupleTableSlot(RelationGetDescr(rel),
2694 : : table_slot_callbacks(rel));
2695 : 18 : old_update_tuple = MakeSingleTupleTableSlot(RelationGetDescr(rel),
2696 : : &TTSOpsVirtual);
2697 : :
2698 [ + + ]: 18 : oldcxt = MemoryContextSwitchTo(GetPerTupleMemoryContext(chgcxt->cc_estate));
2699 : :
2700 : : while (true)
2701 : 42 : {
2702 : : size_t nread;
2703 : 60 : ConcurrentChangeKind prevkind = kind;
2704 : :
2705 [ - + ]: 60 : CHECK_FOR_INTERRUPTS();
2706 : :
2707 : 60 : nread = BufFileReadMaybeEOF(file, &kind, 1, true);
2708 [ + + ]: 60 : if (nread == 0) /* done with the file? */
2709 : 18 : break;
2710 : :
2711 : : /*
2712 : : * If this is the old tuple for an update, read it into the tuple slot
2713 : : * and go to the next one. The update itself will be executed on the
2714 : : * next iteration, when we receive the NEW tuple.
2715 : : */
2716 [ + + ]: 42 : if (kind == CHANGE_UPDATE_OLD)
2717 : : {
2718 : 8 : restore_tuple(file, rel, old_update_tuple);
2719 : 8 : have_old_tuple = true;
2720 : 8 : continue;
2721 : : }
2722 : :
2723 : : /*
2724 : : * Just before an UPDATE or DELETE, we must update the command
2725 : : * counter, because the change could refer to a tuple that we have
2726 : : * just inserted; and before an INSERT, we have to do this also if the
2727 : : * previous command was either update or delete.
2728 : : *
2729 : : * With this approach we don't spend so many CCIs for long strings of
2730 : : * only INSERTs, which can't affect one another.
2731 : : */
2732 [ + + + + ]: 34 : if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE ||
2733 [ + - + + : 7 : (kind == CHANGE_INSERT && (prevkind == CHANGE_UPDATE_NEW ||
+ + ]
2734 : : prevkind == CHANGE_DELETE)))
2735 : : {
2736 : 31 : CommandCounterIncrement();
2737 : 31 : UpdateActiveSnapshotCommandId();
2738 : : }
2739 : :
2740 : : /*
2741 : : * Now restore the tuple into the slot and execute the change.
2742 : : */
2743 : 34 : restore_tuple(file, rel, spilled_tuple);
2744 : :
2745 [ + + ]: 34 : if (kind == CHANGE_INSERT)
2746 : : {
2747 : 7 : apply_concurrent_insert(rel, spilled_tuple, chgcxt);
2748 : : }
2749 [ + + ]: 27 : else if (kind == CHANGE_DELETE)
2750 : : {
2751 : : bool found;
2752 : :
2753 : : /* Find the tuple to be deleted */
2754 : 3 : found = find_target_tuple(rel, chgcxt, spilled_tuple, ondisk_tuple);
2755 [ - + ]: 3 : if (!found)
115 alvherre@kurilemu.de 2756 [ # # ]:UBC 0 : elog(ERROR, "could not find target tuple");
167 alvherre@kurilemu.de 2757 :CBC 3 : apply_concurrent_delete(rel, ondisk_tuple);
2758 : : }
2759 [ + - ]: 24 : else if (kind == CHANGE_UPDATE_NEW)
2760 : : {
2761 : : TupleTableSlot *key;
2762 : : bool found;
2763 : :
2764 [ + + ]: 24 : if (have_old_tuple)
2765 : 8 : key = old_update_tuple;
2766 : : else
2767 : 16 : key = spilled_tuple;
2768 : :
2769 : : /* Find the tuple to be updated or deleted. */
2770 : 24 : found = find_target_tuple(rel, chgcxt, key, ondisk_tuple);
2771 [ - + ]: 24 : if (!found)
115 alvherre@kurilemu.de 2772 [ # # ]:UBC 0 : elog(ERROR, "could not find target tuple");
2773 : :
2774 : : /*
2775 : : * If 'tup' contains TOAST pointers, they point to the old
2776 : : * relation's toast. Copy the corresponding TOAST pointers for the
2777 : : * new relation from the existing tuple. (The fact that we
2778 : : * received a TOAST pointer here implies that the attribute hasn't
2779 : : * changed.)
2780 : : */
167 alvherre@kurilemu.de 2781 :CBC 24 : adjust_toast_pointers(rel, spilled_tuple, ondisk_tuple);
2782 : :
2783 : 24 : apply_concurrent_update(rel, spilled_tuple, ondisk_tuple, chgcxt);
2784 : :
2785 : 24 : ExecClearTuple(old_update_tuple);
2786 : 24 : have_old_tuple = false;
2787 : : }
2788 : : else
167 alvherre@kurilemu.de 2789 [ # # ]:UBC 0 : elog(ERROR, "unrecognized kind of change: %d", kind);
2790 : :
167 alvherre@kurilemu.de 2791 [ + - ]:CBC 34 : ResetPerTupleExprContext(chgcxt->cc_estate);
2792 : : }
2793 : :
2794 : : /* Cleanup. */
2795 : 18 : ExecDropSingleTupleTableSlot(spilled_tuple);
2796 : 18 : ExecDropSingleTupleTableSlot(ondisk_tuple);
2797 : 18 : ExecDropSingleTupleTableSlot(old_update_tuple);
2798 : :
2799 : 18 : MemoryContextSwitchTo(oldcxt);
2800 : 18 : }
2801 : :
2802 : : /*
2803 : : * Apply an insert from the spill of concurrent changes to the new copy of the
2804 : : * table.
2805 : : */
2806 : : static void
2807 : 7 : apply_concurrent_insert(Relation rel, TupleTableSlot *slot,
2808 : : ChangeContext *chgcxt)
2809 : : {
2810 : : /* Put the tuple in the table, but make sure it won't be decoded */
2811 : 7 : table_tuple_insert(rel, slot, GetCurrentCommandId(true),
2812 : : TABLE_INSERT_NO_LOGICAL, NULL);
2813 : :
2814 : : /* Update indexes with this new tuple. */
2815 : 7 : ExecInsertIndexTuples(chgcxt->cc_rri,
2816 : : chgcxt->cc_estate,
2817 : : 0,
2818 : : slot,
2819 : : NIL, NULL);
2820 : 7 : pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED, 1);
2821 : 7 : }
2822 : :
2823 : : /*
2824 : : * Apply an update from the spill of concurrent changes to the new copy of the
2825 : : * table.
2826 : : */
2827 : : static void
2828 : 24 : apply_concurrent_update(Relation rel, TupleTableSlot *spilled_tuple,
2829 : : TupleTableSlot *ondisk_tuple,
2830 : : ChangeContext *chgcxt)
2831 : : {
2832 : : LockTupleMode lockmode;
2833 : : TM_FailureData tmfd;
2834 : : TU_UpdateIndexes update_indexes;
2835 : : TM_Result res;
2836 : :
2837 : : /*
2838 : : * Carry out the update, skipping logical decoding for it.
2839 : : */
2840 : 24 : res = table_tuple_update(rel, &(ondisk_tuple->tts_tid), spilled_tuple,
2841 : : GetCurrentCommandId(true),
2842 : : TABLE_UPDATE_NO_LOGICAL,
2843 : : InvalidSnapshot,
2844 : : InvalidSnapshot,
2845 : : false,
2846 : : &tmfd, &lockmode, &update_indexes);
2847 [ - + ]: 24 : if (res != TM_Ok)
167 alvherre@kurilemu.de 2848 [ # # ]:UBC 0 : ereport(ERROR,
2849 : : errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
2850 : : errmsg("could not apply concurrent %s on relation \"%s\"",
2851 : : "UPDATE", RelationGetRelationName(rel)));
2852 : :
167 alvherre@kurilemu.de 2853 [ + + ]:CBC 24 : if (update_indexes != TU_None)
2854 : : {
2855 : 8 : uint32 flags = EIIT_IS_UPDATE;
2856 : :
2857 [ - + ]: 8 : if (update_indexes == TU_Summarizing)
167 alvherre@kurilemu.de 2858 :UBC 0 : flags |= EIIT_ONLY_SUMMARIZING;
167 alvherre@kurilemu.de 2859 :CBC 8 : ExecInsertIndexTuples(chgcxt->cc_rri,
2860 : : chgcxt->cc_estate,
2861 : : flags,
2862 : : spilled_tuple,
2863 : : NIL, NULL);
2864 : : }
2865 : :
2866 : 24 : pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_UPDATED, 1);
2867 : 24 : }
2868 : :
2869 : : static void
2870 : 3 : apply_concurrent_delete(Relation rel, TupleTableSlot *slot)
2871 : : {
2872 : : TM_Result res;
2873 : : TM_FailureData tmfd;
2874 : :
2875 : : /*
2876 : : * Delete tuple from the new heap, skipping logical decoding for it.
2877 : : */
2878 : 3 : res = table_tuple_delete(rel, &(slot->tts_tid),
2879 : : GetCurrentCommandId(true),
2880 : : TABLE_DELETE_NO_LOGICAL,
2881 : : InvalidSnapshot, InvalidSnapshot,
2882 : : false,
2883 : : &tmfd);
2884 : :
2885 [ - + ]: 3 : if (res != TM_Ok)
167 alvherre@kurilemu.de 2886 [ # # ]:UBC 0 : ereport(ERROR,
2887 : : errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
2888 : : errmsg("could not apply concurrent %s on relation \"%s\"",
2889 : : "DELETE", RelationGetRelationName(rel)));
2890 : :
167 alvherre@kurilemu.de 2891 :CBC 3 : pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
2892 : 3 : }
2893 : :
2894 : : /*
2895 : : * Read tuple from file and put it in the input slot. All memory is allocated
2896 : : * in the current memory context; caller is responsible for freeing it as
2897 : : * appropriate.
2898 : : *
2899 : : * External attributes are stored in separate memory chunks, in order to avoid
2900 : : * exceeding MaxAllocSize - that could happen if the individual attributes are
2901 : : * smaller than MaxAllocSize but the whole tuple is bigger.
2902 : : */
2903 : : static void
2904 : 42 : restore_tuple(BufFile *file, Relation relation, TupleTableSlot *slot)
2905 : : {
2906 : : uint32 t_len;
2907 : : HeapTuple tup;
2908 : : int natt_ext;
2909 : :
2910 : : /* Read the tuple. */
2911 : 42 : BufFileReadExact(file, &t_len, sizeof(t_len));
2912 : 42 : tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
2913 : 42 : tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
2914 : 42 : BufFileReadExact(file, tup->t_data, t_len);
2915 : 42 : tup->t_len = t_len;
2916 : 42 : ItemPointerSetInvalid(&tup->t_self);
2917 : 42 : tup->t_tableOid = RelationGetRelid(relation);
2918 : :
2919 : : /*
2920 : : * Put the tuple we read in a slot. This deforms it, so that we can hack
2921 : : * the external attributes in place.
2922 : : */
2923 : 42 : ExecForceStoreHeapTuple(tup, slot, false);
2924 : :
2925 : : /*
2926 : : * Next, read any attributes we stored separately into the tts_values
2927 : : * array elements expecting them, if any. This matches
2928 : : * repack_store_change.
2929 : : */
2930 : 42 : BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
2931 [ + + ]: 42 : if (natt_ext > 0)
2932 : : {
2933 : 12 : TupleDesc desc = slot->tts_tupleDescriptor;
2934 : :
2935 [ + + ]: 69 : for (int i = 0; i < desc->natts; i++)
2936 : : {
2937 : 57 : CompactAttribute *attr = TupleDescCompactAttr(desc, i);
2938 : : varlena *varlen;
2939 : : uint64 chunk_header;
2940 : : void *value;
2941 : : Size varlensz;
2942 : :
2943 [ + + + + ]: 57 : if (attr->attisdropped || attr->attlen != -1)
2944 : 41 : continue;
2945 [ - + ]: 23 : if (slot_attisnull(slot, i + 1))
167 alvherre@kurilemu.de 2946 :UBC 0 : continue;
167 alvherre@kurilemu.de 2947 :CBC 23 : varlen = (varlena *) DatumGetPointer(slot->tts_values[i]);
2948 [ + + ]: 23 : if (!VARATT_IS_EXTERNAL_INDIRECT(varlen))
2949 : 7 : continue;
2950 : 16 : slot_getsomeattrs(slot, i + 1);
2951 : :
2952 : 16 : BufFileReadExact(file, &chunk_header, VARHDRSZ);
2953 : 16 : varlensz = VARSIZE_ANY(&chunk_header);
2954 : :
2955 : 16 : value = palloc(varlensz);
145 2956 : 16 : memcpy(value, &chunk_header, VARHDRSZ);
167 2957 : 16 : BufFileReadExact(file, (char *) value + VARHDRSZ, varlensz - VARHDRSZ);
2958 : :
2959 : 16 : slot->tts_values[i] = PointerGetDatum(value);
2960 : 16 : natt_ext--;
2961 [ - + ]: 16 : if (natt_ext < 0)
96 alvherre@kurilemu.de 2962 [ # # ]:UBC 0 : elog(ERROR, "insufficient number of attributes stored separately");
2963 : : }
2964 : :
96 alvherre@kurilemu.de 2965 [ - + ]:CBC 12 : if (natt_ext != 0)
96 alvherre@kurilemu.de 2966 [ # # ]:UBC 0 : elog(ERROR,
2967 : : "unexpected number of attributes stored separately (%d remaining)",
2968 : : natt_ext);
2969 : : }
167 alvherre@kurilemu.de 2970 :CBC 42 : }
2971 : :
2972 : : /*
2973 : : * Adjust 'dest' replacing any EXTERNAL_ONDISK toast pointers with the
2974 : : * corresponding ones from 'src'.
2975 : : */
2976 : : static void
2977 : 24 : adjust_toast_pointers(Relation relation, TupleTableSlot *dest, TupleTableSlot *src)
2978 : : {
2979 : 24 : TupleDesc desc = dest->tts_tupleDescriptor;
2980 : :
2981 [ + + ]: 118 : for (int i = 0; i < desc->natts; i++)
2982 : : {
2983 : 94 : CompactAttribute *attr = TupleDescCompactAttr(desc, i);
2984 : : varlena *varlena_dst;
2985 : :
2986 [ + + ]: 94 : if (attr->attisdropped)
2987 : 24 : continue;
2988 [ + + ]: 70 : if (attr->attlen != -1)
2989 : 38 : continue;
2990 [ - + ]: 32 : if (slot_attisnull(dest, i + 1))
167 alvherre@kurilemu.de 2991 :UBC 0 : continue;
2992 : :
167 alvherre@kurilemu.de 2993 :CBC 32 : slot_getsomeattrs(dest, i + 1);
2994 : :
2995 : 32 : varlena_dst = (varlena *) DatumGetPointer(dest->tts_values[i]);
2996 [ + + ]: 32 : if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
2997 : 30 : continue;
2998 : 2 : slot_getsomeattrs(src, i + 1);
2999 : :
3000 : 2 : dest->tts_values[i] = src->tts_values[i];
3001 : : }
3002 : 24 : }
3003 : :
3004 : : /*
3005 : : * Find the tuple to be updated or deleted by the given data change, whose
3006 : : * tuple has already been loaded into locator.
3007 : : *
3008 : : * If the tuple is found, put it in retrieved and return true. If the tuple is
3009 : : * not found, return false.
3010 : : */
3011 : : static bool
3012 : 27 : find_target_tuple(Relation rel, ChangeContext *chgcxt, TupleTableSlot *locator,
3013 : : TupleTableSlot *retrieved)
3014 : : {
3015 : 27 : Form_pg_index idx = chgcxt->cc_ident_index->rd_index;
3016 : : IndexScanDesc scan;
132 3017 : 27 : bool retval = false;
3018 : :
3019 : : /*
3020 : : * Scan key is passed by caller, so it does not have to be constructed
3021 : : * multiple times. Key entries have all fields initialized, except for
3022 : : * sk_argument.
3023 : : *
3024 : : * Use the incoming tuple to finalize the scan key.
3025 : : */
167 3026 [ + + ]: 56 : for (int i = 0; i < chgcxt->cc_ident_key_nentries; i++)
3027 : : {
3028 : 29 : ScanKey entry = &chgcxt->cc_ident_key[i];
3029 : 29 : AttrNumber attno = idx->indkey.values[i];
3030 : :
3031 : 29 : entry->sk_argument = locator->tts_values[attno - 1];
3032 [ - + ]: 29 : Assert(!locator->tts_isnull[attno - 1]);
3033 : : }
3034 : :
3035 : : /* XXX no instrumentation for now */
5 pg@bowt.ie 3036 :GNC 27 : scan = index_beginscan(rel, chgcxt->cc_ident_index, false, GetActiveSnapshot(),
3037 : : NULL, chgcxt->cc_ident_key_nentries, 0, 0);
167 alvherre@kurilemu.de 3038 :CBC 27 : index_rescan(scan, chgcxt->cc_ident_key, chgcxt->cc_ident_key_nentries, NULL, 0);
5 pg@bowt.ie 3039 [ + - ]:GNC 28 : while (table_index_getnext_slot(scan, ForwardScanDirection, retrieved))
3040 : : {
3041 : : /* Be wary of temporal constraints */
132 alvherre@kurilemu.de 3042 [ + + + + ]:CBC 28 : if (scan->xs_recheck && !identity_key_equal(chgcxt, locator, retrieved))
3043 : : {
3044 [ - + ]: 1 : CHECK_FOR_INTERRUPTS();
3045 : 1 : continue;
3046 : : }
3047 : :
3048 : 27 : retval = true;
3049 : 27 : break;
3050 : : }
167 3051 : 27 : index_endscan(scan);
3052 : :
3053 : 27 : return retval;
3054 : : }
3055 : :
3056 : : /*
3057 : : * Check whether the candidate tuple matches the locator tuple on all replica
3058 : : * identity key columns, using the same equality operators as the identity
3059 : : * index scan. The locator tuple has already been loaded into cc_ident_key.
3060 : : *
3061 : : * This is needed to filter lossy index matches, such as GiST multirange scans
3062 : : * used for temporal constraints.
3063 : : */
3064 : : static bool
132 3065 : 2 : identity_key_equal(ChangeContext *chgcxt, TupleTableSlot *locator,
3066 : : TupleTableSlot *candidate)
3067 : : {
3068 : 2 : slot_getsomeattrs(locator, chgcxt->cc_last_key_attno);
3069 : 2 : slot_getsomeattrs(candidate, chgcxt->cc_last_key_attno);
3070 : :
3071 [ + + ]: 4 : for (int i = 0; i < chgcxt->cc_ident_key_nentries; i++)
3072 : : {
3073 : 3 : ScanKey entry = &chgcxt->cc_ident_key[i];
3074 : 3 : AttrNumber attno = chgcxt->cc_ident_index->rd_index->indkey.values[i];
3075 : :
3076 [ - + ]: 3 : Assert(attno > 0);
3077 : :
3078 [ - + ]: 3 : if (locator->tts_isnull[attno - 1] != candidate->tts_isnull[attno - 1])
132 alvherre@kurilemu.de 3079 :UBC 0 : return false;
3080 : :
132 alvherre@kurilemu.de 3081 [ - + ]:CBC 3 : if (locator->tts_isnull[attno - 1])
132 alvherre@kurilemu.de 3082 :UBC 0 : continue;
3083 : :
132 alvherre@kurilemu.de 3084 [ + + ]:CBC 3 : if (!DatumGetBool(FunctionCall2Coll(&entry->sk_func,
3085 : : entry->sk_collation,
3086 : 3 : candidate->tts_values[attno - 1],
3087 : : entry->sk_argument)))
3088 : 1 : return false;
3089 : : }
3090 : :
3091 : 1 : return true;
3092 : : }
3093 : :
3094 : : /*
3095 : : * Decode and apply concurrent changes, up to (and including) the record whose
3096 : : * LSN is 'end_of_wal'.
3097 : : *
3098 : : * XXX the names "process_concurrent_changes" and "apply_concurrent_changes"
3099 : : * are far too similar to each other.
3100 : : */
3101 : : static void
167 3102 : 18 : process_concurrent_changes(XLogRecPtr end_of_wal, ChangeContext *chgcxt, bool done)
3103 : : {
3104 : : DecodingWorkerShared *shared;
3105 : : char fname[MAXPGPATH];
3106 : : BufFile *file;
3107 : :
3108 : 18 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
3109 : : PROGRESS_REPACK_PHASE_CATCH_UP);
3110 : :
3111 : : /* Ask the worker for the file. */
3112 : 18 : shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
3113 : 18 : SpinLockAcquire(&shared->mutex);
3114 : 18 : shared->lsn_upto = end_of_wal;
3115 : 18 : shared->done = done;
3116 : 18 : SpinLockRelease(&shared->mutex);
3117 : :
3118 : : /*
3119 : : * The worker needs to finish processing of the current WAL record. Even
3120 : : * if it's idle, it'll need to close the output file. Thus we're likely to
3121 : : * wait, so prepare for sleep.
3122 : : */
3123 : 18 : ConditionVariablePrepareToSleep(&shared->cv);
3124 : : for (;;)
3125 : 18 : {
3126 : : int last_exported;
3127 : :
3128 : 36 : SpinLockAcquire(&shared->mutex);
3129 : 36 : last_exported = shared->last_exported;
3130 : 36 : SpinLockRelease(&shared->mutex);
3131 : :
3132 : : /*
3133 : : * Has the worker exported the file we are waiting for?
3134 : : */
3135 [ + + ]: 36 : if (last_exported == chgcxt->cc_file_seq)
3136 : 18 : break;
3137 : :
3138 : 18 : ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
3139 : : }
3140 : 18 : ConditionVariableCancelSleep();
3141 : :
3142 : : /* Open the file. */
3143 : 18 : DecodingWorkerFileName(fname, shared->relid, chgcxt->cc_file_seq);
3144 : 18 : file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
3145 : 18 : apply_concurrent_changes(file, chgcxt);
3146 : :
3147 : 18 : BufFileClose(file);
3148 : :
3149 : : /* Get ready for the next file. */
3150 : 18 : chgcxt->cc_file_seq++;
3151 : 18 : }
3152 : :
3153 : : /*
3154 : : * Initialize the ChangeContext struct for the given relation, with
3155 : : * the given index as identity index.
3156 : : */
3157 : : static void
3158 : 9 : initialize_change_context(ChangeContext *chgcxt,
3159 : : Relation relation, Oid ident_index_id)
3160 : : {
3161 : 9 : chgcxt->cc_rel = relation;
3162 : :
3163 : : /* Only initialize fields needed by ExecInsertIndexTuples(). */
3164 : 9 : chgcxt->cc_estate = CreateExecutorState();
3165 : :
3166 : : /*
3167 : : * Set up a range table for the executor, containing our repacked table as
3168 : : * its only member.
3169 : : */
3170 : : {
3171 : : RangeTblEntry *rte;
79 3172 : 9 : TupleDesc desc = RelationGetDescr(relation);
3173 : 9 : List *perminfos = NIL;
3174 : 9 : Bitmapset *updatedCols = NULL;
3175 : : RTEPermissionInfo *perminfo;
3176 : :
3177 : : /*
3178 : : * For our use, the RTE only needs to have perminfoindex initialized,
3179 : : * but there's no reason to not set the fields whose values we have at
3180 : : * hand.
3181 : : */
3182 : 9 : rte = makeNode(RangeTblEntry);
3183 : 9 : rte->rtekind = RTE_RELATION;
3184 : 9 : rte->relid = RelationGetRelid(relation);
3185 : 9 : rte->relkind = RelationGetForm(relation)->relkind;
3186 : : /* Create the RTEPermissionInfo instance (and set ->perminfoindex). */
3187 : 9 : addRTEPermissionInfo(&perminfos, rte);
3188 : :
3189 : : /*
3190 : : * Initialize updatedCols to show that all columns are updated. This
3191 : : * is of course not necessarily true, and we cannot know this early;
3192 : : * but this is only used by ExecInsertIndexTuples to flag index
3193 : : * updates with no logical value changes, so if it's wrong, nothing
3194 : : * terribly bad happens. We may want to improve this someday though.
3195 : : *
3196 : : * Don't claim that dropped columns are changed though.
3197 : : */
3198 [ + + ]: 31 : for (int i = 0; i < desc->natts; i++)
3199 : : {
3200 : 22 : CompactAttribute *attr = TupleDescCompactAttr(desc, i);
3201 : :
3202 [ + + ]: 22 : if (attr->attisdropped)
3203 : 2 : continue;
3204 : 20 : updatedCols = bms_add_member(updatedCols,
3205 : : i + 1 - FirstLowInvalidHeapAttributeNumber);
3206 : : }
3207 : :
3208 : : /* install updatedCols in the right place */
3209 : 9 : perminfo = getRTEPermissionInfo(perminfos, rte);
3210 : 9 : perminfo->updatedCols = updatedCols;
3211 : :
3212 : : /* finally we can initialize the range table proper */
3213 : 9 : ExecInitRangeTable(chgcxt->cc_estate, list_make1(rte), perminfos,
3214 : : bms_make_singleton(1));
3215 : : }
3216 : :
3217 : : /* Set up our ResultRelInfo to use for index updates */
3218 : 9 : chgcxt->cc_rri = makeNode(ResultRelInfo);
3219 : 9 : InitResultRelInfo(chgcxt->cc_rri, relation, 1, NULL, 0);
167 3220 : 9 : ExecOpenIndices(chgcxt->cc_rri, false);
3221 : :
3222 : : /*
3223 : : * The table's relcache entry already has the relcache entry for the
3224 : : * identity index; find that.
3225 : : */
3226 : 9 : chgcxt->cc_ident_index = NULL;
3227 [ + - ]: 9 : for (int i = 0; i < chgcxt->cc_rri->ri_NumIndices; i++)
3228 : : {
3229 : : Relation ind_rel;
3230 : :
3231 : 9 : ind_rel = chgcxt->cc_rri->ri_IndexRelationDescs[i];
3232 [ + - ]: 9 : if (ind_rel->rd_id == ident_index_id)
3233 : : {
3234 : 9 : chgcxt->cc_ident_index = ind_rel;
3235 : 9 : break;
3236 : : }
3237 : : }
3238 [ - + ]: 9 : if (chgcxt->cc_ident_index == NULL)
115 alvherre@kurilemu.de 3239 [ # # ]:UBC 0 : elog(ERROR, "could not find identity index");
3240 : :
3241 : : /* Set up for scanning said identity index */
3242 : : {
3243 : : Form_pg_index indexForm;
3244 : :
167 alvherre@kurilemu.de 3245 :CBC 9 : indexForm = chgcxt->cc_ident_index->rd_index;
3246 : 9 : chgcxt->cc_ident_key_nentries = indexForm->indnkeyatts;
3247 : 9 : chgcxt->cc_ident_key = (ScanKey) palloc_array(ScanKeyData, indexForm->indnkeyatts);
3248 [ + + ]: 20 : for (int i = 0; i < indexForm->indnkeyatts; i++)
3249 : : {
3250 : : ScanKey entry;
3251 : : Oid opfamily,
3252 : : opcintype,
3253 : : opno,
3254 : : opcode;
3255 : : StrategyNumber eq_strategy;
3256 : :
3257 : 11 : entry = &chgcxt->cc_ident_key[i];
3258 : :
3259 : 11 : opfamily = chgcxt->cc_ident_index->rd_opfamily[i];
3260 : 11 : opcintype = chgcxt->cc_ident_index->rd_opcintype[i];
132 3261 : 11 : eq_strategy = IndexAmTranslateCompareType(COMPARE_EQ,
3262 : 11 : chgcxt->cc_ident_index->rd_rel->relam,
3263 : : opfamily, false);
3264 [ - + ]: 11 : if (eq_strategy == InvalidStrategy)
132 alvherre@kurilemu.de 3265 [ # # ]:UBC 0 : elog(ERROR, "could not find equality strategy for index operator family %u for type %u",
3266 : : opfamily, opcintype);
167 alvherre@kurilemu.de 3267 :CBC 11 : opno = get_opfamily_member(opfamily, opcintype, opcintype,
3268 : : eq_strategy);
3269 [ - + ]: 11 : if (!OidIsValid(opno))
132 alvherre@kurilemu.de 3270 [ # # ]:UBC 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
3271 : : eq_strategy, opcintype, opcintype, opfamily);
167 alvherre@kurilemu.de 3272 :CBC 11 : opcode = get_opcode(opno);
3273 [ - + ]: 11 : if (!OidIsValid(opcode))
132 alvherre@kurilemu.de 3274 [ # # ]:UBC 0 : elog(ERROR, "missing oprcode for operator %u", opno);
3275 : :
3276 : : /* Initialize everything but argument. */
167 alvherre@kurilemu.de 3277 :CBC 11 : ScanKeyInit(entry,
3278 : 11 : i + 1,
3279 : : eq_strategy, opcode,
3280 : : (Datum) 0);
3281 : 11 : entry->sk_collation = chgcxt->cc_ident_index->rd_indcollation[i];
3282 : : }
3283 : : }
3284 : :
3285 : : /* Determine the last column we must deform to read the identity */
132 3286 : 9 : chgcxt->cc_last_key_attno = InvalidAttrNumber;
3287 [ + + ]: 20 : for (int i = 0; i < chgcxt->cc_ident_key_nentries; i++)
3288 : : {
3289 : 11 : AttrNumber attno = chgcxt->cc_ident_index->rd_index->indkey.values[i];
3290 : :
3291 [ - + ]: 11 : Assert(attno > 0);
3292 : 11 : chgcxt->cc_last_key_attno = Max(chgcxt->cc_last_key_attno, attno);
3293 : : }
3294 : :
167 3295 : 9 : chgcxt->cc_file_seq = WORKER_FILE_SNAPSHOT + 1;
3296 : 9 : }
3297 : :
3298 : : /*
3299 : : * Free up resources taken by a ChangeContext.
3300 : : */
3301 : : static void
3302 : 9 : release_change_context(ChangeContext *chgcxt)
3303 : : {
3304 : 9 : ExecCloseIndices(chgcxt->cc_rri);
3305 : 9 : FreeExecutorState(chgcxt->cc_estate);
3306 : : /* XXX are these pfrees necessary? */
3307 : 9 : pfree(chgcxt->cc_rri);
3308 : 9 : pfree(chgcxt->cc_ident_key);
3309 : 9 : }
3310 : :
3311 : : /*
3312 : : * The final steps of rebuild_relation() for concurrent processing.
3313 : : *
3314 : : * On entry, NewHeap is locked in AccessExclusiveLock mode. OldHeap and its
3315 : : * clustering index (if one is passed) are still locked in a mode that allows
3316 : : * concurrent data changes. On exit, both tables and their indexes are closed,
3317 : : * but locked in AccessExclusiveLock mode.
3318 : : */
3319 : : static void
3320 : 9 : rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
3321 : : Oid identIdx, TransactionId frozenXid,
3322 : : MultiXactId cutoffMulti)
3323 : : {
3324 : : List *ind_oids_new;
3325 : 9 : Oid old_table_oid = RelationGetRelid(OldHeap);
3326 : 9 : Oid new_table_oid = RelationGetRelid(NewHeap);
3327 : 9 : List *ind_oids_old = RelationGetIndexList(OldHeap);
3328 : : ListCell *lc,
3329 : : *lc2;
3330 : : char relpersistence;
3331 : : bool is_system_catalog;
3332 : : Oid ident_idx_new;
3333 : : XLogRecPtr end_of_wal;
3334 : : List *indexrels;
3335 : : ChangeContext chgcxt;
3336 : :
3337 [ - + ]: 9 : Assert(CheckRelationLockedByMe(OldHeap, ShareUpdateExclusiveLock, false));
3338 [ - + ]: 9 : Assert(CheckRelationLockedByMe(NewHeap, AccessExclusiveLock, false));
3339 : :
3340 : : /*
3341 : : * Unlike the exclusive case, we build new indexes for the new relation
3342 : : * rather than swapping the storage and reindexing the old relation. The
3343 : : * point is that the index build can take some time, so we do it before we
3344 : : * get AccessExclusiveLock on the old heap and therefore we cannot swap
3345 : : * the heap storage yet.
3346 : : *
3347 : : * index_create() will lock the new indexes using AccessExclusiveLock - no
3348 : : * need to change that. At the same time, we use ShareUpdateExclusiveLock
3349 : : * to lock the existing indexes - that should be enough to prevent others
3350 : : * from changing them while we're repacking the relation. The lock on
3351 : : * table should prevent others from changing the index column list, but
3352 : : * might not be enough for commands like ALTER INDEX ... SET ... (Those
3353 : : * are not necessarily dangerous, but can make user confused if the
3354 : : * changes they do get lost due to REPACK.)
3355 : : */
3356 : 9 : ind_oids_new = build_new_indexes(NewHeap, OldHeap, ind_oids_old);
3357 : :
3358 : : /*
3359 : : * The identity index in the new relation appears in the same relative
3360 : : * position as the corresponding index in the old relation. Find it.
3361 : : */
3362 : 9 : ident_idx_new = InvalidOid;
3363 [ + - + - : 18 : foreach_oid(ind_old, ind_oids_old)
+ + ]
3364 : : {
3365 [ + - ]: 9 : if (identIdx == ind_old)
3366 : : {
3367 : 9 : int pos = foreach_current_index(ind_old);
3368 : :
139 3369 [ - + ]: 9 : if (list_length(ind_oids_new) <= pos)
167 alvherre@kurilemu.de 3370 [ # # ]:UBC 0 : elog(ERROR, "list of new indexes too short");
167 alvherre@kurilemu.de 3371 :CBC 9 : ident_idx_new = list_nth_oid(ind_oids_new, pos);
3372 : 9 : break;
3373 : : }
3374 : : }
3375 [ - + ]: 9 : if (!OidIsValid(ident_idx_new))
167 alvherre@kurilemu.de 3376 [ # # ]:UBC 0 : elog(ERROR, "could not find index matching \"%s\" at the new relation",
3377 : : get_rel_name(identIdx));
3378 : :
3379 : : /* Gather information to apply concurrent changes. */
167 alvherre@kurilemu.de 3380 :CBC 9 : initialize_change_context(&chgcxt, NewHeap, ident_idx_new);
3381 : :
3382 : : /*
3383 : : * During testing, wait for another backend to perform concurrent data
3384 : : * changes which we will process below.
3385 : : */
3386 : 9 : INJECTION_POINT("repack-concurrently-before-lock", NULL);
3387 : :
3388 : : /*
3389 : : * Flush all WAL records inserted so far (possibly except for the last
3390 : : * incomplete page; see GetInsertRecPtr), to minimize the amount of data
3391 : : * we need to flush while holding exclusive lock on the source table.
3392 : : */
3393 : 9 : XLogFlush(GetXLogInsertEndRecPtr());
3394 : 9 : end_of_wal = GetFlushRecPtr(NULL);
3395 : :
3396 : : /*
3397 : : * Apply concurrent changes first time, to minimize the time we need to
3398 : : * hold AccessExclusiveLock. (Quite some amount of WAL could have been
3399 : : * written during the data copying and index creation.)
3400 : : */
3401 : 9 : process_concurrent_changes(end_of_wal, &chgcxt, false);
3402 : :
3403 : : /*
3404 : : * Acquire AccessExclusiveLock on the table, its TOAST relation (if there
3405 : : * is one), all its indexes, so that we can swap the files.
3406 : : */
3407 : 9 : LockRelationOid(old_table_oid, AccessExclusiveLock);
3408 : :
3409 : : /*
3410 : : * Lock all indexes now, not only the clustering one: all indexes need to
3411 : : * have their files swapped. While doing that, store their relation
3412 : : * references in a zero-terminated array, to handle predicate locks below.
3413 : : */
3414 : 9 : indexrels = NIL;
3415 [ + - + + : 28 : foreach_oid(ind_oid, ind_oids_old)
+ + ]
3416 : : {
3417 : : Relation index;
3418 : :
3419 : 10 : index = index_open(ind_oid, AccessExclusiveLock);
3420 : :
3421 : : /*
3422 : : * Some things about the index may have changed before we locked the
3423 : : * index, such as ALTER INDEX RENAME. We don't need to do anything
3424 : : * here to absorb those changes in the new index.
3425 : : */
3426 : 10 : indexrels = lappend(indexrels, index);
3427 : : }
3428 : :
3429 : : /*
3430 : : * Lock the OldHeap's TOAST relation exclusively - again, the lock is
3431 : : * needed to swap the files.
3432 : : */
3433 [ + + ]: 9 : if (OidIsValid(OldHeap->rd_rel->reltoastrelid))
3434 : 5 : LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
3435 : :
3436 : : /*
3437 : : * Tuples and pages of the old heap will be gone, but the heap will stay.
3438 : : */
3439 : 9 : TransferPredicateLocksToHeapRelation(OldHeap);
3440 [ + - + + : 28 : foreach_ptr(RelationData, index, indexrels)
+ + ]
3441 : : {
3442 : 10 : TransferPredicateLocksToHeapRelation(index);
3443 : 10 : index_close(index, NoLock);
3444 : : }
3445 : 9 : list_free(indexrels);
3446 : :
3447 : : /*
3448 : : * Flush WAL again, to make sure that all changes committed while we were
3449 : : * waiting for the exclusive lock are available for decoding.
3450 : : */
3451 : 9 : XLogFlush(GetXLogInsertEndRecPtr());
3452 : 9 : end_of_wal = GetFlushRecPtr(NULL);
3453 : :
3454 : : /*
3455 : : * Apply the concurrent changes again. Indicate that the decoding worker
3456 : : * won't be needed anymore.
3457 : : */
3458 : 9 : process_concurrent_changes(end_of_wal, &chgcxt, true);
3459 : :
3460 : : /* Remember info about rel before closing OldHeap */
3461 : 9 : relpersistence = OldHeap->rd_rel->relpersistence;
3462 : 9 : is_system_catalog = IsSystemRelation(OldHeap);
3463 : :
3464 : 9 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
3465 : : PROGRESS_REPACK_PHASE_SWAP_REL_FILES);
3466 : :
3467 : : /*
3468 : : * Even ShareUpdateExclusiveLock should have prevented others from
3469 : : * creating / dropping indexes (even using the CONCURRENTLY option), so we
3470 : : * do not need to check whether the lists match.
3471 : : */
3472 [ + - + + : 19 : forboth(lc, ind_oids_old, lc2, ind_oids_new)
+ - + + +
+ + - +
+ ]
3473 : : {
3474 : 10 : Oid ind_old = lfirst_oid(lc);
3475 : 10 : Oid ind_new = lfirst_oid(lc2);
3476 : 10 : Oid mapped_tables[4] = {0};
3477 : :
3478 : 10 : swap_relation_files(ind_old, ind_new,
3479 : : (old_table_oid == RelationRelationId),
3480 : : false, /* swap_toast_by_content */
3481 : : true,
3482 : : InvalidTransactionId,
3483 : : InvalidMultiXactId,
3484 : : mapped_tables);
3485 : :
3486 : : #ifdef USE_ASSERT_CHECKING
3487 : :
3488 : : /*
3489 : : * Concurrent processing is not supported for system relations, so
3490 : : * there should be no mapped tables.
3491 : : */
3492 [ + + ]: 50 : for (int i = 0; i < 4; i++)
3493 [ - + ]: 40 : Assert(!OidIsValid(mapped_tables[i]));
3494 : : #endif
3495 : : }
3496 : :
3497 : : /* The new indexes must be visible for deletion. */
3498 : 9 : CommandCounterIncrement();
3499 : :
3500 : : /* Close the old heap but keep lock until transaction commit. */
3501 : 9 : table_close(OldHeap, NoLock);
3502 : : /* Close the new heap. (We didn't have to open its indexes). */
3503 : 9 : table_close(NewHeap, NoLock);
3504 : :
3505 : : /* Cleanup what we don't need anymore. (And close the identity index.) */
3506 : 9 : release_change_context(&chgcxt);
3507 : :
3508 : : /*
3509 : : * Swap the relations and their TOAST relations and TOAST indexes. This
3510 : : * also drops the new relation and its indexes.
3511 : : *
3512 : : * (System catalogs are currently not supported.)
3513 : : */
3514 [ - + ]: 9 : Assert(!is_system_catalog);
3515 : 9 : finish_heap_swap(old_table_oid, new_table_oid,
3516 : : is_system_catalog,
3517 : : false, /* swap_toast_by_content */
3518 : : false,
3519 : : true,
3520 : : false, /* reindex */
3521 : : frozenXid, cutoffMulti,
3522 : : relpersistence);
3523 : 9 : }
3524 : :
3525 : : /*
3526 : : * Build indexes on NewHeap according to those on OldHeap.
3527 : : *
3528 : : * OldIndexes is the list of index OIDs on OldHeap. The contained indexes end
3529 : : * up locked using ShareUpdateExclusiveLock.
3530 : : *
3531 : : * A list of OIDs of the corresponding indexes created on NewHeap is
3532 : : * returned. The order of items does match, so we can use these arrays to swap
3533 : : * index storage.
3534 : : */
3535 : : static List *
3536 : 9 : build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
3537 : : {
3538 : 9 : List *result = NIL;
3539 : :
3540 : 9 : pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
3541 : : PROGRESS_REPACK_PHASE_REBUILD_INDEX);
3542 : :
3543 [ + - + + : 28 : foreach_oid(oldindex, OldIndexes)
+ + ]
3544 : : {
3545 : : Oid newindex;
3546 : : char *newName;
3547 : : Relation ind;
3548 : :
3549 : 10 : ind = index_open(oldindex, ShareUpdateExclusiveLock);
3550 : :
3551 : 10 : newName = ChooseRelationName(get_rel_name(oldindex),
3552 : : NULL,
3553 : : "repacknew",
3554 : 10 : get_rel_namespace(ind->rd_index->indrelid),
3555 : : false);
3556 : 10 : newindex = index_create_copy(NewHeap, INDEX_CREATE_SUPPRESS_PROGRESS,
3557 : 10 : oldindex, ind->rd_rel->reltablespace,
3558 : : newName);
3559 : 10 : copy_index_constraints(ind, newindex, RelationGetRelid(NewHeap));
3560 : 10 : result = lappend_oid(result, newindex);
3561 : :
3562 : 10 : index_close(ind, NoLock);
3563 : :
4 fujii@postgresql.org 3564 : 10 : pgstat_progress_incr_param(PROGRESS_REPACK_INDEX_REBUILD_COUNT, 1);
3565 : : }
3566 : :
167 alvherre@kurilemu.de 3567 : 9 : return result;
3568 : : }
3569 : :
3570 : : /*
3571 : : * Create a transient copy of a constraint -- supported by a transient
3572 : : * copy of the index that supports the original constraint.
3573 : : *
3574 : : * When repacking a table that contains exclusion constraints, the executor
3575 : : * relies on these constraints being properly catalogued. These copies are
3576 : : * to support that.
3577 : : *
3578 : : * We don't need the constraints for anything else (the original constraints
3579 : : * will be there once repack completes), so we add pg_depend entries so that
3580 : : * they are dropped when the transient table is dropped.
3581 : : */
3582 : : static void
3583 : 10 : copy_index_constraints(Relation old_index, Oid new_index_id, Oid new_heap_id)
3584 : : {
3585 : : ScanKeyData skey;
3586 : : Relation rel;
3587 : : TupleDesc desc;
3588 : : SysScanDesc scan;
3589 : : HeapTuple tup;
3590 : : ObjectAddress objrel;
3591 : :
3592 : 10 : rel = table_open(ConstraintRelationId, RowExclusiveLock);
3593 : 10 : ObjectAddressSet(objrel, RelationRelationId, new_heap_id);
3594 : :
3595 : : /*
3596 : : * Retrieve the constraints supported by the old index and create an
3597 : : * identical one that points to the new index.
3598 : : */
3599 : 10 : ScanKeyInit(&skey,
3600 : : Anum_pg_constraint_conrelid,
3601 : : BTEqualStrategyNumber, F_OIDEQ,
3602 : 10 : ObjectIdGetDatum(old_index->rd_index->indrelid));
3603 : 10 : scan = systable_beginscan(rel, ConstraintRelidTypidNameIndexId, true,
3604 : : NULL, 1, &skey);
3605 : 10 : desc = RelationGetDescr(rel);
3606 [ + + ]: 32 : while (HeapTupleIsValid(tup = systable_getnext(scan)))
3607 : : {
3608 : 22 : Form_pg_constraint conform = (Form_pg_constraint) GETSTRUCT(tup);
3609 : : Oid oid;
3610 : 22 : Datum values[Natts_pg_constraint] = {0};
3611 : 22 : bool nulls[Natts_pg_constraint] = {0};
3612 : 22 : bool replaces[Natts_pg_constraint] = {0};
3613 : : HeapTuple new_tup;
3614 : : ObjectAddress objcon;
3615 : :
3616 [ + + ]: 22 : if (conform->conindid != RelationGetRelid(old_index))
3617 : 13 : continue;
3618 : :
3619 : 9 : oid = GetNewOidWithIndex(rel, ConstraintOidIndexId,
3620 : : Anum_pg_constraint_oid);
3621 : 9 : values[Anum_pg_constraint_oid - 1] = ObjectIdGetDatum(oid);
3622 : 9 : replaces[Anum_pg_constraint_oid - 1] = true;
3623 : 9 : values[Anum_pg_constraint_conrelid - 1] = ObjectIdGetDatum(new_heap_id);
3624 : 9 : replaces[Anum_pg_constraint_conrelid - 1] = true;
3625 : 9 : values[Anum_pg_constraint_conindid - 1] = ObjectIdGetDatum(new_index_id);
3626 : 9 : replaces[Anum_pg_constraint_conindid - 1] = true;
3627 : :
3628 : 9 : new_tup = heap_modify_tuple(tup, desc, values, nulls, replaces);
3629 : :
3630 : : /* Insert it into the catalog. */
3631 : 9 : CatalogTupleInsert(rel, new_tup);
3632 : :
3633 : : /* Create a dependency so it's removed when we drop the new heap. */
3634 : 9 : ObjectAddressSet(objcon, ConstraintRelationId, oid);
3635 : 9 : recordDependencyOn(&objcon, &objrel, DEPENDENCY_AUTO);
3636 : : }
3637 : 10 : systable_endscan(scan);
3638 : :
3639 : 10 : table_close(rel, RowExclusiveLock);
3640 : :
3641 : 10 : CommandCounterIncrement();
3642 : 10 : }
3643 : :
3644 : : /*
3645 : : * Create a transient copy of attribute defaults.
3646 : : *
3647 : : * When repacking a table that has stored generated columns, the executor
3648 : : * relies on these entries to generate the values for them during apply of
3649 : : * concurrent operations. These copies are there to support that.
3650 : : *
3651 : : * We don't need the defaults for anything else, so we add pg_depend entries
3652 : : * so that they are dropped when the transient table is dropped.
3653 : : */
3654 : : static void
79 3655 : 9 : copy_attribute_defaults(Oid old_heap_oid, Oid new_heap_oid)
3656 : : {
3657 : : ScanKeyData skey;
3658 : : Relation rel;
3659 : : Relation att_rel;
3660 : : SysScanDesc scan;
3661 : : HeapTuple def_tup;
3662 : : ObjectAddress objrel;
3663 : :
3664 : 9 : rel = table_open(AttrDefaultRelationId, RowExclusiveLock);
3665 : 9 : att_rel = table_open(AttributeRelationId, RowExclusiveLock);
3666 : :
3667 : 9 : ObjectAddressSet(objrel, RelationRelationId, new_heap_oid);
3668 : :
3669 : 9 : ScanKeyInit(&skey,
3670 : : Anum_pg_attrdef_adrelid,
3671 : : BTEqualStrategyNumber, F_OIDEQ,
3672 : : ObjectIdGetDatum(old_heap_oid));
3673 : 9 : scan = systable_beginscan(rel, AttrDefaultIndexId, true,
3674 : : NULL, 1, &skey);
3675 [ + + ]: 11 : while (HeapTupleIsValid(def_tup = systable_getnext(scan)))
3676 : : {
3677 : : Form_pg_attrdef adform;
3678 : : Oid oid;
3679 : : Datum def_values[Natts_pg_attrdef];
3680 : : bool def_nulls[Natts_pg_attrdef];
3681 : 2 : bool def_replaces[Natts_pg_attrdef] = {0};
3682 : : Datum att_values[Natts_pg_attribute];
3683 : : bool att_nulls[Natts_pg_attribute];
3684 : 2 : bool att_replaces[Natts_pg_attribute] = {0};
3685 : : HeapTuple new_def_tup,
3686 : : att_tup,
3687 : : new_att_tup;
3688 : : ObjectAddress objad;
3689 : :
3690 : 2 : adform = (Form_pg_attrdef) GETSTRUCT(def_tup);
3691 [ - + ]: 2 : Assert(adform->adrelid == old_heap_oid);
3692 : :
3693 : : /*
3694 : : * Insert a new tuple that's identical to the existing one, other than
3695 : : * its OID and the relation it refers to.
3696 : : */
3697 : 2 : oid = GetNewOidWithIndex(rel, AttrDefaultOidIndexId,
3698 : : Anum_pg_attrdef_oid);
3699 : 2 : def_values[Anum_pg_attrdef_oid - 1] = ObjectIdGetDatum(oid);
3700 : 2 : def_nulls[Anum_pg_attrdef_oid - 1] = false;
3701 : 2 : def_replaces[Anum_pg_attrdef_oid - 1] = true;
3702 : 2 : def_values[Anum_pg_attrdef_adrelid - 1] = ObjectIdGetDatum(new_heap_oid);
3703 : 2 : def_nulls[Anum_pg_attrdef_adrelid - 1] = false;
3704 : 2 : def_replaces[Anum_pg_attrdef_adrelid - 1] = true;
3705 : 2 : new_def_tup = heap_modify_tuple(def_tup, RelationGetDescr(rel),
3706 : : def_values, def_nulls, def_replaces);
3707 : 2 : CatalogTupleInsert(rel, new_def_tup);
3708 : :
3709 : : /* Set atthasdef for this attribute in the transient table */
3710 : 2 : att_tup = SearchSysCache2(ATTNUM,
3711 : : ObjectIdGetDatum(new_heap_oid),
3712 : 2 : ObjectIdGetDatum(adform->adnum));
3713 [ - + ]: 2 : if (!HeapTupleIsValid(att_tup))
79 alvherre@kurilemu.de 3714 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
3715 : : adform->adnum, new_heap_oid);
79 alvherre@kurilemu.de 3716 :CBC 2 : att_values[Anum_pg_attribute_atthasdef - 1] = BoolGetDatum(true);
3717 : 2 : att_nulls[Anum_pg_attribute_atthasdef - 1] = false;
3718 : 2 : att_replaces[Anum_pg_attribute_atthasdef - 1] = true;
3719 : 2 : new_att_tup = heap_modify_tuple(att_tup, RelationGetDescr(att_rel),
3720 : : att_values, att_nulls, att_replaces);
3721 : 2 : CatalogTupleUpdate(att_rel, &new_att_tup->t_self, new_att_tup);
3722 : 2 : ReleaseSysCache(att_tup);
3723 : :
3724 : : /* Add a pg_depend record so it's removed with the transient table */
3725 : 2 : ObjectAddressSet(objad, AttrDefaultRelationId, oid);
3726 : 2 : recordDependencyOn(&objad, &objrel, DEPENDENCY_AUTO);
3727 : : }
3728 : 9 : systable_endscan(scan);
3729 : :
3730 : 9 : table_close(rel, RowExclusiveLock);
3731 : 9 : table_close(att_rel, RowExclusiveLock);
3732 : :
3733 : 9 : CommandCounterIncrement();
3734 : 9 : }
3735 : :
3736 : : /*
3737 : : * Try to start a background worker to perform logical decoding of data
3738 : : * changes applied to relation while REPACK CONCURRENTLY is copying its
3739 : : * contents to a new table.
3740 : : */
3741 : : static void
167 3742 : 9 : start_repack_decoding_worker(Oid relid)
3743 : : {
3744 : : Size size;
3745 : : DecodingWorkerShared *shared;
3746 : : shm_mq *mq;
3747 : : BackgroundWorker bgw;
3748 : :
117 3749 : 9 : decoding_worker = palloc0_object(DecodingWorker);
3750 : :
3751 : : /* Setup shared memory. */
167 3752 : 9 : size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
3753 : : BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
117 3754 : 9 : decoding_worker->seg = dsm_create(size, 0);
3755 : :
3756 : 9 : shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
157 3757 : 9 : shared->initialized = false;
167 3758 : 9 : shared->lsn_upto = InvalidXLogRecPtr;
3759 : 9 : shared->done = false;
117 3760 : 9 : SharedFileSetInit(&shared->sfs, decoding_worker->seg);
167 3761 : 9 : shared->last_exported = -1;
3762 : 9 : SpinLockInit(&shared->mutex);
3763 : 9 : shared->dbid = MyDatabaseId;
3764 : :
3765 : : /*
3766 : : * This is the UserId set in cluster_rel(). Security context shouldn't be
3767 : : * needed for decoding worker.
3768 : : */
3769 : 9 : shared->roleid = GetUserId();
3770 : 9 : shared->relid = relid;
3771 : 9 : ConditionVariableInit(&shared->cv);
3772 : 9 : shared->backend_pid = MyProcPid;
3773 : 9 : shared->backend_proc_number = MyProcNumber;
3774 : :
3775 : : /* Transmit our timeouts to the worker too */
3 3776 : 9 : shared->lock_timeout = LockTimeout;
3777 : 9 : shared->transaction_timeout = TransactionTimeout;
3778 : :
167 3779 : 9 : mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
3780 : : REPACK_ERROR_QUEUE_SIZE);
3781 : 9 : shm_mq_set_receiver(mq, MyProc);
3782 : :
117 3783 : 9 : decoding_worker->error_mqh = shm_mq_attach(mq, decoding_worker->seg, NULL);
3784 : :
167 3785 : 9 : memset(&bgw, 0, sizeof(bgw));
3786 : 9 : snprintf(bgw.bgw_name, BGW_MAXLEN,
3787 : : "REPACK decoding worker for relation \"%s\"",
3788 : : get_rel_name(relid));
3789 : 9 : snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
3790 : 9 : bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
3791 : : BGWORKER_BACKEND_DATABASE_CONNECTION;
3792 : 9 : bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
3793 : 9 : bgw.bgw_restart_time = BGW_NEVER_RESTART;
3794 : 9 : snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
3795 : 9 : snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
117 3796 : 9 : bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(decoding_worker->seg));
167 3797 : 9 : bgw.bgw_notify_pid = MyProcPid;
3798 : :
3799 [ - + ]: 9 : if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
167 alvherre@kurilemu.de 3800 [ # # ]:UBC 0 : ereport(ERROR,
3801 : : errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
3802 : : errmsg("out of background worker slots"),
3803 : : errhint("You might need to increase \"%s\".", "max_worker_processes"));
3804 : :
3805 : : /*
3806 : : * Now that the worker is registered, connect the error message queue to
3807 : : * it.
3808 : : */
3 alvherre@kurilemu.de 3809 :CBC 9 : shm_mq_set_handle(decoding_worker->error_mqh, decoding_worker->handle);
3810 : :
3811 : : /*
3812 : : * Make sure the worker has started before we wait for it to initialize
3813 : : * decoding below, so that the failure-to-start case does not hang
3814 : : * forever.
3815 : : */
3816 : 9 : wait_for_repack_decoding_worker();
3817 : :
3818 : : /*
3819 : : * The decoding setup must be done before the caller can have XID assigned
3820 : : * for any reason, otherwise the worker might end up in a deadlock,
3821 : : * waiting for the caller's transaction to end. Therefore wait here until
3822 : : * the worker indicates that it has the logical decoding initialized.
3823 : : */
167 3824 : 9 : ConditionVariablePrepareToSleep(&shared->cv);
3825 : : for (;;)
3826 : 19 : {
3827 : : bool initialized;
3828 : :
3829 : 28 : SpinLockAcquire(&shared->mutex);
3830 : 28 : initialized = shared->initialized;
3831 : 28 : SpinLockRelease(&shared->mutex);
3832 : :
3833 [ + + ]: 28 : if (initialized)
3834 : 9 : break;
3835 : :
3836 : 19 : ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
3837 : : }
3838 : 9 : ConditionVariableCancelSleep();
3839 : 9 : }
3840 : :
3841 : : /*
3842 : : * Wait for the decoding worker to start up, and throw an error if it fails
3843 : : * to do so.
3844 : : *
3845 : : * This is similar to WaitForParallelWorkersToAttach(). The only reliable way
3846 : : * to tell a worker that failed to start (fork failure, or an exit before it
3847 : : * attached) from one that is merely slow is to check whether it became the
3848 : : * sender on the error message queue. If it stopped without attaching, nothing
3849 : : * was queued and we report the generic failure ourselves. If it attached, any
3850 : : * error it reported is in the queue and is thrown when we process pending
3851 : : * messages, either here or later while we wait for it to initialize decoding.
3852 : : */
3853 : : static void
3 3854 : 9 : wait_for_repack_decoding_worker(void)
3855 : : {
3856 : : for (;;)
3857 : 18 : {
3858 : : BgwHandleStatus status;
3859 : : shm_mq *mq;
3860 : : int rc;
3861 : : pid_t pid;
3862 : :
3863 : : /*
3864 : : * This will process any repack messages that are pending and it may
3865 : : * also throw an error propagated from a worker.
3866 : : */
3867 [ - + ]: 27 : CHECK_FOR_INTERRUPTS();
3868 : :
3869 : : /* If error_mqh is NULL, the worker has exited cleanly */
3870 [ - + ]: 27 : if (decoding_worker->error_mqh == NULL)
3 alvherre@kurilemu.de 3871 :UBC 0 : break;
3872 : :
3 alvherre@kurilemu.de 3873 :CBC 27 : status = GetBackgroundWorkerPid(decoding_worker->handle, &pid);
3874 [ + + ]: 27 : if (status == BGWH_STARTED)
3875 : : {
3876 : : /* Has the worker attached to the error message queue? */
3877 : 20 : mq = shm_mq_get_queue(decoding_worker->error_mqh);
3878 [ + + ]: 20 : if (shm_mq_get_sender(mq) != NULL)
3879 : 9 : break;
3880 : : }
3881 [ - + ]: 7 : else if (status == BGWH_STOPPED)
3882 : : {
3883 : : /*
3884 : : * If the worker stopped without attaching to the error message
3885 : : * queue, throw an error. Otherwise, assume it attached and
3886 : : * reported an error before exiting, so mark it attached and let
3887 : : * the next attempt to process pending messages, here or later
3888 : : * while the initial snapshot is set up, throw that error.
3889 : : */
3 alvherre@kurilemu.de 3890 :UBC 0 : mq = shm_mq_get_queue(decoding_worker->error_mqh);
3891 [ # # ]: 0 : if (shm_mq_get_sender(mq) == NULL)
3892 [ # # ]: 0 : ereport(ERROR,
3893 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3894 : : errmsg("REPACK decoding worker failed to start"),
3895 : : errhint("More details may be available in the server log."));
3896 : 0 : break;
3897 : : }
3898 : :
3899 : : /* Worker neither started or stopped yet, so wait. */
3 alvherre@kurilemu.de 3900 :CBC 18 : rc = WaitLatch(MyLatch,
3901 : : WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
3902 : : -1, WAIT_EVENT_BGWORKER_STARTUP);
3903 : :
3904 [ + - ]: 18 : if (rc & WL_LATCH_SET)
3905 : 18 : ResetLatch(MyLatch);
3906 : : }
3907 : 9 : }
3908 : :
3909 : : /*
3910 : : * Stop the decoding worker and cleanup the related resources.
3911 : : *
3912 : : * The worker stops on its own when it knows there is no more work to do, but
3913 : : * we need to stop it explicitly at least on ERROR in the launching backend.
3914 : : */
3915 : : static void
167 3916 : 9 : stop_repack_decoding_worker(void)
3917 : : {
3918 : : /* Nothing to do if no worker was set up. */
3919 [ - + ]: 9 : if (decoding_worker == NULL)
167 alvherre@kurilemu.de 3920 :UBC 0 : return;
3921 : :
3922 : : /* Terminate the decoding worker, if one is running */
3 alvherre@kurilemu.de 3923 [ + - ]:CBC 9 : if (decoding_worker->handle != NULL)
3924 : 9 : TerminateBackgroundWorker(decoding_worker->handle);
3925 : :
3926 : : /*
3927 : : * The error queue should have already been nulled out during worker
3928 : : * shutdown, but if that didn't happen, do it now.
3929 : : */
3930 [ - + ]: 9 : if (decoding_worker->error_mqh != NULL)
3931 : : {
3 alvherre@kurilemu.de 3932 :UBC 0 : shm_mq_detach(decoding_worker->error_mqh);
3933 : 0 : decoding_worker->error_mqh = NULL;
3934 : : }
3935 : :
3936 : : /*
3937 : : * Cancel any sleep on the condition variable before detaching the shared
3938 : : * memory segment, because the CV lives in that segment. Otherwise later
3939 : : * cleanup would touch freed memory.
3940 : : */
3 alvherre@kurilemu.de 3941 :CBC 9 : ConditionVariableCancelSleep();
3942 : :
3943 : : /*
3944 : : * We can't finish the REPACK command until the worker has exited. This
3945 : : * means, in particular, that we can't respond to interrupts at this
3946 : : * stage.
3947 : : */
117 3948 [ + - ]: 9 : if (decoding_worker->handle != NULL)
3949 : : {
3950 : : BgwHandleStatus status;
3951 : :
3952 : 9 : HOLD_INTERRUPTS();
3953 : 9 : status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
3954 [ - + ]: 9 : RESUME_INTERRUPTS();
3955 : :
3956 [ - + ]: 9 : if (status == BGWH_POSTMASTER_DIED)
117 alvherre@kurilemu.de 3957 [ # # ]:UBC 0 : ereport(FATAL,
3958 : : errcode(ERRCODE_ADMIN_SHUTDOWN),
3959 : : errmsg("postmaster exited during REPACK command"));
3960 : : }
3961 : :
3962 : : /*
3963 : : * Detach from the shared memory segment only now that the worker is gone.
3964 : : * The worker attaches to the shared file set after it maps the segment,
3965 : : * so detaching any earlier can destroy the file set under a worker that
3966 : : * is still starting up.
3967 : : */
117 alvherre@kurilemu.de 3968 [ + - ]:CBC 9 : if (decoding_worker->seg != NULL)
3969 : : {
3970 : 9 : dsm_detach(decoding_worker->seg);
3 3971 : 9 : decoding_worker->seg = NULL;
3972 : : }
3973 : :
167 3974 : 9 : pfree(decoding_worker);
3975 : 9 : decoding_worker = NULL;
3976 : : }
3977 : :
3978 : : /* stop_repack_decoding_worker, wrapped as a before_shmem_exit callback */
3979 : : static void
124 alvherre@kurilemu.de 3980 :UBC 0 : stop_repack_decoding_worker_cb(int code, Datum arg)
3981 : : {
3982 : 0 : stop_repack_decoding_worker();
3983 : 0 : }
3984 : :
3985 : : /*
3986 : : * Get the initial snapshot from the decoding worker.
3987 : : */
3988 : : static Snapshot
167 alvherre@kurilemu.de 3989 :CBC 9 : get_initial_snapshot(DecodingWorker *worker)
3990 : : {
3991 : : DecodingWorkerShared *shared;
3992 : : char fname[MAXPGPATH];
3993 : : BufFile *file;
3994 : : Size snap_size;
3995 : : char *snap_space;
3996 : : Snapshot snapshot;
3997 : :
3998 : 9 : shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
3999 : :
4000 : : /*
4001 : : * The worker needs to initialize the logical decoding, which usually
4002 : : * takes some time. Therefore it makes sense to prepare for the sleep
4003 : : * first.
4004 : : */
4005 : 9 : ConditionVariablePrepareToSleep(&shared->cv);
4006 : : for (;;)
4007 : 8 : {
4008 : : int last_exported;
4009 : :
4010 : 17 : SpinLockAcquire(&shared->mutex);
4011 : 17 : last_exported = shared->last_exported;
4012 : 17 : SpinLockRelease(&shared->mutex);
4013 : :
4014 : : /*
4015 : : * Has the worker exported the file we are waiting for?
4016 : : */
4017 [ + + ]: 17 : if (last_exported == WORKER_FILE_SNAPSHOT)
4018 : 9 : break;
4019 : :
4020 : 8 : ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
4021 : : }
4022 : 9 : ConditionVariableCancelSleep();
4023 : :
4024 : : /* Read the snapshot from a file. */
4025 : 9 : DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
4026 : 9 : file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
4027 : 9 : BufFileReadExact(file, &snap_size, sizeof(snap_size));
4028 : 9 : snap_space = (char *) palloc(snap_size);
4029 : 9 : BufFileReadExact(file, snap_space, snap_size);
4030 : 9 : BufFileClose(file);
4031 : :
4032 : : /* Restore it. */
4033 : 9 : snapshot = RestoreSnapshot(snap_space);
4034 : 9 : pfree(snap_space);
4035 : :
4036 : 9 : return snapshot;
4037 : : }
4038 : :
4039 : : /*
4040 : : * Generate worker's file name into 'fname', which must be of size MAXPGPATH.
4041 : : * If relations of the same 'relid' happen to be processed at the same time,
4042 : : * they must be from different databases and therefore different backends must
4043 : : * be involved.
4044 : : */
4045 : : void
4046 : 54 : DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
4047 : : {
4048 : : /* The PID is already present in the fileset name, so we needn't add it */
4049 : 54 : snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
4050 : 54 : }
4051 : :
4052 : : /*
4053 : : * Handle receipt of an interrupt indicating a repack worker message.
4054 : : *
4055 : : * Note: this is called within a signal handler! All we can do is set
4056 : : * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
4057 : : * ProcessRepackMessages().
4058 : : */
4059 : : void
4060 : 10 : HandleRepackMessageInterrupt(void)
4061 : : {
4062 : 10 : InterruptPending = true;
4063 : 10 : RepackMessagePending = true;
4064 : 10 : SetLatch(MyLatch);
4065 : 10 : }
4066 : :
4067 : : /*
4068 : : * Process any queued protocol messages received from the repack worker.
4069 : : */
4070 : : void
4071 : 9 : ProcessRepackMessages(void)
4072 : : {
4073 : : MemoryContext oldcontext;
4074 : : static MemoryContext hpm_context = NULL;
4075 : :
4076 : : /*
4077 : : * Nothing to do if we haven't launched the worker yet or have already
4078 : : * terminated it. Stopping the worker detaches the error message queue
4079 : : * before clearing decoding_worker, so also bail out once error_mqh is
4080 : : * gone.
4081 : : */
3 4082 [ + - - + ]: 9 : if (decoding_worker == NULL || decoding_worker->error_mqh == NULL)
167 alvherre@kurilemu.de 4083 :UBC 0 : return;
4084 : :
4085 : : /*
4086 : : * This is invoked from ProcessInterrupts(), and since some of the
4087 : : * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
4088 : : * for recursive calls if more signals are received while this runs. It's
4089 : : * unclear that recursive entry would be safe, and it doesn't seem useful
4090 : : * even if it is safe, so let's block interrupts until done.
4091 : : */
167 alvherre@kurilemu.de 4092 :CBC 9 : HOLD_INTERRUPTS();
4093 : :
4094 : : /*
4095 : : * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
4096 : : * don't want to risk leaking data into long-lived contexts, so let's do
4097 : : * our work here in a private context that we can reset on each use.
4098 : : */
4099 [ + + ]: 9 : if (hpm_context == NULL) /* first time through? */
4100 : 7 : hpm_context = AllocSetContextCreate(TopMemoryContext,
4101 : : "ProcessRepackMessages",
4102 : : ALLOCSET_DEFAULT_SIZES);
4103 : : else
4104 : 2 : MemoryContextReset(hpm_context);
4105 : :
4106 : 9 : oldcontext = MemoryContextSwitchTo(hpm_context);
4107 : :
4108 : : /* OK to process messages. Reset the flag saying there are more to do. */
4109 : 9 : RepackMessagePending = false;
4110 : :
4111 : : /*
4112 : : * Read messages from the worker, but stop if the error queue disappears,
4113 : : * which happens when a PqRepackMsg_Terminate is received; or as soon as
4114 : : * no more messages can be read without blocking. Messages are
4115 : : * infrequent, so no point optimizing stringinfo allocation.
4116 : : */
3 4117 [ + + ]: 18 : while (decoding_worker->error_mqh != NULL)
4118 : : {
4119 : : shm_mq_result res;
4120 : : Size nbytes;
4121 : : void *data;
4122 : : StringInfoData msg;
4123 : :
167 4124 : 9 : res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
4125 : : &data, true);
3 4126 [ + - - - ]: 9 : switch (res)
4127 : : {
4128 : 9 : case SHM_MQ_SUCCESS:
4129 : 9 : initStringInfo(&msg);
4130 : 9 : appendBinaryStringInfo(&msg, data, nbytes);
4131 : 9 : ProcessRepackMessage(&msg);
4132 : 9 : pfree(msg.data);
4133 : 9 : break;
4134 : :
3 alvherre@kurilemu.de 4135 :UBC 0 : case SHM_MQ_WOULD_BLOCK:
4136 : : /* no more messages to process for now */
4137 : 0 : goto out;
4138 : :
4139 : 0 : case SHM_MQ_DETACHED:
4140 : :
4141 : : /*
4142 : : * Normal worker stop sends a Terminate message, causing
4143 : : * ProcessRepackMessage to set error_mqh to NULL, thus exiting
4144 : : * this loop; so this case should never happen. If it does,
4145 : : * the worker terminated abnormally, so report that.
4146 : : */
4147 [ # # ]: 0 : ereport(ERROR,
4148 : : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4149 : : errmsg("lost connection to REPACK decoding worker"),
4150 : : errhint("More details may be available in the server log."));
4151 : : }
4152 : : }
4153 : :
3 alvherre@kurilemu.de 4154 :CBC 9 : out:
167 4155 : 9 : MemoryContextSwitchTo(oldcontext);
4156 : :
4157 : : /* Might as well clear the context on our way out */
4158 : 9 : MemoryContextReset(hpm_context);
4159 : :
4160 [ - + ]: 9 : RESUME_INTERRUPTS();
4161 : : }
4162 : :
4163 : : /*
4164 : : * Process a single protocol message received from a repack worker.
4165 : : */
4166 : : static void
4167 : 9 : ProcessRepackMessage(StringInfo msg)
4168 : : {
4169 : : char msgtype;
4170 : :
4171 : 9 : msgtype = pq_getmsgbyte(msg);
4172 : :
4173 [ - + - ]: 9 : switch (msgtype)
4174 : : {
167 alvherre@kurilemu.de 4175 :UBC 0 : case PqMsg_ErrorResponse:
4176 : : case PqMsg_NoticeResponse:
4177 : : {
4178 : : ErrorData edata;
4179 : :
4180 : : /* Parse ErrorResponse or NoticeResponse. */
4181 : 0 : pq_parse_errornotice(msg, &edata);
4182 : :
4183 : : /* Death of a worker isn't enough justification for suicide. */
4184 : 0 : edata.elevel = Min(edata.elevel, ERROR);
4185 : :
4186 : : /*
4187 : : * Add a context line to show that this is a message
4188 : : * propagated from the worker. Otherwise, it can sometimes be
4189 : : * confusing to understand what actually happened.
4190 : : */
4191 [ # # ]: 0 : if (edata.context)
4192 : 0 : edata.context = psprintf("%s\n%s", edata.context,
4193 : : _("REPACK decoding worker"));
4194 : : else
4195 : 0 : edata.context = pstrdup(_("REPACK decoding worker"));
4196 : :
4197 : : /* Rethrow error or print notice. */
4198 : 0 : ThrowErrorData(&edata);
4199 : :
4200 : 0 : break;
4201 : : }
4202 : :
3 alvherre@kurilemu.de 4203 :CBC 9 : case PqRepackMsg_Terminate:
4204 : :
4205 : : /*
4206 : : * The worker has completed its work; stop watching its message
4207 : : * queue now for orderly shutdown.
4208 : : */
4209 : 9 : shm_mq_detach(decoding_worker->error_mqh);
4210 : 9 : decoding_worker->error_mqh = NULL;
4211 : 9 : break;
4212 : :
167 alvherre@kurilemu.de 4213 :UBC 0 : default:
3 4214 [ # # ]: 0 : elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
4215 : : msgtype, msg->len);
4216 : : }
167 alvherre@kurilemu.de 4217 :CBC 9 : }
|
