Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * createas.c
4 : * Execution of CREATE TABLE ... AS, a/k/a SELECT INTO.
5 : * Since CREATE MATERIALIZED VIEW shares syntax and most behaviors,
6 : * we implement that here, too.
7 : *
8 : * We implement this by diverting the query's normal output to a
9 : * specialized DestReceiver type.
10 : *
11 : * Formerly, CTAS was implemented as a variant of SELECT, which led
12 : * to assorted legacy behaviors that we still try to preserve, notably that
13 : * we must return a tuples-processed count in the QueryCompletion. (We no
14 : * longer do that for CTAS ... WITH NO DATA, however.)
15 : *
16 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
17 : * Portions Copyright (c) 1994, Regents of the University of California
18 : *
19 : *
20 : * IDENTIFICATION
21 : * src/backend/commands/createas.c
22 : *
23 : *-------------------------------------------------------------------------
24 : */
25 : #include "postgres.h"
26 :
27 : #include "access/heapam.h"
28 : #include "access/reloptions.h"
29 : #include "access/tableam.h"
30 : #include "access/xact.h"
31 : #include "catalog/namespace.h"
32 : #include "catalog/toasting.h"
33 : #include "commands/createas.h"
34 : #include "commands/matview.h"
35 : #include "commands/prepare.h"
36 : #include "commands/tablecmds.h"
37 : #include "commands/view.h"
38 : #include "miscadmin.h"
39 : #include "nodes/makefuncs.h"
40 : #include "nodes/nodeFuncs.h"
41 : #include "rewrite/rewriteHandler.h"
42 : #include "tcop/tcopprot.h"
43 : #include "utils/builtins.h"
44 : #include "utils/lsyscache.h"
45 : #include "utils/rel.h"
46 : #include "utils/rls.h"
47 : #include "utils/snapmgr.h"
48 :
49 : typedef struct
50 : {
51 : DestReceiver pub; /* publicly-known function pointers */
52 : IntoClause *into; /* target relation specification */
53 : /* These fields are filled by intorel_startup: */
54 : Relation rel; /* relation to write to */
55 : ObjectAddress reladdr; /* address of rel, for ExecCreateTableAs */
56 : CommandId output_cid; /* cmin to insert in output tuples */
57 : int ti_options; /* table_tuple_insert performance options */
58 : BulkInsertState bistate; /* bulk insert state */
59 : } DR_intorel;
60 :
61 : /* utility functions for CTAS definition creation */
62 : static ObjectAddress create_ctas_internal(List *attrList, IntoClause *into);
63 : static ObjectAddress create_ctas_nodata(List *tlist, IntoClause *into);
64 :
65 : /* DestReceiver routines for collecting data */
66 : static void intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo);
67 : static bool intorel_receive(TupleTableSlot *slot, DestReceiver *self);
68 : static void intorel_shutdown(DestReceiver *self);
69 : static void intorel_destroy(DestReceiver *self);
70 :
71 :
72 : /*
73 : * create_ctas_internal
74 : *
75 : * Internal utility used for the creation of the definition of a relation
76 : * created via CREATE TABLE AS or a materialized view. Caller needs to
77 : * provide a list of attributes (ColumnDef nodes).
78 : */
79 : static ObjectAddress
80 1604 : create_ctas_internal(List *attrList, IntoClause *into)
81 : {
82 1604 : CreateStmt *create = makeNode(CreateStmt);
83 : bool is_matview;
84 : char relkind;
85 : Datum toast_options;
86 : static char *validnsps[] = HEAP_RELOPT_NAMESPACES;
87 : ObjectAddress intoRelationAddr;
88 :
89 : /* This code supports both CREATE TABLE AS and CREATE MATERIALIZED VIEW */
90 1604 : is_matview = (into->viewQuery != NULL);
91 1604 : relkind = is_matview ? RELKIND_MATVIEW : RELKIND_RELATION;
92 :
93 : /*
94 : * Create the target relation by faking up a CREATE TABLE parsetree and
95 : * passing it to DefineRelation.
96 : */
97 1604 : create->relation = into->rel;
98 1604 : create->tableElts = attrList;
99 1604 : create->inhRelations = NIL;
100 1604 : create->ofTypename = NULL;
101 1604 : create->constraints = NIL;
102 1604 : create->options = into->options;
103 1604 : create->oncommit = into->onCommit;
104 1604 : create->tablespacename = into->tableSpaceName;
105 1604 : create->if_not_exists = false;
106 1604 : create->accessMethod = into->accessMethod;
107 :
108 : /*
109 : * Create the relation. (This will error out if there's an existing view,
110 : * so we don't need more code to complain if "replace" is false.)
111 : */
112 1604 : intoRelationAddr = DefineRelation(create, relkind, InvalidOid, NULL, NULL);
113 :
114 : /*
115 : * If necessary, create a TOAST table for the target table. Note that
116 : * NewRelationCreateToastTable ends with CommandCounterIncrement(), so
117 : * that the TOAST table will be visible for insertion.
118 : */
119 1586 : CommandCounterIncrement();
120 :
121 : /* parse and validate reloptions for the toast table */
122 1586 : toast_options = transformRelOptions((Datum) 0,
123 : create->options,
124 : "toast",
125 : validnsps,
126 : true, false);
127 :
128 1586 : (void) heap_reloptions(RELKIND_TOASTVALUE, toast_options, true);
129 :
130 1586 : NewRelationCreateToastTable(intoRelationAddr.objectId, toast_options);
131 :
132 : /* Create the "view" part of a materialized view. */
133 1586 : if (is_matview)
134 : {
135 : /* StoreViewQuery scribbles on tree, so make a copy */
136 438 : Query *query = (Query *) copyObject(into->viewQuery);
137 :
138 438 : StoreViewQuery(intoRelationAddr.objectId, query, false);
139 438 : CommandCounterIncrement();
140 : }
141 :
142 1586 : return intoRelationAddr;
143 : }
144 :
145 :
146 : /*
147 : * create_ctas_nodata
148 : *
149 : * Create CTAS or materialized view when WITH NO DATA is used, starting from
150 : * the targetlist of the SELECT or view definition.
151 : */
152 : static ObjectAddress
153 110 : create_ctas_nodata(List *tlist, IntoClause *into)
154 : {
155 : List *attrList;
156 : ListCell *t,
157 : *lc;
158 :
159 : /*
160 : * Build list of ColumnDefs from non-junk elements of the tlist. If a
161 : * column name list was specified in CREATE TABLE AS, override the column
162 : * names in the query. (Too few column names are OK, too many are not.)
163 : */
164 110 : attrList = NIL;
165 110 : lc = list_head(into->colNames);
166 352 : foreach(t, tlist)
167 : {
168 242 : TargetEntry *tle = (TargetEntry *) lfirst(t);
169 :
170 242 : if (!tle->resjunk)
171 : {
172 : ColumnDef *col;
173 : char *colname;
174 :
175 242 : if (lc)
176 : {
177 76 : colname = strVal(lfirst(lc));
178 76 : lc = lnext(into->colNames, lc);
179 : }
180 : else
181 166 : colname = tle->resname;
182 :
183 242 : col = makeColumnDef(colname,
184 242 : exprType((Node *) tle->expr),
185 242 : exprTypmod((Node *) tle->expr),
186 242 : exprCollation((Node *) tle->expr));
187 :
188 : /*
189 : * It's possible that the column is of a collatable type but the
190 : * collation could not be resolved, so double-check. (We must
191 : * check this here because DefineRelation would adopt the type's
192 : * default collation rather than complaining.)
193 : */
194 464 : if (!OidIsValid(col->collOid) &&
195 222 : type_is_collatable(col->typeName->typeOid))
196 0 : ereport(ERROR,
197 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
198 : errmsg("no collation was derived for column \"%s\" with collatable type %s",
199 : col->colname,
200 : format_type_be(col->typeName->typeOid)),
201 : errhint("Use the COLLATE clause to set the collation explicitly.")));
202 :
203 242 : attrList = lappend(attrList, col);
204 : }
205 : }
206 :
207 110 : if (lc != NULL)
208 12 : ereport(ERROR,
209 : (errcode(ERRCODE_SYNTAX_ERROR),
210 : errmsg("too many column names were specified")));
211 :
212 : /* Create the relation definition using the ColumnDef list */
213 98 : return create_ctas_internal(attrList, into);
214 : }
215 :
216 :
217 : /*
218 : * ExecCreateTableAs -- execute a CREATE TABLE AS command
219 : */
220 : ObjectAddress
221 1684 : ExecCreateTableAs(ParseState *pstate, CreateTableAsStmt *stmt,
222 : ParamListInfo params, QueryEnvironment *queryEnv,
223 : QueryCompletion *qc)
224 : {
225 1684 : Query *query = castNode(Query, stmt->query);
226 1684 : IntoClause *into = stmt->into;
227 1684 : bool is_matview = (into->viewQuery != NULL);
228 : DestReceiver *dest;
229 1684 : Oid save_userid = InvalidOid;
230 1684 : int save_sec_context = 0;
231 1684 : int save_nestlevel = 0;
232 : ObjectAddress address;
233 : List *rewritten;
234 : PlannedStmt *plan;
235 : QueryDesc *queryDesc;
236 :
237 : /* Check if the relation exists or not */
238 1684 : if (CreateTableAsRelExists(stmt))
239 46 : return InvalidObjectAddress;
240 :
241 : /*
242 : * Create the tuple receiver object and insert info it will need
243 : */
244 1592 : dest = CreateIntoRelDestReceiver(into);
245 :
246 : /*
247 : * The contained Query could be a SELECT, or an EXECUTE utility command.
248 : * If the latter, we just pass it off to ExecuteQuery.
249 : */
250 1592 : if (query->commandType == CMD_UTILITY &&
251 42 : IsA(query->utilityStmt, ExecuteStmt))
252 : {
253 42 : ExecuteStmt *estmt = castNode(ExecuteStmt, query->utilityStmt);
254 :
255 : Assert(!is_matview); /* excluded by syntax */
256 42 : ExecuteQuery(pstate, estmt, into, params, dest, qc);
257 :
258 : /* get object address that intorel_startup saved for us */
259 42 : address = ((DR_intorel *) dest)->reladdr;
260 :
261 42 : return address;
262 : }
263 : Assert(query->commandType == CMD_SELECT);
264 :
265 : /*
266 : * For materialized views, lock down security-restricted operations and
267 : * arrange to make GUC variable changes local to this command. This is
268 : * not necessary for security, but this keeps the behavior similar to
269 : * REFRESH MATERIALIZED VIEW. Otherwise, one could create a materialized
270 : * view not possible to refresh.
271 : */
272 1550 : if (is_matview)
273 : {
274 444 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
275 444 : SetUserIdAndSecContext(save_userid,
276 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
277 444 : save_nestlevel = NewGUCNestLevel();
278 : }
279 :
280 1550 : if (into->skipData)
281 : {
282 : /*
283 : * If WITH NO DATA was specified, do not go through the rewriter,
284 : * planner and executor. Just define the relation using a code path
285 : * similar to CREATE VIEW. This avoids dump/restore problems stemming
286 : * from running the planner before all dependencies are set up.
287 : */
288 110 : address = create_ctas_nodata(query->targetList, into);
289 : }
290 : else
291 : {
292 : /*
293 : * Parse analysis was done already, but we still have to run the rule
294 : * rewriter. We do not do AcquireRewriteLocks: we assume the query
295 : * either came straight from the parser, or suitable locks were
296 : * acquired by plancache.c.
297 : */
298 1440 : rewritten = QueryRewrite(query);
299 :
300 : /* SELECT should never rewrite to more or less than one SELECT query */
301 1440 : if (list_length(rewritten) != 1)
302 0 : elog(ERROR, "unexpected rewrite result for %s",
303 : is_matview ? "CREATE MATERIALIZED VIEW" :
304 : "CREATE TABLE AS SELECT");
305 1440 : query = linitial_node(Query, rewritten);
306 : Assert(query->commandType == CMD_SELECT);
307 :
308 : /* plan the query */
309 1440 : plan = pg_plan_query(query, pstate->p_sourcetext,
310 : CURSOR_OPT_PARALLEL_OK, params);
311 :
312 : /*
313 : * Use a snapshot with an updated command ID to ensure this query sees
314 : * results of any previously executed queries. (This could only
315 : * matter if the planner executed an allegedly-stable function that
316 : * changed the database contents, but let's do it anyway to be
317 : * parallel to the EXPLAIN code path.)
318 : */
319 1434 : PushCopiedSnapshot(GetActiveSnapshot());
320 1434 : UpdateActiveSnapshotCommandId();
321 :
322 : /* Create a QueryDesc, redirecting output to our tuple receiver */
323 1434 : queryDesc = CreateQueryDesc(plan, pstate->p_sourcetext,
324 : GetActiveSnapshot(), InvalidSnapshot,
325 : dest, params, queryEnv, 0);
326 :
327 : /* call ExecutorStart to prepare the plan for execution */
328 1434 : ExecutorStart(queryDesc, GetIntoRelEFlags(into));
329 :
330 : /* run the plan to completion */
331 1434 : ExecutorRun(queryDesc, ForwardScanDirection, 0, true);
332 :
333 : /* save the rowcount if we're given a qc to fill */
334 1392 : if (qc)
335 1372 : SetQueryCompletion(qc, CMDTAG_SELECT, queryDesc->estate->es_processed);
336 :
337 : /* get object address that intorel_startup saved for us */
338 1392 : address = ((DR_intorel *) dest)->reladdr;
339 :
340 : /* and clean up */
341 1392 : ExecutorFinish(queryDesc);
342 1392 : ExecutorEnd(queryDesc);
343 :
344 1392 : FreeQueryDesc(queryDesc);
345 :
346 1392 : PopActiveSnapshot();
347 : }
348 :
349 1490 : if (is_matview)
350 : {
351 : /* Roll back any GUC changes */
352 426 : AtEOXact_GUC(false, save_nestlevel);
353 :
354 : /* Restore userid and security context */
355 426 : SetUserIdAndSecContext(save_userid, save_sec_context);
356 : }
357 :
358 1490 : return address;
359 : }
360 :
361 : /*
362 : * GetIntoRelEFlags --- compute executor flags needed for CREATE TABLE AS
363 : *
364 : * This is exported because EXPLAIN and PREPARE need it too. (Note: those
365 : * callers still need to deal explicitly with the skipData flag; since they
366 : * use different methods for suppressing execution, it doesn't seem worth
367 : * trying to encapsulate that part.)
368 : */
369 : int
370 1572 : GetIntoRelEFlags(IntoClause *intoClause)
371 : {
372 1572 : int flags = 0;
373 :
374 1572 : if (intoClause->skipData)
375 42 : flags |= EXEC_FLAG_WITH_NO_DATA;
376 :
377 1572 : return flags;
378 : }
379 :
380 : /*
381 : * CreateTableAsRelExists --- check existence of relation for CreateTableAsStmt
382 : *
383 : * Utility wrapper checking if the relation pending for creation in this
384 : * CreateTableAsStmt query already exists or not. Returns true if the
385 : * relation exists, otherwise false.
386 : */
387 : bool
388 1840 : CreateTableAsRelExists(CreateTableAsStmt *ctas)
389 : {
390 : Oid nspid;
391 : Oid oldrelid;
392 : ObjectAddress address;
393 1840 : IntoClause *into = ctas->into;
394 :
395 1840 : nspid = RangeVarGetCreationNamespace(into->rel);
396 :
397 1840 : oldrelid = get_relname_relid(into->rel->relname, nspid);
398 1840 : if (OidIsValid(oldrelid))
399 : {
400 152 : if (!ctas->if_not_exists)
401 72 : ereport(ERROR,
402 : (errcode(ERRCODE_DUPLICATE_TABLE),
403 : errmsg("relation \"%s\" already exists",
404 : into->rel->relname)));
405 :
406 : /*
407 : * The relation exists and IF NOT EXISTS has been specified.
408 : *
409 : * If we are in an extension script, insist that the pre-existing
410 : * object be a member of the extension, to avoid security risks.
411 : */
412 80 : ObjectAddressSet(address, RelationRelationId, oldrelid);
413 80 : checkMembershipInCurrentExtension(&address);
414 :
415 : /* OK to skip */
416 76 : ereport(NOTICE,
417 : (errcode(ERRCODE_DUPLICATE_TABLE),
418 : errmsg("relation \"%s\" already exists, skipping",
419 : into->rel->relname)));
420 76 : return true;
421 : }
422 :
423 : /* Relation does not exist, it can be created */
424 1688 : return false;
425 : }
426 :
427 : /*
428 : * CreateIntoRelDestReceiver -- create a suitable DestReceiver object
429 : *
430 : * intoClause will be NULL if called from CreateDestReceiver(), in which
431 : * case it has to be provided later. However, it is convenient to allow
432 : * self->into to be filled in immediately for other callers.
433 : */
434 : DestReceiver *
435 1688 : CreateIntoRelDestReceiver(IntoClause *intoClause)
436 : {
437 1688 : DR_intorel *self = (DR_intorel *) palloc0(sizeof(DR_intorel));
438 :
439 1688 : self->pub.receiveSlot = intorel_receive;
440 1688 : self->pub.rStartup = intorel_startup;
441 1688 : self->pub.rShutdown = intorel_shutdown;
442 1688 : self->pub.rDestroy = intorel_destroy;
443 1688 : self->pub.mydest = DestIntoRel;
444 1688 : self->into = intoClause;
445 : /* other private fields will be set during intorel_startup */
446 :
447 1688 : return (DestReceiver *) self;
448 : }
449 :
450 : /*
451 : * intorel_startup --- executor startup
452 : */
453 : static void
454 1530 : intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo)
455 : {
456 1530 : DR_intorel *myState = (DR_intorel *) self;
457 1530 : IntoClause *into = myState->into;
458 : bool is_matview;
459 : List *attrList;
460 : ObjectAddress intoRelationAddr;
461 : Relation intoRelationDesc;
462 : ListCell *lc;
463 : int attnum;
464 :
465 : Assert(into != NULL); /* else somebody forgot to set it */
466 :
467 : /* This code supports both CREATE TABLE AS and CREATE MATERIALIZED VIEW */
468 1530 : is_matview = (into->viewQuery != NULL);
469 :
470 : /*
471 : * Build column definitions using "pre-cooked" type and collation info. If
472 : * a column name list was specified in CREATE TABLE AS, override the
473 : * column names derived from the query. (Too few column names are OK, too
474 : * many are not.)
475 : */
476 1530 : attrList = NIL;
477 1530 : lc = list_head(into->colNames);
478 5932 : for (attnum = 0; attnum < typeinfo->natts; attnum++)
479 : {
480 4414 : Form_pg_attribute attribute = TupleDescAttr(typeinfo, attnum);
481 : ColumnDef *col;
482 : char *colname;
483 :
484 4414 : if (lc)
485 : {
486 264 : colname = strVal(lfirst(lc));
487 264 : lc = lnext(into->colNames, lc);
488 : }
489 : else
490 4150 : colname = NameStr(attribute->attname);
491 :
492 4414 : col = makeColumnDef(colname,
493 : attribute->atttypid,
494 : attribute->atttypmod,
495 : attribute->attcollation);
496 :
497 : /*
498 : * It's possible that the column is of a collatable type but the
499 : * collation could not be resolved, so double-check. (We must check
500 : * this here because DefineRelation would adopt the type's default
501 : * collation rather than complaining.)
502 : */
503 8144 : if (!OidIsValid(col->collOid) &&
504 3730 : type_is_collatable(col->typeName->typeOid))
505 12 : ereport(ERROR,
506 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
507 : errmsg("no collation was derived for column \"%s\" with collatable type %s",
508 : col->colname,
509 : format_type_be(col->typeName->typeOid)),
510 : errhint("Use the COLLATE clause to set the collation explicitly.")));
511 :
512 4402 : attrList = lappend(attrList, col);
513 : }
514 :
515 1518 : if (lc != NULL)
516 12 : ereport(ERROR,
517 : (errcode(ERRCODE_SYNTAX_ERROR),
518 : errmsg("too many column names were specified")));
519 :
520 : /*
521 : * Actually create the target table
522 : */
523 1506 : intoRelationAddr = create_ctas_internal(attrList, into);
524 :
525 : /*
526 : * Finally we can open the target table
527 : */
528 1488 : intoRelationDesc = table_open(intoRelationAddr.objectId, AccessExclusiveLock);
529 :
530 : /*
531 : * Make sure the constructed table does not have RLS enabled.
532 : *
533 : * check_enable_rls() will ereport(ERROR) itself if the user has requested
534 : * something invalid, and otherwise will return RLS_ENABLED if RLS should
535 : * be enabled here. We don't actually support that currently, so throw
536 : * our own ereport(ERROR) if that happens.
537 : */
538 1488 : if (check_enable_rls(intoRelationAddr.objectId, InvalidOid, false) == RLS_ENABLED)
539 0 : ereport(ERROR,
540 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
541 : errmsg("policies not yet implemented for this command")));
542 :
543 : /*
544 : * Tentatively mark the target as populated, if it's a matview and we're
545 : * going to fill it; otherwise, no change needed.
546 : */
547 1488 : if (is_matview && !into->skipData)
548 356 : SetMatViewPopulatedState(intoRelationDesc, true);
549 :
550 : /*
551 : * Fill private fields of myState for use by later routines
552 : */
553 1488 : myState->rel = intoRelationDesc;
554 1488 : myState->reladdr = intoRelationAddr;
555 1488 : myState->output_cid = GetCurrentCommandId(true);
556 1488 : myState->ti_options = TABLE_INSERT_SKIP_FSM;
557 :
558 : /*
559 : * If WITH NO DATA is specified, there is no need to set up the state for
560 : * bulk inserts as there are no tuples to insert.
561 : */
562 1488 : if (!into->skipData)
563 1452 : myState->bistate = GetBulkInsertState();
564 : else
565 36 : myState->bistate = NULL;
566 :
567 : /*
568 : * Valid smgr_targblock implies something already wrote to the relation.
569 : * This may be harmless, but this function hasn't planned for it.
570 : */
571 : Assert(RelationGetTargetBlock(intoRelationDesc) == InvalidBlockNumber);
572 1488 : }
573 :
574 : /*
575 : * intorel_receive --- receive one tuple
576 : */
577 : static bool
578 2123712 : intorel_receive(TupleTableSlot *slot, DestReceiver *self)
579 : {
580 2123712 : DR_intorel *myState = (DR_intorel *) self;
581 :
582 : /* Nothing to insert if WITH NO DATA is specified. */
583 2123712 : if (!myState->into->skipData)
584 : {
585 : /*
586 : * Note that the input slot might not be of the type of the target
587 : * relation. That's supported by table_tuple_insert(), but slightly
588 : * less efficient than inserting with the right slot - but the
589 : * alternative would be to copy into a slot of the right type, which
590 : * would not be cheap either. This also doesn't allow accessing per-AM
591 : * data (say a tuple's xmin), but since we don't do that here...
592 : */
593 2123712 : table_tuple_insert(myState->rel,
594 : slot,
595 : myState->output_cid,
596 : myState->ti_options,
597 : myState->bistate);
598 : }
599 :
600 : /* We know this is a newly created relation, so there are no indexes */
601 :
602 2123712 : return true;
603 : }
604 :
605 : /*
606 : * intorel_shutdown --- executor end
607 : */
608 : static void
609 1488 : intorel_shutdown(DestReceiver *self)
610 : {
611 1488 : DR_intorel *myState = (DR_intorel *) self;
612 1488 : IntoClause *into = myState->into;
613 :
614 1488 : if (!into->skipData)
615 : {
616 1452 : FreeBulkInsertState(myState->bistate);
617 1452 : table_finish_bulk_insert(myState->rel, myState->ti_options);
618 : }
619 :
620 : /* close rel, but keep lock until commit */
621 1488 : table_close(myState->rel, NoLock);
622 1488 : myState->rel = NULL;
623 1488 : }
624 :
625 : /*
626 : * intorel_destroy --- release DestReceiver object
627 : */
628 : static void
629 96 : intorel_destroy(DestReceiver *self)
630 : {
631 96 : pfree(self);
632 96 : }
|