1 | /*------------------------------------------------------------------------- |
2 | * |
3 | * nodeIndexscan.c |
4 | * Routines to support indexed scans of relations |
5 | * |
6 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
7 | * Portions Copyright (c) 1994, Regents of the University of California |
8 | * |
9 | * |
10 | * IDENTIFICATION |
11 | * src/backend/executor/nodeIndexscan.c |
12 | * |
13 | *------------------------------------------------------------------------- |
14 | */ |
15 | /* |
16 | * INTERFACE ROUTINES |
17 | * ExecIndexScan scans a relation using an index |
18 | * IndexNext retrieve next tuple using index |
19 | * IndexNextWithReorder same, but recheck ORDER BY expressions |
20 | * ExecInitIndexScan creates and initializes state info. |
21 | * ExecReScanIndexScan rescans the indexed relation. |
22 | * ExecEndIndexScan releases all storage. |
23 | * ExecIndexMarkPos marks scan position. |
24 | * ExecIndexRestrPos restores scan position. |
25 | * ExecIndexScanEstimate estimates DSM space needed for parallel index scan |
26 | * ExecIndexScanInitializeDSM initialize DSM for parallel indexscan |
27 | * ExecIndexScanReInitializeDSM reinitialize DSM for fresh scan |
28 | * ExecIndexScanInitializeWorker attach to DSM info in parallel worker |
29 | */ |
30 | #include "postgres.h" |
31 | |
32 | #include "access/nbtree.h" |
33 | #include "access/relscan.h" |
34 | #include "access/tableam.h" |
35 | #include "catalog/pg_am.h" |
36 | #include "executor/execdebug.h" |
37 | #include "executor/nodeIndexscan.h" |
38 | #include "lib/pairingheap.h" |
39 | #include "miscadmin.h" |
40 | #include "nodes/nodeFuncs.h" |
41 | #include "utils/array.h" |
42 | #include "utils/datum.h" |
43 | #include "utils/lsyscache.h" |
44 | #include "utils/memutils.h" |
45 | #include "utils/rel.h" |
46 | |
47 | /* |
48 | * When an ordering operator is used, tuples fetched from the index that |
49 | * need to be reordered are queued in a pairing heap, as ReorderTuples. |
50 | */ |
51 | typedef struct |
52 | { |
53 | pairingheap_node ph_node; |
54 | HeapTuple htup; |
55 | Datum *orderbyvals; |
56 | bool *orderbynulls; |
57 | } ReorderTuple; |
58 | |
59 | static TupleTableSlot *IndexNext(IndexScanState *node); |
60 | static TupleTableSlot *IndexNextWithReorder(IndexScanState *node); |
61 | static void EvalOrderByExpressions(IndexScanState *node, ExprContext *econtext); |
62 | static bool IndexRecheck(IndexScanState *node, TupleTableSlot *slot); |
63 | static int cmp_orderbyvals(const Datum *adist, const bool *anulls, |
64 | const Datum *bdist, const bool *bnulls, |
65 | IndexScanState *node); |
66 | static int reorderqueue_cmp(const pairingheap_node *a, |
67 | const pairingheap_node *b, void *arg); |
68 | static void reorderqueue_push(IndexScanState *node, TupleTableSlot *slot, |
69 | Datum *orderbyvals, bool *orderbynulls); |
70 | static HeapTuple reorderqueue_pop(IndexScanState *node); |
71 | |
72 | |
73 | /* ---------------------------------------------------------------- |
74 | * IndexNext |
75 | * |
76 | * Retrieve a tuple from the IndexScan node's currentRelation |
77 | * using the index specified in the IndexScanState information. |
78 | * ---------------------------------------------------------------- |
79 | */ |
80 | static TupleTableSlot * |
81 | IndexNext(IndexScanState *node) |
82 | { |
83 | EState *estate; |
84 | ExprContext *econtext; |
85 | ScanDirection direction; |
86 | IndexScanDesc scandesc; |
87 | TupleTableSlot *slot; |
88 | |
89 | /* |
90 | * extract necessary information from index scan node |
91 | */ |
92 | estate = node->ss.ps.state; |
93 | direction = estate->es_direction; |
94 | /* flip direction if this is an overall backward scan */ |
95 | if (ScanDirectionIsBackward(((IndexScan *) node->ss.ps.plan)->indexorderdir)) |
96 | { |
97 | if (ScanDirectionIsForward(direction)) |
98 | direction = BackwardScanDirection; |
99 | else if (ScanDirectionIsBackward(direction)) |
100 | direction = ForwardScanDirection; |
101 | } |
102 | scandesc = node->iss_ScanDesc; |
103 | econtext = node->ss.ps.ps_ExprContext; |
104 | slot = node->ss.ss_ScanTupleSlot; |
105 | |
106 | if (scandesc == NULL) |
107 | { |
108 | /* |
109 | * We reach here if the index scan is not parallel, or if we're |
110 | * serially executing an index scan that was planned to be parallel. |
111 | */ |
112 | scandesc = index_beginscan(node->ss.ss_currentRelation, |
113 | node->iss_RelationDesc, |
114 | estate->es_snapshot, |
115 | node->iss_NumScanKeys, |
116 | node->iss_NumOrderByKeys); |
117 | |
118 | node->iss_ScanDesc = scandesc; |
119 | |
120 | /* |
121 | * If no run-time keys to calculate or they are ready, go ahead and |
122 | * pass the scankeys to the index AM. |
123 | */ |
124 | if (node->iss_NumRuntimeKeys == 0 || node->iss_RuntimeKeysReady) |
125 | index_rescan(scandesc, |
126 | node->iss_ScanKeys, node->iss_NumScanKeys, |
127 | node->iss_OrderByKeys, node->iss_NumOrderByKeys); |
128 | } |
129 | |
130 | /* |
131 | * ok, now that we have what we need, fetch the next tuple. |
132 | */ |
133 | while (index_getnext_slot(scandesc, direction, slot)) |
134 | { |
135 | CHECK_FOR_INTERRUPTS(); |
136 | |
137 | /* |
138 | * If the index was lossy, we have to recheck the index quals using |
139 | * the fetched tuple. |
140 | */ |
141 | if (scandesc->xs_recheck) |
142 | { |
143 | econtext->ecxt_scantuple = slot; |
144 | if (!ExecQualAndReset(node->indexqualorig, econtext)) |
145 | { |
146 | /* Fails recheck, so drop it and loop back for another */ |
147 | InstrCountFiltered2(node, 1); |
148 | continue; |
149 | } |
150 | } |
151 | |
152 | return slot; |
153 | } |
154 | |
155 | /* |
156 | * if we get here it means the index scan failed so we are at the end of |
157 | * the scan.. |
158 | */ |
159 | node->iss_ReachedEnd = true; |
160 | return ExecClearTuple(slot); |
161 | } |
162 | |
163 | /* ---------------------------------------------------------------- |
164 | * IndexNextWithReorder |
165 | * |
166 | * Like IndexNext, but this version can also re-check ORDER BY |
167 | * expressions, and reorder the tuples as necessary. |
168 | * ---------------------------------------------------------------- |
169 | */ |
170 | static TupleTableSlot * |
171 | IndexNextWithReorder(IndexScanState *node) |
172 | { |
173 | EState *estate; |
174 | ExprContext *econtext; |
175 | IndexScanDesc scandesc; |
176 | TupleTableSlot *slot; |
177 | ReorderTuple *topmost = NULL; |
178 | bool was_exact; |
179 | Datum *lastfetched_vals; |
180 | bool *lastfetched_nulls; |
181 | int cmp; |
182 | |
183 | estate = node->ss.ps.state; |
184 | |
185 | /* |
186 | * Only forward scan is supported with reordering. Note: we can get away |
187 | * with just Asserting here because the system will not try to run the |
188 | * plan backwards if ExecSupportsBackwardScan() says it won't work. |
189 | * Currently, that is guaranteed because no index AMs support both |
190 | * amcanorderbyop and amcanbackward; if any ever do, |
191 | * ExecSupportsBackwardScan() will need to consider indexorderbys |
192 | * explicitly. |
193 | */ |
194 | Assert(!ScanDirectionIsBackward(((IndexScan *) node->ss.ps.plan)->indexorderdir)); |
195 | Assert(ScanDirectionIsForward(estate->es_direction)); |
196 | |
197 | scandesc = node->iss_ScanDesc; |
198 | econtext = node->ss.ps.ps_ExprContext; |
199 | slot = node->ss.ss_ScanTupleSlot; |
200 | |
201 | if (scandesc == NULL) |
202 | { |
203 | /* |
204 | * We reach here if the index scan is not parallel, or if we're |
205 | * serially executing an index scan that was planned to be parallel. |
206 | */ |
207 | scandesc = index_beginscan(node->ss.ss_currentRelation, |
208 | node->iss_RelationDesc, |
209 | estate->es_snapshot, |
210 | node->iss_NumScanKeys, |
211 | node->iss_NumOrderByKeys); |
212 | |
213 | node->iss_ScanDesc = scandesc; |
214 | |
215 | /* |
216 | * If no run-time keys to calculate or they are ready, go ahead and |
217 | * pass the scankeys to the index AM. |
218 | */ |
219 | if (node->iss_NumRuntimeKeys == 0 || node->iss_RuntimeKeysReady) |
220 | index_rescan(scandesc, |
221 | node->iss_ScanKeys, node->iss_NumScanKeys, |
222 | node->iss_OrderByKeys, node->iss_NumOrderByKeys); |
223 | } |
224 | |
225 | for (;;) |
226 | { |
227 | CHECK_FOR_INTERRUPTS(); |
228 | |
229 | /* |
230 | * Check the reorder queue first. If the topmost tuple in the queue |
231 | * has an ORDER BY value smaller than (or equal to) the value last |
232 | * returned by the index, we can return it now. |
233 | */ |
234 | if (!pairingheap_is_empty(node->iss_ReorderQueue)) |
235 | { |
236 | topmost = (ReorderTuple *) pairingheap_first(node->iss_ReorderQueue); |
237 | |
238 | if (node->iss_ReachedEnd || |
239 | cmp_orderbyvals(topmost->orderbyvals, |
240 | topmost->orderbynulls, |
241 | scandesc->xs_orderbyvals, |
242 | scandesc->xs_orderbynulls, |
243 | node) <= 0) |
244 | { |
245 | HeapTuple tuple; |
246 | |
247 | tuple = reorderqueue_pop(node); |
248 | |
249 | /* Pass 'true', as the tuple in the queue is a palloc'd copy */ |
250 | ExecForceStoreHeapTuple(tuple, slot, true); |
251 | return slot; |
252 | } |
253 | } |
254 | else if (node->iss_ReachedEnd) |
255 | { |
256 | /* Queue is empty, and no more tuples from index. We're done. */ |
257 | return ExecClearTuple(slot); |
258 | } |
259 | |
260 | /* |
261 | * Fetch next tuple from the index. |
262 | */ |
263 | next_indextuple: |
264 | if (!index_getnext_slot(scandesc, ForwardScanDirection, slot)) |
265 | { |
266 | /* |
267 | * No more tuples from the index. But we still need to drain any |
268 | * remaining tuples from the queue before we're done. |
269 | */ |
270 | node->iss_ReachedEnd = true; |
271 | continue; |
272 | } |
273 | |
274 | /* |
275 | * If the index was lossy, we have to recheck the index quals and |
276 | * ORDER BY expressions using the fetched tuple. |
277 | */ |
278 | if (scandesc->xs_recheck) |
279 | { |
280 | econtext->ecxt_scantuple = slot; |
281 | if (!ExecQualAndReset(node->indexqualorig, econtext)) |
282 | { |
283 | /* Fails recheck, so drop it and loop back for another */ |
284 | InstrCountFiltered2(node, 1); |
285 | /* allow this loop to be cancellable */ |
286 | CHECK_FOR_INTERRUPTS(); |
287 | goto next_indextuple; |
288 | } |
289 | } |
290 | |
291 | if (scandesc->xs_recheckorderby) |
292 | { |
293 | econtext->ecxt_scantuple = slot; |
294 | ResetExprContext(econtext); |
295 | EvalOrderByExpressions(node, econtext); |
296 | |
297 | /* |
298 | * Was the ORDER BY value returned by the index accurate? The |
299 | * recheck flag means that the index can return inaccurate values, |
300 | * but then again, the value returned for any particular tuple |
301 | * could also be exactly correct. Compare the value returned by |
302 | * the index with the recalculated value. (If the value returned |
303 | * by the index happened to be exact right, we can often avoid |
304 | * pushing the tuple to the queue, just to pop it back out again.) |
305 | */ |
306 | cmp = cmp_orderbyvals(node->iss_OrderByValues, |
307 | node->iss_OrderByNulls, |
308 | scandesc->xs_orderbyvals, |
309 | scandesc->xs_orderbynulls, |
310 | node); |
311 | if (cmp < 0) |
312 | elog(ERROR, "index returned tuples in wrong order" ); |
313 | else if (cmp == 0) |
314 | was_exact = true; |
315 | else |
316 | was_exact = false; |
317 | lastfetched_vals = node->iss_OrderByValues; |
318 | lastfetched_nulls = node->iss_OrderByNulls; |
319 | } |
320 | else |
321 | { |
322 | was_exact = true; |
323 | lastfetched_vals = scandesc->xs_orderbyvals; |
324 | lastfetched_nulls = scandesc->xs_orderbynulls; |
325 | } |
326 | |
327 | /* |
328 | * Can we return this tuple immediately, or does it need to be pushed |
329 | * to the reorder queue? If the ORDER BY expression values returned |
330 | * by the index were inaccurate, we can't return it yet, because the |
331 | * next tuple from the index might need to come before this one. Also, |
332 | * we can't return it yet if there are any smaller tuples in the queue |
333 | * already. |
334 | */ |
335 | if (!was_exact || (topmost && cmp_orderbyvals(lastfetched_vals, |
336 | lastfetched_nulls, |
337 | topmost->orderbyvals, |
338 | topmost->orderbynulls, |
339 | node) > 0)) |
340 | { |
341 | /* Put this tuple to the queue */ |
342 | reorderqueue_push(node, slot, lastfetched_vals, lastfetched_nulls); |
343 | continue; |
344 | } |
345 | else |
346 | { |
347 | /* Can return this tuple immediately. */ |
348 | return slot; |
349 | } |
350 | } |
351 | |
352 | /* |
353 | * if we get here it means the index scan failed so we are at the end of |
354 | * the scan.. |
355 | */ |
356 | return ExecClearTuple(slot); |
357 | } |
358 | |
359 | /* |
360 | * Calculate the expressions in the ORDER BY clause, based on the heap tuple. |
361 | */ |
362 | static void |
363 | EvalOrderByExpressions(IndexScanState *node, ExprContext *econtext) |
364 | { |
365 | int i; |
366 | ListCell *l; |
367 | MemoryContext oldContext; |
368 | |
369 | oldContext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory); |
370 | |
371 | i = 0; |
372 | foreach(l, node->indexorderbyorig) |
373 | { |
374 | ExprState *orderby = (ExprState *) lfirst(l); |
375 | |
376 | node->iss_OrderByValues[i] = ExecEvalExpr(orderby, |
377 | econtext, |
378 | &node->iss_OrderByNulls[i]); |
379 | i++; |
380 | } |
381 | |
382 | MemoryContextSwitchTo(oldContext); |
383 | } |
384 | |
385 | /* |
386 | * IndexRecheck -- access method routine to recheck a tuple in EvalPlanQual |
387 | */ |
388 | static bool |
389 | IndexRecheck(IndexScanState *node, TupleTableSlot *slot) |
390 | { |
391 | ExprContext *econtext; |
392 | |
393 | /* |
394 | * extract necessary information from index scan node |
395 | */ |
396 | econtext = node->ss.ps.ps_ExprContext; |
397 | |
398 | /* Does the tuple meet the indexqual condition? */ |
399 | econtext->ecxt_scantuple = slot; |
400 | return ExecQualAndReset(node->indexqualorig, econtext); |
401 | } |
402 | |
403 | |
404 | /* |
405 | * Compare ORDER BY expression values. |
406 | */ |
407 | static int |
408 | cmp_orderbyvals(const Datum *adist, const bool *anulls, |
409 | const Datum *bdist, const bool *bnulls, |
410 | IndexScanState *node) |
411 | { |
412 | int i; |
413 | int result; |
414 | |
415 | for (i = 0; i < node->iss_NumOrderByKeys; i++) |
416 | { |
417 | SortSupport ssup = &node->iss_SortSupport[i]; |
418 | |
419 | /* |
420 | * Handle nulls. We only need to support NULLS LAST ordering, because |
421 | * match_pathkeys_to_index() doesn't consider indexorderby |
422 | * implementation otherwise. |
423 | */ |
424 | if (anulls[i] && !bnulls[i]) |
425 | return 1; |
426 | else if (!anulls[i] && bnulls[i]) |
427 | return -1; |
428 | else if (anulls[i] && bnulls[i]) |
429 | return 0; |
430 | |
431 | result = ssup->comparator(adist[i], bdist[i], ssup); |
432 | if (result != 0) |
433 | return result; |
434 | } |
435 | |
436 | return 0; |
437 | } |
438 | |
439 | /* |
440 | * Pairing heap provides getting topmost (greatest) element while KNN provides |
441 | * ascending sort. That's why we invert the sort order. |
442 | */ |
443 | static int |
444 | reorderqueue_cmp(const pairingheap_node *a, const pairingheap_node *b, |
445 | void *arg) |
446 | { |
447 | ReorderTuple *rta = (ReorderTuple *) a; |
448 | ReorderTuple *rtb = (ReorderTuple *) b; |
449 | IndexScanState *node = (IndexScanState *) arg; |
450 | |
451 | /* exchange argument order to invert the sort order */ |
452 | return cmp_orderbyvals(rtb->orderbyvals, rtb->orderbynulls, |
453 | rta->orderbyvals, rta->orderbynulls, |
454 | node); |
455 | } |
456 | |
457 | /* |
458 | * Helper function to push a tuple to the reorder queue. |
459 | */ |
460 | static void |
461 | reorderqueue_push(IndexScanState *node, TupleTableSlot *slot, |
462 | Datum *orderbyvals, bool *orderbynulls) |
463 | { |
464 | IndexScanDesc scandesc = node->iss_ScanDesc; |
465 | EState *estate = node->ss.ps.state; |
466 | MemoryContext oldContext = MemoryContextSwitchTo(estate->es_query_cxt); |
467 | ReorderTuple *rt; |
468 | int i; |
469 | |
470 | rt = (ReorderTuple *) palloc(sizeof(ReorderTuple)); |
471 | rt->htup = ExecCopySlotHeapTuple(slot); |
472 | rt->orderbyvals = |
473 | (Datum *) palloc(sizeof(Datum) * scandesc->numberOfOrderBys); |
474 | rt->orderbynulls = |
475 | (bool *) palloc(sizeof(bool) * scandesc->numberOfOrderBys); |
476 | for (i = 0; i < node->iss_NumOrderByKeys; i++) |
477 | { |
478 | if (!orderbynulls[i]) |
479 | rt->orderbyvals[i] = datumCopy(orderbyvals[i], |
480 | node->iss_OrderByTypByVals[i], |
481 | node->iss_OrderByTypLens[i]); |
482 | else |
483 | rt->orderbyvals[i] = (Datum) 0; |
484 | rt->orderbynulls[i] = orderbynulls[i]; |
485 | } |
486 | pairingheap_add(node->iss_ReorderQueue, &rt->ph_node); |
487 | |
488 | MemoryContextSwitchTo(oldContext); |
489 | } |
490 | |
491 | /* |
492 | * Helper function to pop the next tuple from the reorder queue. |
493 | */ |
494 | static HeapTuple |
495 | reorderqueue_pop(IndexScanState *node) |
496 | { |
497 | HeapTuple result; |
498 | ReorderTuple *topmost; |
499 | int i; |
500 | |
501 | topmost = (ReorderTuple *) pairingheap_remove_first(node->iss_ReorderQueue); |
502 | |
503 | result = topmost->htup; |
504 | for (i = 0; i < node->iss_NumOrderByKeys; i++) |
505 | { |
506 | if (!node->iss_OrderByTypByVals[i] && !topmost->orderbynulls[i]) |
507 | pfree(DatumGetPointer(topmost->orderbyvals[i])); |
508 | } |
509 | pfree(topmost->orderbyvals); |
510 | pfree(topmost->orderbynulls); |
511 | pfree(topmost); |
512 | |
513 | return result; |
514 | } |
515 | |
516 | |
517 | /* ---------------------------------------------------------------- |
518 | * ExecIndexScan(node) |
519 | * ---------------------------------------------------------------- |
520 | */ |
521 | static TupleTableSlot * |
522 | ExecIndexScan(PlanState *pstate) |
523 | { |
524 | IndexScanState *node = castNode(IndexScanState, pstate); |
525 | |
526 | /* |
527 | * If we have runtime keys and they've not already been set up, do it now. |
528 | */ |
529 | if (node->iss_NumRuntimeKeys != 0 && !node->iss_RuntimeKeysReady) |
530 | ExecReScan((PlanState *) node); |
531 | |
532 | if (node->iss_NumOrderByKeys > 0) |
533 | return ExecScan(&node->ss, |
534 | (ExecScanAccessMtd) IndexNextWithReorder, |
535 | (ExecScanRecheckMtd) IndexRecheck); |
536 | else |
537 | return ExecScan(&node->ss, |
538 | (ExecScanAccessMtd) IndexNext, |
539 | (ExecScanRecheckMtd) IndexRecheck); |
540 | } |
541 | |
542 | /* ---------------------------------------------------------------- |
543 | * ExecReScanIndexScan(node) |
544 | * |
545 | * Recalculates the values of any scan keys whose value depends on |
546 | * information known at runtime, then rescans the indexed relation. |
547 | * |
548 | * Updating the scan key was formerly done separately in |
549 | * ExecUpdateIndexScanKeys. Integrating it into ReScan makes |
550 | * rescans of indices and relations/general streams more uniform. |
551 | * ---------------------------------------------------------------- |
552 | */ |
553 | void |
554 | ExecReScanIndexScan(IndexScanState *node) |
555 | { |
556 | /* |
557 | * If we are doing runtime key calculations (ie, any of the index key |
558 | * values weren't simple Consts), compute the new key values. But first, |
559 | * reset the context so we don't leak memory as each outer tuple is |
560 | * scanned. Note this assumes that we will recalculate *all* runtime keys |
561 | * on each call. |
562 | */ |
563 | if (node->iss_NumRuntimeKeys != 0) |
564 | { |
565 | ExprContext *econtext = node->iss_RuntimeContext; |
566 | |
567 | ResetExprContext(econtext); |
568 | ExecIndexEvalRuntimeKeys(econtext, |
569 | node->iss_RuntimeKeys, |
570 | node->iss_NumRuntimeKeys); |
571 | } |
572 | node->iss_RuntimeKeysReady = true; |
573 | |
574 | /* flush the reorder queue */ |
575 | if (node->iss_ReorderQueue) |
576 | { |
577 | while (!pairingheap_is_empty(node->iss_ReorderQueue)) |
578 | reorderqueue_pop(node); |
579 | } |
580 | |
581 | /* reset index scan */ |
582 | if (node->iss_ScanDesc) |
583 | index_rescan(node->iss_ScanDesc, |
584 | node->iss_ScanKeys, node->iss_NumScanKeys, |
585 | node->iss_OrderByKeys, node->iss_NumOrderByKeys); |
586 | node->iss_ReachedEnd = false; |
587 | |
588 | ExecScanReScan(&node->ss); |
589 | } |
590 | |
591 | |
592 | /* |
593 | * ExecIndexEvalRuntimeKeys |
594 | * Evaluate any runtime key values, and update the scankeys. |
595 | */ |
596 | void |
597 | ExecIndexEvalRuntimeKeys(ExprContext *econtext, |
598 | IndexRuntimeKeyInfo *runtimeKeys, int numRuntimeKeys) |
599 | { |
600 | int j; |
601 | MemoryContext oldContext; |
602 | |
603 | /* We want to keep the key values in per-tuple memory */ |
604 | oldContext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory); |
605 | |
606 | for (j = 0; j < numRuntimeKeys; j++) |
607 | { |
608 | ScanKey scan_key = runtimeKeys[j].scan_key; |
609 | ExprState *key_expr = runtimeKeys[j].key_expr; |
610 | Datum scanvalue; |
611 | bool isNull; |
612 | |
613 | /* |
614 | * For each run-time key, extract the run-time expression and evaluate |
615 | * it with respect to the current context. We then stick the result |
616 | * into the proper scan key. |
617 | * |
618 | * Note: the result of the eval could be a pass-by-ref value that's |
619 | * stored in some outer scan's tuple, not in |
620 | * econtext->ecxt_per_tuple_memory. We assume that the outer tuple |
621 | * will stay put throughout our scan. If this is wrong, we could copy |
622 | * the result into our context explicitly, but I think that's not |
623 | * necessary. |
624 | * |
625 | * It's also entirely possible that the result of the eval is a |
626 | * toasted value. In this case we should forcibly detoast it, to |
627 | * avoid repeat detoastings each time the value is examined by an |
628 | * index support function. |
629 | */ |
630 | scanvalue = ExecEvalExpr(key_expr, |
631 | econtext, |
632 | &isNull); |
633 | if (isNull) |
634 | { |
635 | scan_key->sk_argument = scanvalue; |
636 | scan_key->sk_flags |= SK_ISNULL; |
637 | } |
638 | else |
639 | { |
640 | if (runtimeKeys[j].key_toastable) |
641 | scanvalue = PointerGetDatum(PG_DETOAST_DATUM(scanvalue)); |
642 | scan_key->sk_argument = scanvalue; |
643 | scan_key->sk_flags &= ~SK_ISNULL; |
644 | } |
645 | } |
646 | |
647 | MemoryContextSwitchTo(oldContext); |
648 | } |
649 | |
650 | /* |
651 | * ExecIndexEvalArrayKeys |
652 | * Evaluate any array key values, and set up to iterate through arrays. |
653 | * |
654 | * Returns true if there are array elements to consider; false means there |
655 | * is at least one null or empty array, so no match is possible. On true |
656 | * result, the scankeys are initialized with the first elements of the arrays. |
657 | */ |
658 | bool |
659 | ExecIndexEvalArrayKeys(ExprContext *econtext, |
660 | IndexArrayKeyInfo *arrayKeys, int numArrayKeys) |
661 | { |
662 | bool result = true; |
663 | int j; |
664 | MemoryContext oldContext; |
665 | |
666 | /* We want to keep the arrays in per-tuple memory */ |
667 | oldContext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory); |
668 | |
669 | for (j = 0; j < numArrayKeys; j++) |
670 | { |
671 | ScanKey scan_key = arrayKeys[j].scan_key; |
672 | ExprState *array_expr = arrayKeys[j].array_expr; |
673 | Datum arraydatum; |
674 | bool isNull; |
675 | ArrayType *arrayval; |
676 | int16 elmlen; |
677 | bool elmbyval; |
678 | char elmalign; |
679 | int num_elems; |
680 | Datum *elem_values; |
681 | bool *elem_nulls; |
682 | |
683 | /* |
684 | * Compute and deconstruct the array expression. (Notes in |
685 | * ExecIndexEvalRuntimeKeys() apply here too.) |
686 | */ |
687 | arraydatum = ExecEvalExpr(array_expr, |
688 | econtext, |
689 | &isNull); |
690 | if (isNull) |
691 | { |
692 | result = false; |
693 | break; /* no point in evaluating more */ |
694 | } |
695 | arrayval = DatumGetArrayTypeP(arraydatum); |
696 | /* We could cache this data, but not clear it's worth it */ |
697 | get_typlenbyvalalign(ARR_ELEMTYPE(arrayval), |
698 | &elmlen, &elmbyval, &elmalign); |
699 | deconstruct_array(arrayval, |
700 | ARR_ELEMTYPE(arrayval), |
701 | elmlen, elmbyval, elmalign, |
702 | &elem_values, &elem_nulls, &num_elems); |
703 | if (num_elems <= 0) |
704 | { |
705 | result = false; |
706 | break; /* no point in evaluating more */ |
707 | } |
708 | |
709 | /* |
710 | * Note: we expect the previous array data, if any, to be |
711 | * automatically freed by resetting the per-tuple context; hence no |
712 | * pfree's here. |
713 | */ |
714 | arrayKeys[j].elem_values = elem_values; |
715 | arrayKeys[j].elem_nulls = elem_nulls; |
716 | arrayKeys[j].num_elems = num_elems; |
717 | scan_key->sk_argument = elem_values[0]; |
718 | if (elem_nulls[0]) |
719 | scan_key->sk_flags |= SK_ISNULL; |
720 | else |
721 | scan_key->sk_flags &= ~SK_ISNULL; |
722 | arrayKeys[j].next_elem = 1; |
723 | } |
724 | |
725 | MemoryContextSwitchTo(oldContext); |
726 | |
727 | return result; |
728 | } |
729 | |
730 | /* |
731 | * ExecIndexAdvanceArrayKeys |
732 | * Advance to the next set of array key values, if any. |
733 | * |
734 | * Returns true if there is another set of values to consider, false if not. |
735 | * On true result, the scankeys are initialized with the next set of values. |
736 | */ |
737 | bool |
738 | ExecIndexAdvanceArrayKeys(IndexArrayKeyInfo *arrayKeys, int numArrayKeys) |
739 | { |
740 | bool found = false; |
741 | int j; |
742 | |
743 | /* |
744 | * Note we advance the rightmost array key most quickly, since it will |
745 | * correspond to the lowest-order index column among the available |
746 | * qualifications. This is hypothesized to result in better locality of |
747 | * access in the index. |
748 | */ |
749 | for (j = numArrayKeys - 1; j >= 0; j--) |
750 | { |
751 | ScanKey scan_key = arrayKeys[j].scan_key; |
752 | int next_elem = arrayKeys[j].next_elem; |
753 | int num_elems = arrayKeys[j].num_elems; |
754 | Datum *elem_values = arrayKeys[j].elem_values; |
755 | bool *elem_nulls = arrayKeys[j].elem_nulls; |
756 | |
757 | if (next_elem >= num_elems) |
758 | { |
759 | next_elem = 0; |
760 | found = false; /* need to advance next array key */ |
761 | } |
762 | else |
763 | found = true; |
764 | scan_key->sk_argument = elem_values[next_elem]; |
765 | if (elem_nulls[next_elem]) |
766 | scan_key->sk_flags |= SK_ISNULL; |
767 | else |
768 | scan_key->sk_flags &= ~SK_ISNULL; |
769 | arrayKeys[j].next_elem = next_elem + 1; |
770 | if (found) |
771 | break; |
772 | } |
773 | |
774 | return found; |
775 | } |
776 | |
777 | |
778 | /* ---------------------------------------------------------------- |
779 | * ExecEndIndexScan |
780 | * ---------------------------------------------------------------- |
781 | */ |
782 | void |
783 | ExecEndIndexScan(IndexScanState *node) |
784 | { |
785 | Relation indexRelationDesc; |
786 | IndexScanDesc indexScanDesc; |
787 | |
788 | /* |
789 | * extract information from the node |
790 | */ |
791 | indexRelationDesc = node->iss_RelationDesc; |
792 | indexScanDesc = node->iss_ScanDesc; |
793 | |
794 | /* |
795 | * Free the exprcontext(s) ... now dead code, see ExecFreeExprContext |
796 | */ |
797 | #ifdef NOT_USED |
798 | ExecFreeExprContext(&node->ss.ps); |
799 | if (node->iss_RuntimeContext) |
800 | FreeExprContext(node->iss_RuntimeContext, true); |
801 | #endif |
802 | |
803 | /* |
804 | * clear out tuple table slots |
805 | */ |
806 | if (node->ss.ps.ps_ResultTupleSlot) |
807 | ExecClearTuple(node->ss.ps.ps_ResultTupleSlot); |
808 | ExecClearTuple(node->ss.ss_ScanTupleSlot); |
809 | |
810 | /* |
811 | * close the index relation (no-op if we didn't open it) |
812 | */ |
813 | if (indexScanDesc) |
814 | index_endscan(indexScanDesc); |
815 | if (indexRelationDesc) |
816 | index_close(indexRelationDesc, NoLock); |
817 | } |
818 | |
819 | /* ---------------------------------------------------------------- |
820 | * ExecIndexMarkPos |
821 | * |
822 | * Note: we assume that no caller attempts to set a mark before having read |
823 | * at least one tuple. Otherwise, iss_ScanDesc might still be NULL. |
824 | * ---------------------------------------------------------------- |
825 | */ |
826 | void |
827 | ExecIndexMarkPos(IndexScanState *node) |
828 | { |
829 | EState *estate = node->ss.ps.state; |
830 | EPQState *epqstate = estate->es_epq_active; |
831 | |
832 | if (epqstate != NULL) |
833 | { |
834 | /* |
835 | * We are inside an EvalPlanQual recheck. If a test tuple exists for |
836 | * this relation, then we shouldn't access the index at all. We would |
837 | * instead need to save, and later restore, the state of the |
838 | * relsubs_done flag, so that re-fetching the test tuple is possible. |
839 | * However, given the assumption that no caller sets a mark at the |
840 | * start of the scan, we can only get here with relsubs_done[i] |
841 | * already set, and so no state need be saved. |
842 | */ |
843 | Index scanrelid = ((Scan *) node->ss.ps.plan)->scanrelid; |
844 | |
845 | Assert(scanrelid > 0); |
846 | if (epqstate->relsubs_slot[scanrelid - 1] != NULL || |
847 | epqstate->relsubs_rowmark[scanrelid - 1] != NULL) |
848 | { |
849 | /* Verify the claim above */ |
850 | if (!epqstate->relsubs_done[scanrelid - 1]) |
851 | elog(ERROR, "unexpected ExecIndexMarkPos call in EPQ recheck" ); |
852 | return; |
853 | } |
854 | } |
855 | |
856 | index_markpos(node->iss_ScanDesc); |
857 | } |
858 | |
859 | /* ---------------------------------------------------------------- |
860 | * ExecIndexRestrPos |
861 | * ---------------------------------------------------------------- |
862 | */ |
863 | void |
864 | ExecIndexRestrPos(IndexScanState *node) |
865 | { |
866 | EState *estate = node->ss.ps.state; |
867 | EPQState *epqstate = estate->es_epq_active; |
868 | |
869 | if (estate->es_epq_active != NULL) |
870 | { |
871 | /* See comments in ExecIndexMarkPos */ |
872 | Index scanrelid = ((Scan *) node->ss.ps.plan)->scanrelid; |
873 | |
874 | Assert(scanrelid > 0); |
875 | if (epqstate->relsubs_slot[scanrelid - 1] != NULL || |
876 | epqstate->relsubs_rowmark[scanrelid - 1] != NULL) |
877 | { |
878 | /* Verify the claim above */ |
879 | if (!epqstate->relsubs_done[scanrelid - 1]) |
880 | elog(ERROR, "unexpected ExecIndexRestrPos call in EPQ recheck" ); |
881 | return; |
882 | } |
883 | } |
884 | |
885 | index_restrpos(node->iss_ScanDesc); |
886 | } |
887 | |
888 | /* ---------------------------------------------------------------- |
889 | * ExecInitIndexScan |
890 | * |
891 | * Initializes the index scan's state information, creates |
892 | * scan keys, and opens the base and index relations. |
893 | * |
894 | * Note: index scans have 2 sets of state information because |
895 | * we have to keep track of the base relation and the |
896 | * index relation. |
897 | * ---------------------------------------------------------------- |
898 | */ |
899 | IndexScanState * |
900 | ExecInitIndexScan(IndexScan *node, EState *estate, int eflags) |
901 | { |
902 | IndexScanState *indexstate; |
903 | Relation currentRelation; |
904 | LOCKMODE lockmode; |
905 | |
906 | /* |
907 | * create state structure |
908 | */ |
909 | indexstate = makeNode(IndexScanState); |
910 | indexstate->ss.ps.plan = (Plan *) node; |
911 | indexstate->ss.ps.state = estate; |
912 | indexstate->ss.ps.ExecProcNode = ExecIndexScan; |
913 | |
914 | /* |
915 | * Miscellaneous initialization |
916 | * |
917 | * create expression context for node |
918 | */ |
919 | ExecAssignExprContext(estate, &indexstate->ss.ps); |
920 | |
921 | /* |
922 | * open the scan relation |
923 | */ |
924 | currentRelation = ExecOpenScanRelation(estate, node->scan.scanrelid, eflags); |
925 | |
926 | indexstate->ss.ss_currentRelation = currentRelation; |
927 | indexstate->ss.ss_currentScanDesc = NULL; /* no heap scan here */ |
928 | |
929 | /* |
930 | * get the scan type from the relation descriptor. |
931 | */ |
932 | ExecInitScanTupleSlot(estate, &indexstate->ss, |
933 | RelationGetDescr(currentRelation), |
934 | table_slot_callbacks(currentRelation)); |
935 | |
936 | /* |
937 | * Initialize result type and projection. |
938 | */ |
939 | ExecInitResultTypeTL(&indexstate->ss.ps); |
940 | ExecAssignScanProjectionInfo(&indexstate->ss); |
941 | |
942 | /* |
943 | * initialize child expressions |
944 | * |
945 | * Note: we don't initialize all of the indexqual expression, only the |
946 | * sub-parts corresponding to runtime keys (see below). Likewise for |
947 | * indexorderby, if any. But the indexqualorig expression is always |
948 | * initialized even though it will only be used in some uncommon cases --- |
949 | * would be nice to improve that. (Problem is that any SubPlans present |
950 | * in the expression must be found now...) |
951 | */ |
952 | indexstate->ss.ps.qual = |
953 | ExecInitQual(node->scan.plan.qual, (PlanState *) indexstate); |
954 | indexstate->indexqualorig = |
955 | ExecInitQual(node->indexqualorig, (PlanState *) indexstate); |
956 | indexstate->indexorderbyorig = |
957 | ExecInitExprList(node->indexorderbyorig, (PlanState *) indexstate); |
958 | |
959 | /* |
960 | * If we are just doing EXPLAIN (ie, aren't going to run the plan), stop |
961 | * here. This allows an index-advisor plugin to EXPLAIN a plan containing |
962 | * references to nonexistent indexes. |
963 | */ |
964 | if (eflags & EXEC_FLAG_EXPLAIN_ONLY) |
965 | return indexstate; |
966 | |
967 | /* Open the index relation. */ |
968 | lockmode = exec_rt_fetch(node->scan.scanrelid, estate)->rellockmode; |
969 | indexstate->iss_RelationDesc = index_open(node->indexid, lockmode); |
970 | |
971 | /* |
972 | * Initialize index-specific scan state |
973 | */ |
974 | indexstate->iss_RuntimeKeysReady = false; |
975 | indexstate->iss_RuntimeKeys = NULL; |
976 | indexstate->iss_NumRuntimeKeys = 0; |
977 | |
978 | /* |
979 | * build the index scan keys from the index qualification |
980 | */ |
981 | ExecIndexBuildScanKeys((PlanState *) indexstate, |
982 | indexstate->iss_RelationDesc, |
983 | node->indexqual, |
984 | false, |
985 | &indexstate->iss_ScanKeys, |
986 | &indexstate->iss_NumScanKeys, |
987 | &indexstate->iss_RuntimeKeys, |
988 | &indexstate->iss_NumRuntimeKeys, |
989 | NULL, /* no ArrayKeys */ |
990 | NULL); |
991 | |
992 | /* |
993 | * any ORDER BY exprs have to be turned into scankeys in the same way |
994 | */ |
995 | ExecIndexBuildScanKeys((PlanState *) indexstate, |
996 | indexstate->iss_RelationDesc, |
997 | node->indexorderby, |
998 | true, |
999 | &indexstate->iss_OrderByKeys, |
1000 | &indexstate->iss_NumOrderByKeys, |
1001 | &indexstate->iss_RuntimeKeys, |
1002 | &indexstate->iss_NumRuntimeKeys, |
1003 | NULL, /* no ArrayKeys */ |
1004 | NULL); |
1005 | |
1006 | /* Initialize sort support, if we need to re-check ORDER BY exprs */ |
1007 | if (indexstate->iss_NumOrderByKeys > 0) |
1008 | { |
1009 | int numOrderByKeys = indexstate->iss_NumOrderByKeys; |
1010 | int i; |
1011 | ListCell *lco; |
1012 | ListCell *lcx; |
1013 | |
1014 | /* |
1015 | * Prepare sort support, and look up the data type for each ORDER BY |
1016 | * expression. |
1017 | */ |
1018 | Assert(numOrderByKeys == list_length(node->indexorderbyops)); |
1019 | Assert(numOrderByKeys == list_length(node->indexorderbyorig)); |
1020 | indexstate->iss_SortSupport = (SortSupportData *) |
1021 | palloc0(numOrderByKeys * sizeof(SortSupportData)); |
1022 | indexstate->iss_OrderByTypByVals = (bool *) |
1023 | palloc(numOrderByKeys * sizeof(bool)); |
1024 | indexstate->iss_OrderByTypLens = (int16 *) |
1025 | palloc(numOrderByKeys * sizeof(int16)); |
1026 | i = 0; |
1027 | forboth(lco, node->indexorderbyops, lcx, node->indexorderbyorig) |
1028 | { |
1029 | Oid orderbyop = lfirst_oid(lco); |
1030 | Node *orderbyexpr = (Node *) lfirst(lcx); |
1031 | Oid orderbyType = exprType(orderbyexpr); |
1032 | Oid orderbyColl = exprCollation(orderbyexpr); |
1033 | SortSupport orderbysort = &indexstate->iss_SortSupport[i]; |
1034 | |
1035 | /* Initialize sort support */ |
1036 | orderbysort->ssup_cxt = CurrentMemoryContext; |
1037 | orderbysort->ssup_collation = orderbyColl; |
1038 | /* See cmp_orderbyvals() comments on NULLS LAST */ |
1039 | orderbysort->ssup_nulls_first = false; |
1040 | /* ssup_attno is unused here and elsewhere */ |
1041 | orderbysort->ssup_attno = 0; |
1042 | /* No abbreviation */ |
1043 | orderbysort->abbreviate = false; |
1044 | PrepareSortSupportFromOrderingOp(orderbyop, orderbysort); |
1045 | |
1046 | get_typlenbyval(orderbyType, |
1047 | &indexstate->iss_OrderByTypLens[i], |
1048 | &indexstate->iss_OrderByTypByVals[i]); |
1049 | i++; |
1050 | } |
1051 | |
1052 | /* allocate arrays to hold the re-calculated distances */ |
1053 | indexstate->iss_OrderByValues = (Datum *) |
1054 | palloc(numOrderByKeys * sizeof(Datum)); |
1055 | indexstate->iss_OrderByNulls = (bool *) |
1056 | palloc(numOrderByKeys * sizeof(bool)); |
1057 | |
1058 | /* and initialize the reorder queue */ |
1059 | indexstate->iss_ReorderQueue = pairingheap_allocate(reorderqueue_cmp, |
1060 | indexstate); |
1061 | } |
1062 | |
1063 | /* |
1064 | * If we have runtime keys, we need an ExprContext to evaluate them. The |
1065 | * node's standard context won't do because we want to reset that context |
1066 | * for every tuple. So, build another context just like the other one... |
1067 | * -tgl 7/11/00 |
1068 | */ |
1069 | if (indexstate->iss_NumRuntimeKeys != 0) |
1070 | { |
1071 | ExprContext *stdecontext = indexstate->ss.ps.ps_ExprContext; |
1072 | |
1073 | ExecAssignExprContext(estate, &indexstate->ss.ps); |
1074 | indexstate->iss_RuntimeContext = indexstate->ss.ps.ps_ExprContext; |
1075 | indexstate->ss.ps.ps_ExprContext = stdecontext; |
1076 | } |
1077 | else |
1078 | { |
1079 | indexstate->iss_RuntimeContext = NULL; |
1080 | } |
1081 | |
1082 | /* |
1083 | * all done. |
1084 | */ |
1085 | return indexstate; |
1086 | } |
1087 | |
1088 | |
1089 | /* |
1090 | * ExecIndexBuildScanKeys |
1091 | * Build the index scan keys from the index qualification expressions |
1092 | * |
1093 | * The index quals are passed to the index AM in the form of a ScanKey array. |
1094 | * This routine sets up the ScanKeys, fills in all constant fields of the |
1095 | * ScanKeys, and prepares information about the keys that have non-constant |
1096 | * comparison values. We divide index qual expressions into five types: |
1097 | * |
1098 | * 1. Simple operator with constant comparison value ("indexkey op constant"). |
1099 | * For these, we just fill in a ScanKey containing the constant value. |
1100 | * |
1101 | * 2. Simple operator with non-constant value ("indexkey op expression"). |
1102 | * For these, we create a ScanKey with everything filled in except the |
1103 | * expression value, and set up an IndexRuntimeKeyInfo struct to drive |
1104 | * evaluation of the expression at the right times. |
1105 | * |
1106 | * 3. RowCompareExpr ("(indexkey, indexkey, ...) op (expr, expr, ...)"). |
1107 | * For these, we create a header ScanKey plus a subsidiary ScanKey array, |
1108 | * as specified in access/skey.h. The elements of the row comparison |
1109 | * can have either constant or non-constant comparison values. |
1110 | * |
1111 | * 4. ScalarArrayOpExpr ("indexkey op ANY (array-expression)"). If the index |
1112 | * supports amsearcharray, we handle these the same as simple operators, |
1113 | * setting the SK_SEARCHARRAY flag to tell the AM to handle them. Otherwise, |
1114 | * we create a ScanKey with everything filled in except the comparison value, |
1115 | * and set up an IndexArrayKeyInfo struct to drive processing of the qual. |
1116 | * (Note that if we use an IndexArrayKeyInfo struct, the array expression is |
1117 | * always treated as requiring runtime evaluation, even if it's a constant.) |
1118 | * |
1119 | * 5. NullTest ("indexkey IS NULL/IS NOT NULL"). We just fill in the |
1120 | * ScanKey properly. |
1121 | * |
1122 | * This code is also used to prepare ORDER BY expressions for amcanorderbyop |
1123 | * indexes. The behavior is exactly the same, except that we have to look up |
1124 | * the operator differently. Note that only cases 1 and 2 are currently |
1125 | * possible for ORDER BY. |
1126 | * |
1127 | * Input params are: |
1128 | * |
1129 | * planstate: executor state node we are working for |
1130 | * index: the index we are building scan keys for |
1131 | * quals: indexquals (or indexorderbys) expressions |
1132 | * isorderby: true if processing ORDER BY exprs, false if processing quals |
1133 | * *runtimeKeys: ptr to pre-existing IndexRuntimeKeyInfos, or NULL if none |
1134 | * *numRuntimeKeys: number of pre-existing runtime keys |
1135 | * |
1136 | * Output params are: |
1137 | * |
1138 | * *scanKeys: receives ptr to array of ScanKeys |
1139 | * *numScanKeys: receives number of scankeys |
1140 | * *runtimeKeys: receives ptr to array of IndexRuntimeKeyInfos, or NULL if none |
1141 | * *numRuntimeKeys: receives number of runtime keys |
1142 | * *arrayKeys: receives ptr to array of IndexArrayKeyInfos, or NULL if none |
1143 | * *numArrayKeys: receives number of array keys |
1144 | * |
1145 | * Caller may pass NULL for arrayKeys and numArrayKeys to indicate that |
1146 | * IndexArrayKeyInfos are not supported. |
1147 | */ |
1148 | void |
1149 | ExecIndexBuildScanKeys(PlanState *planstate, Relation index, |
1150 | List *quals, bool isorderby, |
1151 | ScanKey *scanKeys, int *numScanKeys, |
1152 | IndexRuntimeKeyInfo **runtimeKeys, int *numRuntimeKeys, |
1153 | IndexArrayKeyInfo **arrayKeys, int *numArrayKeys) |
1154 | { |
1155 | ListCell *qual_cell; |
1156 | ScanKey scan_keys; |
1157 | IndexRuntimeKeyInfo *runtime_keys; |
1158 | IndexArrayKeyInfo *array_keys; |
1159 | int n_scan_keys; |
1160 | int n_runtime_keys; |
1161 | int max_runtime_keys; |
1162 | int n_array_keys; |
1163 | int j; |
1164 | |
1165 | /* Allocate array for ScanKey structs: one per qual */ |
1166 | n_scan_keys = list_length(quals); |
1167 | scan_keys = (ScanKey) palloc(n_scan_keys * sizeof(ScanKeyData)); |
1168 | |
1169 | /* |
1170 | * runtime_keys array is dynamically resized as needed. We handle it this |
1171 | * way so that the same runtime keys array can be shared between |
1172 | * indexquals and indexorderbys, which will be processed in separate calls |
1173 | * of this function. Caller must be sure to pass in NULL/0 for first |
1174 | * call. |
1175 | */ |
1176 | runtime_keys = *runtimeKeys; |
1177 | n_runtime_keys = max_runtime_keys = *numRuntimeKeys; |
1178 | |
1179 | /* Allocate array_keys as large as it could possibly need to be */ |
1180 | array_keys = (IndexArrayKeyInfo *) |
1181 | palloc0(n_scan_keys * sizeof(IndexArrayKeyInfo)); |
1182 | n_array_keys = 0; |
1183 | |
1184 | /* |
1185 | * for each opclause in the given qual, convert the opclause into a single |
1186 | * scan key |
1187 | */ |
1188 | j = 0; |
1189 | foreach(qual_cell, quals) |
1190 | { |
1191 | Expr *clause = (Expr *) lfirst(qual_cell); |
1192 | ScanKey this_scan_key = &scan_keys[j++]; |
1193 | Oid opno; /* operator's OID */ |
1194 | RegProcedure opfuncid; /* operator proc id used in scan */ |
1195 | Oid opfamily; /* opfamily of index column */ |
1196 | int op_strategy; /* operator's strategy number */ |
1197 | Oid op_lefttype; /* operator's declared input types */ |
1198 | Oid op_righttype; |
1199 | Expr *leftop; /* expr on lhs of operator */ |
1200 | Expr *rightop; /* expr on rhs ... */ |
1201 | AttrNumber varattno; /* att number used in scan */ |
1202 | int indnkeyatts; |
1203 | |
1204 | indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); |
1205 | if (IsA(clause, OpExpr)) |
1206 | { |
1207 | /* indexkey op const or indexkey op expression */ |
1208 | int flags = 0; |
1209 | Datum scanvalue; |
1210 | |
1211 | opno = ((OpExpr *) clause)->opno; |
1212 | opfuncid = ((OpExpr *) clause)->opfuncid; |
1213 | |
1214 | /* |
1215 | * leftop should be the index key Var, possibly relabeled |
1216 | */ |
1217 | leftop = (Expr *) get_leftop(clause); |
1218 | |
1219 | if (leftop && IsA(leftop, RelabelType)) |
1220 | leftop = ((RelabelType *) leftop)->arg; |
1221 | |
1222 | Assert(leftop != NULL); |
1223 | |
1224 | if (!(IsA(leftop, Var) && |
1225 | ((Var *) leftop)->varno == INDEX_VAR)) |
1226 | elog(ERROR, "indexqual doesn't have key on left side" ); |
1227 | |
1228 | varattno = ((Var *) leftop)->varattno; |
1229 | if (varattno < 1 || varattno > indnkeyatts) |
1230 | elog(ERROR, "bogus index qualification" ); |
1231 | |
1232 | /* |
1233 | * We have to look up the operator's strategy number. This |
1234 | * provides a cross-check that the operator does match the index. |
1235 | */ |
1236 | opfamily = index->rd_opfamily[varattno - 1]; |
1237 | |
1238 | get_op_opfamily_properties(opno, opfamily, isorderby, |
1239 | &op_strategy, |
1240 | &op_lefttype, |
1241 | &op_righttype); |
1242 | |
1243 | if (isorderby) |
1244 | flags |= SK_ORDER_BY; |
1245 | |
1246 | /* |
1247 | * rightop is the constant or variable comparison value |
1248 | */ |
1249 | rightop = (Expr *) get_rightop(clause); |
1250 | |
1251 | if (rightop && IsA(rightop, RelabelType)) |
1252 | rightop = ((RelabelType *) rightop)->arg; |
1253 | |
1254 | Assert(rightop != NULL); |
1255 | |
1256 | if (IsA(rightop, Const)) |
1257 | { |
1258 | /* OK, simple constant comparison value */ |
1259 | scanvalue = ((Const *) rightop)->constvalue; |
1260 | if (((Const *) rightop)->constisnull) |
1261 | flags |= SK_ISNULL; |
1262 | } |
1263 | else |
1264 | { |
1265 | /* Need to treat this one as a runtime key */ |
1266 | if (n_runtime_keys >= max_runtime_keys) |
1267 | { |
1268 | if (max_runtime_keys == 0) |
1269 | { |
1270 | max_runtime_keys = 8; |
1271 | runtime_keys = (IndexRuntimeKeyInfo *) |
1272 | palloc(max_runtime_keys * sizeof(IndexRuntimeKeyInfo)); |
1273 | } |
1274 | else |
1275 | { |
1276 | max_runtime_keys *= 2; |
1277 | runtime_keys = (IndexRuntimeKeyInfo *) |
1278 | repalloc(runtime_keys, max_runtime_keys * sizeof(IndexRuntimeKeyInfo)); |
1279 | } |
1280 | } |
1281 | runtime_keys[n_runtime_keys].scan_key = this_scan_key; |
1282 | runtime_keys[n_runtime_keys].key_expr = |
1283 | ExecInitExpr(rightop, planstate); |
1284 | runtime_keys[n_runtime_keys].key_toastable = |
1285 | TypeIsToastable(op_righttype); |
1286 | n_runtime_keys++; |
1287 | scanvalue = (Datum) 0; |
1288 | } |
1289 | |
1290 | /* |
1291 | * initialize the scan key's fields appropriately |
1292 | */ |
1293 | ScanKeyEntryInitialize(this_scan_key, |
1294 | flags, |
1295 | varattno, /* attribute number to scan */ |
1296 | op_strategy, /* op's strategy */ |
1297 | op_righttype, /* strategy subtype */ |
1298 | ((OpExpr *) clause)->inputcollid, /* collation */ |
1299 | opfuncid, /* reg proc to use */ |
1300 | scanvalue); /* constant */ |
1301 | } |
1302 | else if (IsA(clause, RowCompareExpr)) |
1303 | { |
1304 | /* (indexkey, indexkey, ...) op (expression, expression, ...) */ |
1305 | RowCompareExpr *rc = (RowCompareExpr *) clause; |
1306 | ScanKey first_sub_key; |
1307 | int n_sub_key; |
1308 | ListCell *largs_cell; |
1309 | ListCell *rargs_cell; |
1310 | ListCell *opnos_cell; |
1311 | ListCell *collids_cell; |
1312 | |
1313 | Assert(!isorderby); |
1314 | |
1315 | first_sub_key = (ScanKey) |
1316 | palloc(list_length(rc->opnos) * sizeof(ScanKeyData)); |
1317 | n_sub_key = 0; |
1318 | |
1319 | /* Scan RowCompare columns and generate subsidiary ScanKey items */ |
1320 | forfour(largs_cell, rc->largs, rargs_cell, rc->rargs, |
1321 | opnos_cell, rc->opnos, collids_cell, rc->inputcollids) |
1322 | { |
1323 | ScanKey this_sub_key = &first_sub_key[n_sub_key]; |
1324 | int flags = SK_ROW_MEMBER; |
1325 | Datum scanvalue; |
1326 | Oid inputcollation; |
1327 | |
1328 | leftop = (Expr *) lfirst(largs_cell); |
1329 | rightop = (Expr *) lfirst(rargs_cell); |
1330 | opno = lfirst_oid(opnos_cell); |
1331 | inputcollation = lfirst_oid(collids_cell); |
1332 | |
1333 | /* |
1334 | * leftop should be the index key Var, possibly relabeled |
1335 | */ |
1336 | if (leftop && IsA(leftop, RelabelType)) |
1337 | leftop = ((RelabelType *) leftop)->arg; |
1338 | |
1339 | Assert(leftop != NULL); |
1340 | |
1341 | if (!(IsA(leftop, Var) && |
1342 | ((Var *) leftop)->varno == INDEX_VAR)) |
1343 | elog(ERROR, "indexqual doesn't have key on left side" ); |
1344 | |
1345 | varattno = ((Var *) leftop)->varattno; |
1346 | |
1347 | /* |
1348 | * We have to look up the operator's associated btree support |
1349 | * function |
1350 | */ |
1351 | if (index->rd_rel->relam != BTREE_AM_OID || |
1352 | varattno < 1 || varattno > indnkeyatts) |
1353 | elog(ERROR, "bogus RowCompare index qualification" ); |
1354 | opfamily = index->rd_opfamily[varattno - 1]; |
1355 | |
1356 | get_op_opfamily_properties(opno, opfamily, isorderby, |
1357 | &op_strategy, |
1358 | &op_lefttype, |
1359 | &op_righttype); |
1360 | |
1361 | if (op_strategy != rc->rctype) |
1362 | elog(ERROR, "RowCompare index qualification contains wrong operator" ); |
1363 | |
1364 | opfuncid = get_opfamily_proc(opfamily, |
1365 | op_lefttype, |
1366 | op_righttype, |
1367 | BTORDER_PROC); |
1368 | if (!RegProcedureIsValid(opfuncid)) |
1369 | elog(ERROR, "missing support function %d(%u,%u) in opfamily %u" , |
1370 | BTORDER_PROC, op_lefttype, op_righttype, opfamily); |
1371 | |
1372 | /* |
1373 | * rightop is the constant or variable comparison value |
1374 | */ |
1375 | if (rightop && IsA(rightop, RelabelType)) |
1376 | rightop = ((RelabelType *) rightop)->arg; |
1377 | |
1378 | Assert(rightop != NULL); |
1379 | |
1380 | if (IsA(rightop, Const)) |
1381 | { |
1382 | /* OK, simple constant comparison value */ |
1383 | scanvalue = ((Const *) rightop)->constvalue; |
1384 | if (((Const *) rightop)->constisnull) |
1385 | flags |= SK_ISNULL; |
1386 | } |
1387 | else |
1388 | { |
1389 | /* Need to treat this one as a runtime key */ |
1390 | if (n_runtime_keys >= max_runtime_keys) |
1391 | { |
1392 | if (max_runtime_keys == 0) |
1393 | { |
1394 | max_runtime_keys = 8; |
1395 | runtime_keys = (IndexRuntimeKeyInfo *) |
1396 | palloc(max_runtime_keys * sizeof(IndexRuntimeKeyInfo)); |
1397 | } |
1398 | else |
1399 | { |
1400 | max_runtime_keys *= 2; |
1401 | runtime_keys = (IndexRuntimeKeyInfo *) |
1402 | repalloc(runtime_keys, max_runtime_keys * sizeof(IndexRuntimeKeyInfo)); |
1403 | } |
1404 | } |
1405 | runtime_keys[n_runtime_keys].scan_key = this_sub_key; |
1406 | runtime_keys[n_runtime_keys].key_expr = |
1407 | ExecInitExpr(rightop, planstate); |
1408 | runtime_keys[n_runtime_keys].key_toastable = |
1409 | TypeIsToastable(op_righttype); |
1410 | n_runtime_keys++; |
1411 | scanvalue = (Datum) 0; |
1412 | } |
1413 | |
1414 | /* |
1415 | * initialize the subsidiary scan key's fields appropriately |
1416 | */ |
1417 | ScanKeyEntryInitialize(this_sub_key, |
1418 | flags, |
1419 | varattno, /* attribute number */ |
1420 | op_strategy, /* op's strategy */ |
1421 | op_righttype, /* strategy subtype */ |
1422 | inputcollation, /* collation */ |
1423 | opfuncid, /* reg proc to use */ |
1424 | scanvalue); /* constant */ |
1425 | n_sub_key++; |
1426 | } |
1427 | |
1428 | /* Mark the last subsidiary scankey correctly */ |
1429 | first_sub_key[n_sub_key - 1].sk_flags |= SK_ROW_END; |
1430 | |
1431 | /* |
1432 | * We don't use ScanKeyEntryInitialize for the header because it |
1433 | * isn't going to contain a valid sk_func pointer. |
1434 | */ |
1435 | MemSet(this_scan_key, 0, sizeof(ScanKeyData)); |
1436 | this_scan_key->sk_flags = SK_ROW_HEADER; |
1437 | this_scan_key->sk_attno = first_sub_key->sk_attno; |
1438 | this_scan_key->sk_strategy = rc->rctype; |
1439 | /* sk_subtype, sk_collation, sk_func not used in a header */ |
1440 | this_scan_key->sk_argument = PointerGetDatum(first_sub_key); |
1441 | } |
1442 | else if (IsA(clause, ScalarArrayOpExpr)) |
1443 | { |
1444 | /* indexkey op ANY (array-expression) */ |
1445 | ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause; |
1446 | int flags = 0; |
1447 | Datum scanvalue; |
1448 | |
1449 | Assert(!isorderby); |
1450 | |
1451 | Assert(saop->useOr); |
1452 | opno = saop->opno; |
1453 | opfuncid = saop->opfuncid; |
1454 | |
1455 | /* |
1456 | * leftop should be the index key Var, possibly relabeled |
1457 | */ |
1458 | leftop = (Expr *) linitial(saop->args); |
1459 | |
1460 | if (leftop && IsA(leftop, RelabelType)) |
1461 | leftop = ((RelabelType *) leftop)->arg; |
1462 | |
1463 | Assert(leftop != NULL); |
1464 | |
1465 | if (!(IsA(leftop, Var) && |
1466 | ((Var *) leftop)->varno == INDEX_VAR)) |
1467 | elog(ERROR, "indexqual doesn't have key on left side" ); |
1468 | |
1469 | varattno = ((Var *) leftop)->varattno; |
1470 | if (varattno < 1 || varattno > indnkeyatts) |
1471 | elog(ERROR, "bogus index qualification" ); |
1472 | |
1473 | /* |
1474 | * We have to look up the operator's strategy number. This |
1475 | * provides a cross-check that the operator does match the index. |
1476 | */ |
1477 | opfamily = index->rd_opfamily[varattno - 1]; |
1478 | |
1479 | get_op_opfamily_properties(opno, opfamily, isorderby, |
1480 | &op_strategy, |
1481 | &op_lefttype, |
1482 | &op_righttype); |
1483 | |
1484 | /* |
1485 | * rightop is the constant or variable array value |
1486 | */ |
1487 | rightop = (Expr *) lsecond(saop->args); |
1488 | |
1489 | if (rightop && IsA(rightop, RelabelType)) |
1490 | rightop = ((RelabelType *) rightop)->arg; |
1491 | |
1492 | Assert(rightop != NULL); |
1493 | |
1494 | if (index->rd_indam->amsearcharray) |
1495 | { |
1496 | /* Index AM will handle this like a simple operator */ |
1497 | flags |= SK_SEARCHARRAY; |
1498 | if (IsA(rightop, Const)) |
1499 | { |
1500 | /* OK, simple constant comparison value */ |
1501 | scanvalue = ((Const *) rightop)->constvalue; |
1502 | if (((Const *) rightop)->constisnull) |
1503 | flags |= SK_ISNULL; |
1504 | } |
1505 | else |
1506 | { |
1507 | /* Need to treat this one as a runtime key */ |
1508 | if (n_runtime_keys >= max_runtime_keys) |
1509 | { |
1510 | if (max_runtime_keys == 0) |
1511 | { |
1512 | max_runtime_keys = 8; |
1513 | runtime_keys = (IndexRuntimeKeyInfo *) |
1514 | palloc(max_runtime_keys * sizeof(IndexRuntimeKeyInfo)); |
1515 | } |
1516 | else |
1517 | { |
1518 | max_runtime_keys *= 2; |
1519 | runtime_keys = (IndexRuntimeKeyInfo *) |
1520 | repalloc(runtime_keys, max_runtime_keys * sizeof(IndexRuntimeKeyInfo)); |
1521 | } |
1522 | } |
1523 | runtime_keys[n_runtime_keys].scan_key = this_scan_key; |
1524 | runtime_keys[n_runtime_keys].key_expr = |
1525 | ExecInitExpr(rightop, planstate); |
1526 | |
1527 | /* |
1528 | * Careful here: the runtime expression is not of |
1529 | * op_righttype, but rather is an array of same; so |
1530 | * TypeIsToastable() isn't helpful. However, we can |
1531 | * assume that all array types are toastable. |
1532 | */ |
1533 | runtime_keys[n_runtime_keys].key_toastable = true; |
1534 | n_runtime_keys++; |
1535 | scanvalue = (Datum) 0; |
1536 | } |
1537 | } |
1538 | else |
1539 | { |
1540 | /* Executor has to expand the array value */ |
1541 | array_keys[n_array_keys].scan_key = this_scan_key; |
1542 | array_keys[n_array_keys].array_expr = |
1543 | ExecInitExpr(rightop, planstate); |
1544 | /* the remaining fields were zeroed by palloc0 */ |
1545 | n_array_keys++; |
1546 | scanvalue = (Datum) 0; |
1547 | } |
1548 | |
1549 | /* |
1550 | * initialize the scan key's fields appropriately |
1551 | */ |
1552 | ScanKeyEntryInitialize(this_scan_key, |
1553 | flags, |
1554 | varattno, /* attribute number to scan */ |
1555 | op_strategy, /* op's strategy */ |
1556 | op_righttype, /* strategy subtype */ |
1557 | saop->inputcollid, /* collation */ |
1558 | opfuncid, /* reg proc to use */ |
1559 | scanvalue); /* constant */ |
1560 | } |
1561 | else if (IsA(clause, NullTest)) |
1562 | { |
1563 | /* indexkey IS NULL or indexkey IS NOT NULL */ |
1564 | NullTest *ntest = (NullTest *) clause; |
1565 | int flags; |
1566 | |
1567 | Assert(!isorderby); |
1568 | |
1569 | /* |
1570 | * argument should be the index key Var, possibly relabeled |
1571 | */ |
1572 | leftop = ntest->arg; |
1573 | |
1574 | if (leftop && IsA(leftop, RelabelType)) |
1575 | leftop = ((RelabelType *) leftop)->arg; |
1576 | |
1577 | Assert(leftop != NULL); |
1578 | |
1579 | if (!(IsA(leftop, Var) && |
1580 | ((Var *) leftop)->varno == INDEX_VAR)) |
1581 | elog(ERROR, "NullTest indexqual has wrong key" ); |
1582 | |
1583 | varattno = ((Var *) leftop)->varattno; |
1584 | |
1585 | /* |
1586 | * initialize the scan key's fields appropriately |
1587 | */ |
1588 | switch (ntest->nulltesttype) |
1589 | { |
1590 | case IS_NULL: |
1591 | flags = SK_ISNULL | SK_SEARCHNULL; |
1592 | break; |
1593 | case IS_NOT_NULL: |
1594 | flags = SK_ISNULL | SK_SEARCHNOTNULL; |
1595 | break; |
1596 | default: |
1597 | elog(ERROR, "unrecognized nulltesttype: %d" , |
1598 | (int) ntest->nulltesttype); |
1599 | flags = 0; /* keep compiler quiet */ |
1600 | break; |
1601 | } |
1602 | |
1603 | ScanKeyEntryInitialize(this_scan_key, |
1604 | flags, |
1605 | varattno, /* attribute number to scan */ |
1606 | InvalidStrategy, /* no strategy */ |
1607 | InvalidOid, /* no strategy subtype */ |
1608 | InvalidOid, /* no collation */ |
1609 | InvalidOid, /* no reg proc for this */ |
1610 | (Datum) 0); /* constant */ |
1611 | } |
1612 | else |
1613 | elog(ERROR, "unsupported indexqual type: %d" , |
1614 | (int) nodeTag(clause)); |
1615 | } |
1616 | |
1617 | Assert(n_runtime_keys <= max_runtime_keys); |
1618 | |
1619 | /* Get rid of any unused arrays */ |
1620 | if (n_array_keys == 0) |
1621 | { |
1622 | pfree(array_keys); |
1623 | array_keys = NULL; |
1624 | } |
1625 | |
1626 | /* |
1627 | * Return info to our caller. |
1628 | */ |
1629 | *scanKeys = scan_keys; |
1630 | *numScanKeys = n_scan_keys; |
1631 | *runtimeKeys = runtime_keys; |
1632 | *numRuntimeKeys = n_runtime_keys; |
1633 | if (arrayKeys) |
1634 | { |
1635 | *arrayKeys = array_keys; |
1636 | *numArrayKeys = n_array_keys; |
1637 | } |
1638 | else if (n_array_keys != 0) |
1639 | elog(ERROR, "ScalarArrayOpExpr index qual found where not allowed" ); |
1640 | } |
1641 | |
1642 | /* ---------------------------------------------------------------- |
1643 | * Parallel Scan Support |
1644 | * ---------------------------------------------------------------- |
1645 | */ |
1646 | |
1647 | /* ---------------------------------------------------------------- |
1648 | * ExecIndexScanEstimate |
1649 | * |
1650 | * Compute the amount of space we'll need in the parallel |
1651 | * query DSM, and inform pcxt->estimator about our needs. |
1652 | * ---------------------------------------------------------------- |
1653 | */ |
1654 | void |
1655 | ExecIndexScanEstimate(IndexScanState *node, |
1656 | ParallelContext *pcxt) |
1657 | { |
1658 | EState *estate = node->ss.ps.state; |
1659 | |
1660 | node->iss_PscanLen = index_parallelscan_estimate(node->iss_RelationDesc, |
1661 | estate->es_snapshot); |
1662 | shm_toc_estimate_chunk(&pcxt->estimator, node->iss_PscanLen); |
1663 | shm_toc_estimate_keys(&pcxt->estimator, 1); |
1664 | } |
1665 | |
1666 | /* ---------------------------------------------------------------- |
1667 | * ExecIndexScanInitializeDSM |
1668 | * |
1669 | * Set up a parallel index scan descriptor. |
1670 | * ---------------------------------------------------------------- |
1671 | */ |
1672 | void |
1673 | ExecIndexScanInitializeDSM(IndexScanState *node, |
1674 | ParallelContext *pcxt) |
1675 | { |
1676 | EState *estate = node->ss.ps.state; |
1677 | ParallelIndexScanDesc piscan; |
1678 | |
1679 | piscan = shm_toc_allocate(pcxt->toc, node->iss_PscanLen); |
1680 | index_parallelscan_initialize(node->ss.ss_currentRelation, |
1681 | node->iss_RelationDesc, |
1682 | estate->es_snapshot, |
1683 | piscan); |
1684 | shm_toc_insert(pcxt->toc, node->ss.ps.plan->plan_node_id, piscan); |
1685 | node->iss_ScanDesc = |
1686 | index_beginscan_parallel(node->ss.ss_currentRelation, |
1687 | node->iss_RelationDesc, |
1688 | node->iss_NumScanKeys, |
1689 | node->iss_NumOrderByKeys, |
1690 | piscan); |
1691 | |
1692 | /* |
1693 | * If no run-time keys to calculate or they are ready, go ahead and pass |
1694 | * the scankeys to the index AM. |
1695 | */ |
1696 | if (node->iss_NumRuntimeKeys == 0 || node->iss_RuntimeKeysReady) |
1697 | index_rescan(node->iss_ScanDesc, |
1698 | node->iss_ScanKeys, node->iss_NumScanKeys, |
1699 | node->iss_OrderByKeys, node->iss_NumOrderByKeys); |
1700 | } |
1701 | |
1702 | /* ---------------------------------------------------------------- |
1703 | * ExecIndexScanReInitializeDSM |
1704 | * |
1705 | * Reset shared state before beginning a fresh scan. |
1706 | * ---------------------------------------------------------------- |
1707 | */ |
1708 | void |
1709 | ExecIndexScanReInitializeDSM(IndexScanState *node, |
1710 | ParallelContext *pcxt) |
1711 | { |
1712 | index_parallelrescan(node->iss_ScanDesc); |
1713 | } |
1714 | |
1715 | /* ---------------------------------------------------------------- |
1716 | * ExecIndexScanInitializeWorker |
1717 | * |
1718 | * Copy relevant information from TOC into planstate. |
1719 | * ---------------------------------------------------------------- |
1720 | */ |
1721 | void |
1722 | ExecIndexScanInitializeWorker(IndexScanState *node, |
1723 | ParallelWorkerContext *pwcxt) |
1724 | { |
1725 | ParallelIndexScanDesc piscan; |
1726 | |
1727 | piscan = shm_toc_lookup(pwcxt->toc, node->ss.ps.plan->plan_node_id, false); |
1728 | node->iss_ScanDesc = |
1729 | index_beginscan_parallel(node->ss.ss_currentRelation, |
1730 | node->iss_RelationDesc, |
1731 | node->iss_NumScanKeys, |
1732 | node->iss_NumOrderByKeys, |
1733 | piscan); |
1734 | |
1735 | /* |
1736 | * If no run-time keys to calculate or they are ready, go ahead and pass |
1737 | * the scankeys to the index AM. |
1738 | */ |
1739 | if (node->iss_NumRuntimeKeys == 0 || node->iss_RuntimeKeysReady) |
1740 | index_rescan(node->iss_ScanDesc, |
1741 | node->iss_ScanKeys, node->iss_NumScanKeys, |
1742 | node->iss_OrderByKeys, node->iss_NumOrderByKeys); |
1743 | } |
1744 | |