| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* nodeIndexonlyscan.c |
| 4 |
* Routines to support index-only scans |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/executor/nodeIndexonlyscan.c |
| 12 |
* |
| 13 |
*------------------------------------------------------------------------- |
| 14 |
*/ |
| 15 |
/* |
| 16 |
* INTERFACE ROUTINES |
| 17 |
* ExecIndexOnlyScan scans an index |
| 18 |
* IndexOnlyNext retrieve next tuple |
| 19 |
* ExecInitIndexOnlyScan creates and initializes state info. |
| 20 |
* ExecReScanIndexOnlyScan rescans the indexed relation. |
| 21 |
* ExecEndIndexOnlyScan releases all storage. |
| 22 |
* ExecIndexOnlyMarkPos marks scan position. |
| 23 |
* ExecIndexOnlyRestrPos restores scan position. |
| 24 |
* ExecIndexOnlyScanEstimate estimates DSM space needed for |
| 25 |
* parallel index-only scan |
| 26 |
* ExecIndexOnlyScanInitializeDSM initialize DSM for parallel |
| 27 |
* index-only scan |
| 28 |
* ExecIndexOnlyScanReInitializeDSM reinitialize DSM for fresh scan |
| 29 |
* ExecIndexOnlyScanInitializeWorker attach to DSM info in parallel worker |
| 30 |
*/ |
| 31 |
#include "postgres.h" |
| 32 |
|
| 33 |
#include "access/genam.h" |
| 34 |
#include "access/htup_details.h" |
| 35 |
#include "access/relscan.h" |
| 36 |
#include "access/tableam.h" |
| 37 |
#include "access/tupdesc.h" |
| 38 |
#include "access/visibilitymap.h" |
| 39 |
#include "catalog/pg_type.h" |
| 40 |
#include "executor/executor.h" |
| 41 |
#include "executor/nodeIndexonlyscan.h" |
| 42 |
#include "executor/nodeIndexscan.h" |
| 43 |
#include "miscadmin.h" |
| 44 |
#include "storage/bufmgr.h" |
| 45 |
#include "storage/predicate.h" |
| 46 |
#include "utils/builtins.h" |
| 47 |
#include "utils/rel.h" |
| 48 |
|
| 49 |
|
| 50 |
static TupleTableSlot *IndexOnlyNext(IndexOnlyScanState *node); |
| 51 |
static void StoreIndexTuple(IndexOnlyScanState *node, TupleTableSlot *slot, |
| 52 |
IndexScanDesc scandesc); |
| 53 |
|
| 54 |
|
| 55 |
/* ---------------------------------------------------------------- |
| 56 |
* IndexOnlyNext |
| 57 |
* |
| 58 |
* Retrieve a tuple from the IndexOnlyScan node's index. |
| 59 |
* ---------------------------------------------------------------- |
| 60 |
*/ |
| 61 |
static TupleTableSlot * |
| 62 |
IndexOnlyNext(IndexOnlyScanState *node) |
| 63 |
{ |
| 64 |
EState *estate; |
| 65 |
ExprContext *econtext; |
| 66 |
ScanDirection direction; |
| 67 |
IndexScanDesc scandesc; |
| 68 |
TupleTableSlot *slot; |
| 69 |
ItemPointer tid; |
| 70 |
|
| 71 |
/* |
| 72 |
* extract necessary information from index scan node |
| 73 |
*/ |
| 74 |
estate = node->ss.ps.state; |
| 75 |
|
| 76 |
/* |
| 77 |
* Determine which direction to scan the index in based on the plan's scan |
| 78 |
* direction and the current direction of execution. |
| 79 |
*/ |
| 80 |
direction = ScanDirectionCombine(estate->es_direction, |
| 81 |
((IndexOnlyScan *) node->ss.ps.plan)->indexorderdir); |
| 82 |
scandesc = node->ioss_ScanDesc; |
| 83 |
econtext = node->ss.ps.ps_ExprContext; |
| 84 |
slot = node->ss.ss_ScanTupleSlot; |
| 85 |
|
| 86 |
if (scandesc == NULL) |
| 87 |
{ |
| 88 |
/* |
| 89 |
* We reach here if the index only scan is not parallel, or if we're |
| 90 |
* serially executing an index only scan that was planned to be |
| 91 |
* parallel. |
| 92 |
*/ |
| 93 |
scandesc = index_beginscan(node->ss.ss_currentRelation, |
| 94 |
node->ioss_RelationDesc, |
| 95 |
estate->es_snapshot, |
| 96 |
&node->ioss_Instrument, |
| 97 |
node->ioss_NumScanKeys, |
| 98 |
node->ioss_NumOrderByKeys); |
| 99 |
|
| 100 |
node->ioss_ScanDesc = scandesc; |
| 101 |
|
| 102 |
|
| 103 |
/* Set it up for index-only scan */ |
| 104 |
node->ioss_ScanDesc->xs_want_itup = true; |
| 105 |
node->ioss_VMBuffer = InvalidBuffer; |
| 106 |
|
| 107 |
/* |
| 108 |
* If no run-time keys to calculate or they are ready, go ahead and |
| 109 |
* pass the scankeys to the index AM. |
| 110 |
*/ |
| 111 |
if (node->ioss_NumRuntimeKeys == 0 || node->ioss_RuntimeKeysReady) |
| 112 |
index_rescan(scandesc, |
| 113 |
node->ioss_ScanKeys, |
| 114 |
node->ioss_NumScanKeys, |
| 115 |
node->ioss_OrderByKeys, |
| 116 |
node->ioss_NumOrderByKeys); |
| 117 |
} |
| 118 |
|
| 119 |
/* |
| 120 |
* OK, now that we have what we need, fetch the next tuple. |
| 121 |
*/ |
| 122 |
while ((tid = index_getnext_tid(scandesc, direction)) != NULL) |
| 123 |
{ |
| 124 |
bool tuple_from_heap = false; |
| 125 |
|
| 126 |
CHECK_FOR_INTERRUPTS(); |
| 127 |
|
| 128 |
/* |
| 129 |
* We can skip the heap fetch if the TID references a heap page on |
| 130 |
* which all tuples are known visible to everybody. In any case, |
| 131 |
* we'll use the index tuple not the heap tuple as the data source. |
| 132 |
* |
| 133 |
* Note on Memory Ordering Effects: visibilitymap_get_status does not |
| 134 |
* lock the visibility map buffer, and therefore the result we read |
| 135 |
* here could be slightly stale. However, it can't be stale enough to |
| 136 |
* matter. |
| 137 |
* |
| 138 |
* We need to detect clearing a VM bit due to an insert right away, |
| 139 |
* because the tuple is present in the index page but not visible. The |
| 140 |
* reading of the TID by this scan (using a shared lock on the index |
| 141 |
* buffer) is serialized with the insert of the TID into the index |
| 142 |
* (using an exclusive lock on the index buffer). Because the VM bit |
| 143 |
* is cleared before updating the index, and locking/unlocking of the |
| 144 |
* index page acts as a full memory barrier, we are sure to see the |
| 145 |
* cleared bit if we see a recently-inserted TID. |
| 146 |
* |
| 147 |
* Deletes do not update the index page (only VACUUM will clear out |
| 148 |
* the TID), so the clearing of the VM bit by a delete is not |
| 149 |
* serialized with this test below, and we may see a value that is |
| 150 |
* significantly stale. However, we don't care about the delete right |
| 151 |
* away, because the tuple is still visible until the deleting |
| 152 |
* transaction commits or the statement ends (if it's our |
| 153 |
* transaction). In either case, the lock on the VM buffer will have |
| 154 |
* been released (acting as a write barrier) after clearing the bit. |
| 155 |
* And for us to have a snapshot that includes the deleting |
| 156 |
* transaction (making the tuple invisible), we must have acquired |
| 157 |
* ProcArrayLock after that time, acting as a read barrier. |
| 158 |
* |
| 159 |
* It's worth going through this complexity to avoid needing to lock |
| 160 |
* the VM buffer, which could cause significant contention. |
| 161 |
*/ |
| 162 |
if (!VM_ALL_VISIBLE(scandesc->heapRelation, |
| 163 |
ItemPointerGetBlockNumber(tid), |
| 164 |
&node->ioss_VMBuffer)) |
| 165 |
{ |
| 166 |
/* |
| 167 |
* Rats, we have to visit the heap to check visibility. |
| 168 |
*/ |
| 169 |
InstrCountTuples2(node, 1); |
| 170 |
if (!index_fetch_heap(scandesc, node->ioss_TableSlot)) |
| 171 |
continue; /* no visible tuple, try next index entry */ |
| 172 |
|
| 173 |
ExecClearTuple(node->ioss_TableSlot); |
| 174 |
|
| 175 |
/* |
| 176 |
* Only MVCC snapshots are supported here, so there should be no |
| 177 |
* need to keep following the HOT chain once a visible entry has |
| 178 |
* been found. If we did want to allow that, we'd need to keep |
| 179 |
* more state to remember not to call index_getnext_tid next time. |
| 180 |
*/ |
| 181 |
if (scandesc->xs_heap_continue) |
| 182 |
elog(ERROR, "non-MVCC snapshots are not supported in index-only scans"); |
| 183 |
|
| 184 |
/* |
| 185 |
* Note: at this point we are holding a pin on the heap page, as |
| 186 |
* recorded in scandesc->xs_cbuf. We could release that pin now, |
| 187 |
* but it's not clear whether it's a win to do so. The next index |
| 188 |
* entry might require a visit to the same heap page. |
| 189 |
*/ |
| 190 |
|
| 191 |
tuple_from_heap = true; |
| 192 |
} |
| 193 |
|
| 194 |
/* Fill the scan tuple slot with data from the index */ |
| 195 |
StoreIndexTuple(node, slot, scandesc); |
| 196 |
|
| 197 |
/* |
| 198 |
* If the index was lossy, we have to recheck the index quals. |
| 199 |
*/ |
| 200 |
if (scandesc->xs_recheck) |
| 201 |
{ |
| 202 |
econtext->ecxt_scantuple = slot; |
| 203 |
if (!ExecQualAndReset(node->recheckqual, econtext)) |
| 204 |
{ |
| 205 |
/* Fails recheck, so drop it and loop back for another */ |
| 206 |
InstrCountFiltered2(node, 1); |
| 207 |
continue; |
| 208 |
} |
| 209 |
} |
| 210 |
|
| 211 |
/* |
| 212 |
* We don't currently support rechecking ORDER BY distances. (In |
| 213 |
* principle, if the index can support retrieval of the originally |
| 214 |
* indexed value, it should be able to produce an exact distance |
| 215 |
* calculation too. So it's not clear that adding code here for |
| 216 |
* recheck/re-sort would be worth the trouble. But we should at least |
| 217 |
* throw an error if someone tries it.) |
| 218 |
*/ |
| 219 |
if (scandesc->numberOfOrderBys > 0 && scandesc->xs_recheckorderby) |
| 220 |
ereport(ERROR, |
| 221 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 222 |
errmsg("lossy distance functions are not supported in index-only scans"))); |
| 223 |
|
| 224 |
/* |
| 225 |
* If we didn't access the heap, then we'll need to take a predicate |
| 226 |
* lock explicitly, as if we had. For now we do that at page level. |
| 227 |
*/ |
| 228 |
if (!tuple_from_heap) |
| 229 |
PredicateLockPage(scandesc->heapRelation, |
| 230 |
ItemPointerGetBlockNumber(tid), |
| 231 |
estate->es_snapshot); |
| 232 |
|
| 233 |
return slot; |
| 234 |
} |
| 235 |
|
| 236 |
/* |
| 237 |
* if we get here it means the index scan failed so we are at the end of |
| 238 |
* the scan.. |
| 239 |
*/ |
| 240 |
return ExecClearTuple(slot); |
| 241 |
} |
| 242 |
|
| 243 |
/* |
| 244 |
* StoreIndexTuple |
| 245 |
* Fill the slot with the data the index AM returned. |
| 246 |
* |
| 247 |
* The data might be provided in either HeapTuple (xs_hitup) or IndexTuple |
| 248 |
* (xs_itup) format. Conceivably an index AM might fill both fields, in which |
| 249 |
* case we prefer the heap format, since it's probably a bit cheaper to fill a |
| 250 |
* slot from. |
| 251 |
* |
| 252 |
* At some point this might be generally-useful functionality, but |
| 253 |
* right now we don't need it elsewhere. |
| 254 |
*/ |
| 255 |
static void |
| 256 |
StoreIndexTuple(IndexOnlyScanState *node, TupleTableSlot *slot, |
| 257 |
IndexScanDesc scandesc) |
| 258 |
{ |
| 259 |
ExecClearTuple(slot); |
| 260 |
|
| 261 |
/* |
| 262 |
* We must deform the tuple using the tupdesc the index AM formed it with |
| 263 |
* (xs_hitupdesc or xs_itupdesc), not the slot's tupdesc. The datums |
| 264 |
* returned by the index AM must be binary compatible, but the descriptors |
| 265 |
* may align each column differently in certain rare cases. (Actually, |
| 266 |
* btree's "name" opclass stores cstring tuples that _aren't_ even binary |
| 267 |
* compatible, in the strictest sense. We directly handle that here.) |
| 268 |
*/ |
| 269 |
if (scandesc->xs_hitup) |
| 270 |
{ |
| 271 |
Assert(slot->tts_tupleDescriptor->natts == scandesc->xs_hitupdesc->natts); |
| 272 |
|
| 273 |
heap_deform_tuple(scandesc->xs_hitup, scandesc->xs_hitupdesc, |
| 274 |
slot->tts_values, slot->tts_isnull); |
| 275 |
} |
| 276 |
else if (scandesc->xs_itup) |
| 277 |
{ |
| 278 |
Assert(slot->tts_tupleDescriptor->natts == scandesc->xs_itupdesc->natts); |
| 279 |
|
| 280 |
index_deform_tuple(scandesc->xs_itup, scandesc->xs_itupdesc, |
| 281 |
slot->tts_values, slot->tts_isnull); |
| 282 |
|
| 283 |
/* |
| 284 |
* Copy all name columns stored as cstrings back into a NAMEDATALEN |
| 285 |
* byte sized allocation. We mark this branch as unlikely as |
| 286 |
* generally "name" is used only for the system catalogs and this |
| 287 |
* would have to be a user query running on those or some other user |
| 288 |
* table with an index on a name column. |
| 289 |
*/ |
| 290 |
if (unlikely(node->ioss_NameCStringAttNums != NULL)) |
| 291 |
{ |
| 292 |
int attcount = node->ioss_NameCStringCount; |
| 293 |
|
| 294 |
for (int idx = 0; idx < attcount; idx++) |
| 295 |
{ |
| 296 |
int attnum = node->ioss_NameCStringAttNums[idx]; |
| 297 |
Name name; |
| 298 |
|
| 299 |
/* skip null Datums */ |
| 300 |
if (slot->tts_isnull[attnum]) |
| 301 |
continue; |
| 302 |
|
| 303 |
/* |
| 304 |
* allocate the NAMEDATALEN and copy the datum into that |
| 305 |
* memory |
| 306 |
*/ |
| 307 |
name = (Name) MemoryContextAlloc(node->ss.ps.ps_ExprContext->ecxt_per_tuple_memory, |
| 308 |
NAMEDATALEN); |
| 309 |
|
| 310 |
/* use namestrcpy to zero-pad all trailing bytes */ |
| 311 |
namestrcpy(name, DatumGetCString(slot->tts_values[attnum])); |
| 312 |
slot->tts_values[attnum] = NameGetDatum(name); |
| 313 |
} |
| 314 |
} |
| 315 |
} |
| 316 |
else |
| 317 |
elog(ERROR, "no data returned for index-only scan"); |
| 318 |
|
| 319 |
ExecStoreVirtualTuple(slot); |
| 320 |
} |
| 321 |
|
| 322 |
/* |
| 323 |
* IndexOnlyRecheck -- access method routine to recheck a tuple in EvalPlanQual |
| 324 |
* |
| 325 |
* This can't really happen, since an index can't supply CTID which would |
| 326 |
* be necessary data for any potential EvalPlanQual target relation. If it |
| 327 |
* did happen, the EPQ code would pass us the wrong data, namely a heap |
| 328 |
* tuple not an index tuple. So throw an error. |
| 329 |
*/ |
| 330 |
static bool |
| 331 |
IndexOnlyRecheck(IndexOnlyScanState *node, TupleTableSlot *slot) |
| 332 |
{ |
| 333 |
elog(ERROR, "EvalPlanQual recheck is not supported in index-only scans"); |
| 334 |
return false; /* keep compiler quiet */ |
| 335 |
} |
| 336 |
|
| 337 |
/* ---------------------------------------------------------------- |
| 338 |
* ExecIndexOnlyScan(node) |
| 339 |
* ---------------------------------------------------------------- |
| 340 |
*/ |
| 341 |
static TupleTableSlot * |
| 342 |
ExecIndexOnlyScan(PlanState *pstate) |
| 343 |
{ |
| 344 |
IndexOnlyScanState *node = castNode(IndexOnlyScanState, pstate); |
| 345 |
|
| 346 |
/* |
| 347 |
* If we have runtime keys and they've not already been set up, do it now. |
| 348 |
*/ |
| 349 |
if (node->ioss_NumRuntimeKeys != 0 && !node->ioss_RuntimeKeysReady) |
| 350 |
ExecReScan((PlanState *) node); |
| 351 |
|
| 352 |
return ExecScan(&node->ss, |
| 353 |
(ExecScanAccessMtd) IndexOnlyNext, |
| 354 |
(ExecScanRecheckMtd) IndexOnlyRecheck); |
| 355 |
} |
| 356 |
|
| 357 |
/* ---------------------------------------------------------------- |
| 358 |
* ExecReScanIndexOnlyScan(node) |
| 359 |
* |
| 360 |
* Recalculates the values of any scan keys whose value depends on |
| 361 |
* information known at runtime, then rescans the indexed relation. |
| 362 |
* |
| 363 |
* Updating the scan key was formerly done separately in |
| 364 |
* ExecUpdateIndexScanKeys. Integrating it into ReScan makes |
| 365 |
* rescans of indices and relations/general streams more uniform. |
| 366 |
* ---------------------------------------------------------------- |
| 367 |
*/ |
| 368 |
void |
| 369 |
ExecReScanIndexOnlyScan(IndexOnlyScanState *node) |
| 370 |
{ |
| 371 |
/* |
| 372 |
* If we are doing runtime key calculations (ie, any of the index key |
| 373 |
* values weren't simple Consts), compute the new key values. But first, |
| 374 |
* reset the context so we don't leak memory as each outer tuple is |
| 375 |
* scanned. Note this assumes that we will recalculate *all* runtime keys |
| 376 |
* on each call. |
| 377 |
*/ |
| 378 |
if (node->ioss_NumRuntimeKeys != 0) |
| 379 |
{ |
| 380 |
ExprContext *econtext = node->ioss_RuntimeContext; |
| 381 |
|
| 382 |
ResetExprContext(econtext); |
| 383 |
ExecIndexEvalRuntimeKeys(econtext, |
| 384 |
node->ioss_RuntimeKeys, |
| 385 |
node->ioss_NumRuntimeKeys); |
| 386 |
} |
| 387 |
node->ioss_RuntimeKeysReady = true; |
| 388 |
|
| 389 |
/* reset index scan */ |
| 390 |
if (node->ioss_ScanDesc) |
| 391 |
index_rescan(node->ioss_ScanDesc, |
| 392 |
node->ioss_ScanKeys, node->ioss_NumScanKeys, |
| 393 |
node->ioss_OrderByKeys, node->ioss_NumOrderByKeys); |
| 394 |
|
| 395 |
ExecScanReScan(&node->ss); |
| 396 |
} |
| 397 |
|
| 398 |
|
| 399 |
/* ---------------------------------------------------------------- |
| 400 |
* ExecEndIndexOnlyScan |
| 401 |
* ---------------------------------------------------------------- |
| 402 |
*/ |
| 403 |
void |
| 404 |
ExecEndIndexOnlyScan(IndexOnlyScanState *node) |
| 405 |
{ |
| 406 |
Relation indexRelationDesc; |
| 407 |
IndexScanDesc indexScanDesc; |
| 408 |
|
| 409 |
/* |
| 410 |
* extract information from the node |
| 411 |
*/ |
| 412 |
indexRelationDesc = node->ioss_RelationDesc; |
| 413 |
indexScanDesc = node->ioss_ScanDesc; |
| 414 |
|
| 415 |
/* Release VM buffer pin, if any. */ |
| 416 |
if (node->ioss_VMBuffer != InvalidBuffer) |
| 417 |
{ |
| 418 |
ReleaseBuffer(node->ioss_VMBuffer); |
| 419 |
node->ioss_VMBuffer = InvalidBuffer; |
| 420 |
} |
| 421 |
|
| 422 |
/* |
| 423 |
* When ending a parallel worker, copy the statistics gathered by the |
| 424 |
* worker back into shared memory so that it can be picked up by the main |
| 425 |
* process to report in EXPLAIN ANALYZE |
| 426 |
*/ |
| 427 |
if (node->ioss_SharedInfo != NULL && IsParallelWorker()) |
| 428 |
{ |
| 429 |
IndexScanInstrumentation *winstrument; |
| 430 |
|
| 431 |
Assert(ParallelWorkerNumber < node->ioss_SharedInfo->num_workers); |
| 432 |
winstrument = &node->ioss_SharedInfo->winstrument[ParallelWorkerNumber]; |
| 433 |
|
| 434 |
/* |
| 435 |
* We have to accumulate the stats rather than performing a memcpy. |
| 436 |
* When a Gather/GatherMerge node finishes it will perform planner |
| 437 |
* shutdown on the workers. On rescan it will spin up new workers |
| 438 |
* which will have a new IndexOnlyScanState and zeroed stats. |
| 439 |
*/ |
| 440 |
winstrument->nsearches += node->ioss_Instrument.nsearches; |
| 441 |
} |
| 442 |
|
| 443 |
/* |
| 444 |
* close the index relation (no-op if we didn't open it) |
| 445 |
*/ |
| 446 |
if (indexScanDesc) |
| 447 |
index_endscan(indexScanDesc); |
| 448 |
if (indexRelationDesc) |
| 449 |
index_close(indexRelationDesc, NoLock); |
| 450 |
} |
| 451 |
|
| 452 |
/* ---------------------------------------------------------------- |
| 453 |
* ExecIndexOnlyMarkPos |
| 454 |
* |
| 455 |
* Note: we assume that no caller attempts to set a mark before having read |
| 456 |
* at least one tuple. Otherwise, ioss_ScanDesc might still be NULL. |
| 457 |
* ---------------------------------------------------------------- |
| 458 |
*/ |
| 459 |
void |
| 460 |
ExecIndexOnlyMarkPos(IndexOnlyScanState *node) |
| 461 |
{ |
| 462 |
EState *estate = node->ss.ps.state; |
| 463 |
EPQState *epqstate = estate->es_epq_active; |
| 464 |
|
| 465 |
if (epqstate != NULL) |
| 466 |
{ |
| 467 |
/* |
| 468 |
* We are inside an EvalPlanQual recheck. If a test tuple exists for |
| 469 |
* this relation, then we shouldn't access the index at all. We would |
| 470 |
* instead need to save, and later restore, the state of the |
| 471 |
* relsubs_done flag, so that re-fetching the test tuple is possible. |
| 472 |
* However, given the assumption that no caller sets a mark at the |
| 473 |
* start of the scan, we can only get here with relsubs_done[i] |
| 474 |
* already set, and so no state need be saved. |
| 475 |
*/ |
| 476 |
Index scanrelid = ((Scan *) node->ss.ps.plan)->scanrelid; |
| 477 |
|
| 478 |
Assert(scanrelid > 0); |
| 479 |
if (epqstate->relsubs_slot[scanrelid - 1] != NULL || |
| 480 |
epqstate->relsubs_rowmark[scanrelid - 1] != NULL) |
| 481 |
{ |
| 482 |
/* Verify the claim above */ |
| 483 |
if (!epqstate->relsubs_done[scanrelid - 1]) |
| 484 |
elog(ERROR, "unexpected ExecIndexOnlyMarkPos call in EPQ recheck"); |
| 485 |
return; |
| 486 |
} |
| 487 |
} |
| 488 |
|
| 489 |
index_markpos(node->ioss_ScanDesc); |
| 490 |
} |
| 491 |
|
| 492 |
/* ---------------------------------------------------------------- |
| 493 |
* ExecIndexOnlyRestrPos |
| 494 |
* ---------------------------------------------------------------- |
| 495 |
*/ |
| 496 |
void |
| 497 |
ExecIndexOnlyRestrPos(IndexOnlyScanState *node) |
| 498 |
{ |
| 499 |
EState *estate = node->ss.ps.state; |
| 500 |
EPQState *epqstate = estate->es_epq_active; |
| 501 |
|
| 502 |
if (estate->es_epq_active != NULL) |
| 503 |
{ |
| 504 |
/* See comments in ExecIndexMarkPos */ |
| 505 |
Index scanrelid = ((Scan *) node->ss.ps.plan)->scanrelid; |
| 506 |
|
| 507 |
Assert(scanrelid > 0); |
| 508 |
if (epqstate->relsubs_slot[scanrelid - 1] != NULL || |
| 509 |
epqstate->relsubs_rowmark[scanrelid - 1] != NULL) |
| 510 |
{ |
| 511 |
/* Verify the claim above */ |
| 512 |
if (!epqstate->relsubs_done[scanrelid - 1]) |
| 513 |
elog(ERROR, "unexpected ExecIndexOnlyRestrPos call in EPQ recheck"); |
| 514 |
return; |
| 515 |
} |
| 516 |
} |
| 517 |
|
| 518 |
index_restrpos(node->ioss_ScanDesc); |
| 519 |
} |
| 520 |
|
| 521 |
/* ---------------------------------------------------------------- |
| 522 |
* ExecInitIndexOnlyScan |
| 523 |
* |
| 524 |
* Initializes the index scan's state information, creates |
| 525 |
* scan keys, and opens the base and index relations. |
| 526 |
* |
| 527 |
* Note: index scans have 2 sets of state information because |
| 528 |
* we have to keep track of the base relation and the |
| 529 |
* index relation. |
| 530 |
* ---------------------------------------------------------------- |
| 531 |
*/ |
| 532 |
IndexOnlyScanState * |
| 533 |
ExecInitIndexOnlyScan(IndexOnlyScan *node, EState *estate, int eflags) |
| 534 |
{ |
| 535 |
IndexOnlyScanState *indexstate; |
| 536 |
Relation currentRelation; |
| 537 |
Relation indexRelation; |
| 538 |
LOCKMODE lockmode; |
| 539 |
TupleDesc tupDesc; |
| 540 |
int indnkeyatts; |
| 541 |
int namecount; |
| 542 |
|
| 543 |
/* |
| 544 |
* create state structure |
| 545 |
*/ |
| 546 |
indexstate = makeNode(IndexOnlyScanState); |
| 547 |
indexstate->ss.ps.plan = (Plan *) node; |
| 548 |
indexstate->ss.ps.state = estate; |
| 549 |
indexstate->ss.ps.ExecProcNode = ExecIndexOnlyScan; |
| 550 |
|
| 551 |
/* |
| 552 |
* Miscellaneous initialization |
| 553 |
* |
| 554 |
* create expression context for node |
| 555 |
*/ |
| 556 |
ExecAssignExprContext(estate, &indexstate->ss.ps); |
| 557 |
|
| 558 |
/* |
| 559 |
* open the scan relation |
| 560 |
*/ |
| 561 |
currentRelation = ExecOpenScanRelation(estate, node->scan.scanrelid, eflags); |
| 562 |
|
| 563 |
indexstate->ss.ss_currentRelation = currentRelation; |
| 564 |
indexstate->ss.ss_currentScanDesc = NULL; /* no heap scan here */ |
| 565 |
|
| 566 |
/* |
| 567 |
* Build the scan tuple type using the indextlist generated by the |
| 568 |
* planner. We use this, rather than the index's physical tuple |
| 569 |
* descriptor, because the latter contains storage column types not the |
| 570 |
* types of the original datums. (It's the AM's responsibility to return |
| 571 |
* suitable data anyway.) |
| 572 |
*/ |
| 573 |
tupDesc = ExecTypeFromTL(node->indextlist); |
| 574 |
ExecInitScanTupleSlot(estate, &indexstate->ss, tupDesc, |
| 575 |
&TTSOpsVirtual); |
| 576 |
|
| 577 |
/* |
| 578 |
* We need another slot, in a format that's suitable for the table AM, for |
| 579 |
* when we need to fetch a tuple from the table for rechecking visibility. |
| 580 |
*/ |
| 581 |
indexstate->ioss_TableSlot = |
| 582 |
ExecAllocTableSlot(&estate->es_tupleTable, |
| 583 |
RelationGetDescr(currentRelation), |
| 584 |
table_slot_callbacks(currentRelation)); |
| 585 |
|
| 586 |
/* |
| 587 |
* Initialize result type and projection info. The node's targetlist will |
| 588 |
* contain Vars with varno = INDEX_VAR, referencing the scan tuple. |
| 589 |
*/ |
| 590 |
ExecInitResultTypeTL(&indexstate->ss.ps); |
| 591 |
ExecAssignScanProjectionInfoWithVarno(&indexstate->ss, INDEX_VAR); |
| 592 |
|
| 593 |
/* |
| 594 |
* initialize child expressions |
| 595 |
* |
| 596 |
* Note: we don't initialize all of the indexorderby expression, only the |
| 597 |
* sub-parts corresponding to runtime keys (see below). |
| 598 |
*/ |
| 599 |
indexstate->ss.ps.qual = |
| 600 |
ExecInitQual(node->scan.plan.qual, (PlanState *) indexstate); |
| 601 |
indexstate->recheckqual = |
| 602 |
ExecInitQual(node->recheckqual, (PlanState *) indexstate); |
| 603 |
|
| 604 |
/* |
| 605 |
* If we are just doing EXPLAIN (ie, aren't going to run the plan), stop |
| 606 |
* here. This allows an index-advisor plugin to EXPLAIN a plan containing |
| 607 |
* references to nonexistent indexes. |
| 608 |
*/ |
| 609 |
if (eflags & EXEC_FLAG_EXPLAIN_ONLY) |
| 610 |
return indexstate; |
| 611 |
|
| 612 |
/* Open the index relation. */ |
| 613 |
lockmode = exec_rt_fetch(node->scan.scanrelid, estate)->rellockmode; |
| 614 |
indexRelation = index_open(node->indexid, lockmode); |
| 615 |
indexstate->ioss_RelationDesc = indexRelation; |
| 616 |
|
| 617 |
/* |
| 618 |
* Initialize index-specific scan state |
| 619 |
*/ |
| 620 |
indexstate->ioss_RuntimeKeysReady = false; |
| 621 |
indexstate->ioss_RuntimeKeys = NULL; |
| 622 |
indexstate->ioss_NumRuntimeKeys = 0; |
| 623 |
|
| 624 |
/* |
| 625 |
* build the index scan keys from the index qualification |
| 626 |
*/ |
| 627 |
ExecIndexBuildScanKeys((PlanState *) indexstate, |
| 628 |
indexRelation, |
| 629 |
node->indexqual, |
| 630 |
false, |
| 631 |
&indexstate->ioss_ScanKeys, |
| 632 |
&indexstate->ioss_NumScanKeys, |
| 633 |
&indexstate->ioss_RuntimeKeys, |
| 634 |
&indexstate->ioss_NumRuntimeKeys, |
| 635 |
NULL, /* no ArrayKeys */ |
| 636 |
NULL); |
| 637 |
|
| 638 |
/* |
| 639 |
* any ORDER BY exprs have to be turned into scankeys in the same way |
| 640 |
*/ |
| 641 |
ExecIndexBuildScanKeys((PlanState *) indexstate, |
| 642 |
indexRelation, |
| 643 |
node->indexorderby, |
| 644 |
true, |
| 645 |
&indexstate->ioss_OrderByKeys, |
| 646 |
&indexstate->ioss_NumOrderByKeys, |
| 647 |
&indexstate->ioss_RuntimeKeys, |
| 648 |
&indexstate->ioss_NumRuntimeKeys, |
| 649 |
NULL, /* no ArrayKeys */ |
| 650 |
NULL); |
| 651 |
|
| 652 |
/* |
| 653 |
* If we have runtime keys, we need an ExprContext to evaluate them. The |
| 654 |
* node's standard context won't do because we want to reset that context |
| 655 |
* for every tuple. So, build another context just like the other one... |
| 656 |
* -tgl 7/11/00 |
| 657 |
*/ |
| 658 |
if (indexstate->ioss_NumRuntimeKeys != 0) |
| 659 |
{ |
| 660 |
ExprContext *stdecontext = indexstate->ss.ps.ps_ExprContext; |
| 661 |
|
| 662 |
ExecAssignExprContext(estate, &indexstate->ss.ps); |
| 663 |
indexstate->ioss_RuntimeContext = indexstate->ss.ps.ps_ExprContext; |
| 664 |
indexstate->ss.ps.ps_ExprContext = stdecontext; |
| 665 |
} |
| 666 |
else |
| 667 |
{ |
| 668 |
indexstate->ioss_RuntimeContext = NULL; |
| 669 |
} |
| 670 |
|
| 671 |
indexstate->ioss_NameCStringAttNums = NULL; |
| 672 |
indnkeyatts = indexRelation->rd_index->indnkeyatts; |
| 673 |
namecount = 0; |
| 674 |
|
| 675 |
/* |
| 676 |
* The "name" type for btree uses text_ops which results in storing |
| 677 |
* cstrings in the indexed keys rather than names. Here we detect that in |
| 678 |
* a generic way in case other index AMs want to do the same optimization. |
| 679 |
* Check for opclasses with an opcintype of NAMEOID and an index tuple |
| 680 |
* descriptor with CSTRINGOID. If any of these are found, create an array |
| 681 |
* marking the index attribute number of each of them. StoreIndexTuple() |
| 682 |
* handles copying the name Datums into a NAMEDATALEN-byte allocation. |
| 683 |
*/ |
| 684 |
|
| 685 |
/* First, count the number of such index keys */ |
| 686 |
for (int attnum = 0; attnum < indnkeyatts; attnum++) |
| 687 |
{ |
| 688 |
if (TupleDescAttr(indexRelation->rd_att, attnum)->atttypid == CSTRINGOID && |
| 689 |
indexRelation->rd_opcintype[attnum] == NAMEOID) |
| 690 |
namecount++; |
| 691 |
} |
| 692 |
|
| 693 |
if (namecount > 0) |
| 694 |
{ |
| 695 |
int idx = 0; |
| 696 |
|
| 697 |
/* |
| 698 |
* Now create an array to mark the attribute numbers of the keys that |
| 699 |
* need to be converted from cstring to name. |
| 700 |
*/ |
| 701 |
indexstate->ioss_NameCStringAttNums = (AttrNumber *) |
| 702 |
palloc(sizeof(AttrNumber) * namecount); |
| 703 |
|
| 704 |
for (int attnum = 0; attnum < indnkeyatts; attnum++) |
| 705 |
{ |
| 706 |
if (TupleDescAttr(indexRelation->rd_att, attnum)->atttypid == CSTRINGOID && |
| 707 |
indexRelation->rd_opcintype[attnum] == NAMEOID) |
| 708 |
indexstate->ioss_NameCStringAttNums[idx++] = (AttrNumber) attnum; |
| 709 |
} |
| 710 |
} |
| 711 |
|
| 712 |
indexstate->ioss_NameCStringCount = namecount; |
| 713 |
|
| 714 |
/* |
| 715 |
* all done. |
| 716 |
*/ |
| 717 |
return indexstate; |
| 718 |
} |
| 719 |
|
| 720 |
/* ---------------------------------------------------------------- |
| 721 |
* Parallel Index-only Scan Support |
| 722 |
* ---------------------------------------------------------------- |
| 723 |
*/ |
| 724 |
|
| 725 |
/* ---------------------------------------------------------------- |
| 726 |
* ExecIndexOnlyScanEstimate |
| 727 |
* |
| 728 |
* Compute the amount of space we'll need in the parallel |
| 729 |
* query DSM, and inform pcxt->estimator about our needs. |
| 730 |
* ---------------------------------------------------------------- |
| 731 |
*/ |
| 732 |
void |
| 733 |
ExecIndexOnlyScanEstimate(IndexOnlyScanState *node, |
| 734 |
ParallelContext *pcxt) |
| 735 |
{ |
| 736 |
EState *estate = node->ss.ps.state; |
| 737 |
bool instrument = (node->ss.ps.instrument != NULL); |
| 738 |
bool parallel_aware = node->ss.ps.plan->parallel_aware; |
| 739 |
|
| 740 |
if (!instrument && !parallel_aware) |
| 741 |
{ |
| 742 |
/* No DSM required by the scan */ |
| 743 |
return; |
| 744 |
} |
| 745 |
|
| 746 |
node->ioss_PscanLen = index_parallelscan_estimate(node->ioss_RelationDesc, |
| 747 |
node->ioss_NumScanKeys, |
| 748 |
node->ioss_NumOrderByKeys, |
| 749 |
estate->es_snapshot, |
| 750 |
instrument, parallel_aware, |
| 751 |
pcxt->nworkers); |
| 752 |
shm_toc_estimate_chunk(&pcxt->estimator, node->ioss_PscanLen); |
| 753 |
shm_toc_estimate_keys(&pcxt->estimator, 1); |
| 754 |
} |
| 755 |
|
| 756 |
/* ---------------------------------------------------------------- |
| 757 |
* ExecIndexOnlyScanInitializeDSM |
| 758 |
* |
| 759 |
* Set up a parallel index-only scan descriptor. |
| 760 |
* ---------------------------------------------------------------- |
| 761 |
*/ |
| 762 |
void |
| 763 |
ExecIndexOnlyScanInitializeDSM(IndexOnlyScanState *node, |
| 764 |
ParallelContext *pcxt) |
| 765 |
{ |
| 766 |
EState *estate = node->ss.ps.state; |
| 767 |
ParallelIndexScanDesc piscan; |
| 768 |
bool instrument = node->ss.ps.instrument != NULL; |
| 769 |
bool parallel_aware = node->ss.ps.plan->parallel_aware; |
| 770 |
|
| 771 |
if (!instrument && !parallel_aware) |
| 772 |
{ |
| 773 |
/* No DSM required by the scan */ |
| 774 |
return; |
| 775 |
} |
| 776 |
|
| 777 |
piscan = shm_toc_allocate(pcxt->toc, node->ioss_PscanLen); |
| 778 |
index_parallelscan_initialize(node->ss.ss_currentRelation, |
| 779 |
node->ioss_RelationDesc, |
| 780 |
estate->es_snapshot, |
| 781 |
instrument, parallel_aware, pcxt->nworkers, |
| 782 |
&node->ioss_SharedInfo, piscan); |
| 783 |
shm_toc_insert(pcxt->toc, node->ss.ps.plan->plan_node_id, piscan); |
| 784 |
|
| 785 |
if (!parallel_aware) |
| 786 |
{ |
| 787 |
/* Only here to initialize SharedInfo in DSM */ |
| 788 |
return; |
| 789 |
} |
| 790 |
|
| 791 |
node->ioss_ScanDesc = |
| 792 |
index_beginscan_parallel(node->ss.ss_currentRelation, |
| 793 |
node->ioss_RelationDesc, |
| 794 |
&node->ioss_Instrument, |
| 795 |
node->ioss_NumScanKeys, |
| 796 |
node->ioss_NumOrderByKeys, |
| 797 |
piscan); |
| 798 |
node->ioss_ScanDesc->xs_want_itup = true; |
| 799 |
node->ioss_VMBuffer = InvalidBuffer; |
| 800 |
|
| 801 |
/* |
| 802 |
* If no run-time keys to calculate or they are ready, go ahead and pass |
| 803 |
* the scankeys to the index AM. |
| 804 |
*/ |
| 805 |
if (node->ioss_NumRuntimeKeys == 0 || node->ioss_RuntimeKeysReady) |
| 806 |
index_rescan(node->ioss_ScanDesc, |
| 807 |
node->ioss_ScanKeys, node->ioss_NumScanKeys, |
| 808 |
node->ioss_OrderByKeys, node->ioss_NumOrderByKeys); |
| 809 |
} |
| 810 |
|
| 811 |
/* ---------------------------------------------------------------- |
| 812 |
* ExecIndexOnlyScanReInitializeDSM |
| 813 |
* |
| 814 |
* Reset shared state before beginning a fresh scan. |
| 815 |
* ---------------------------------------------------------------- |
| 816 |
*/ |
| 817 |
void |
| 818 |
ExecIndexOnlyScanReInitializeDSM(IndexOnlyScanState *node, |
| 819 |
ParallelContext *pcxt) |
| 820 |
{ |
| 821 |
Assert(node->ss.ps.plan->parallel_aware); |
| 822 |
index_parallelrescan(node->ioss_ScanDesc); |
| 823 |
} |
| 824 |
|
| 825 |
/* ---------------------------------------------------------------- |
| 826 |
* ExecIndexOnlyScanInitializeWorker |
| 827 |
* |
| 828 |
* Copy relevant information from TOC into planstate. |
| 829 |
* ---------------------------------------------------------------- |
| 830 |
*/ |
| 831 |
void |
| 832 |
ExecIndexOnlyScanInitializeWorker(IndexOnlyScanState *node, |
| 833 |
ParallelWorkerContext *pwcxt) |
| 834 |
{ |
| 835 |
ParallelIndexScanDesc piscan; |
| 836 |
bool instrument = node->ss.ps.instrument != NULL; |
| 837 |
bool parallel_aware = node->ss.ps.plan->parallel_aware; |
| 838 |
|
| 839 |
if (!instrument && !parallel_aware) |
| 840 |
{ |
| 841 |
/* No DSM required by the scan */ |
| 842 |
return; |
| 843 |
} |
| 844 |
|
| 845 |
piscan = shm_toc_lookup(pwcxt->toc, node->ss.ps.plan->plan_node_id, false); |
| 846 |
|
| 847 |
if (instrument) |
| 848 |
node->ioss_SharedInfo = (SharedIndexScanInstrumentation *) |
| 849 |
OffsetToPointer(piscan, piscan->ps_offset_ins); |
| 850 |
|
| 851 |
if (!parallel_aware) |
| 852 |
{ |
| 853 |
/* Only here to set up worker node's SharedInfo */ |
| 854 |
return; |
| 855 |
} |
| 856 |
|
| 857 |
node->ioss_ScanDesc = |
| 858 |
index_beginscan_parallel(node->ss.ss_currentRelation, |
| 859 |
node->ioss_RelationDesc, |
| 860 |
&node->ioss_Instrument, |
| 861 |
node->ioss_NumScanKeys, |
| 862 |
node->ioss_NumOrderByKeys, |
| 863 |
piscan); |
| 864 |
node->ioss_ScanDesc->xs_want_itup = true; |
| 865 |
|
| 866 |
/* |
| 867 |
* If no run-time keys to calculate or they are ready, go ahead and pass |
| 868 |
* the scankeys to the index AM. |
| 869 |
*/ |
| 870 |
if (node->ioss_NumRuntimeKeys == 0 || node->ioss_RuntimeKeysReady) |
| 871 |
index_rescan(node->ioss_ScanDesc, |
| 872 |
node->ioss_ScanKeys, node->ioss_NumScanKeys, |
| 873 |
node->ioss_OrderByKeys, node->ioss_NumOrderByKeys); |
| 874 |
} |
| 875 |
|
| 876 |
/* ---------------------------------------------------------------- |
| 877 |
* ExecIndexOnlyScanRetrieveInstrumentation |
| 878 |
* |
| 879 |
* Transfer index-only scan statistics from DSM to private memory. |
| 880 |
* ---------------------------------------------------------------- |
| 881 |
*/ |
| 882 |
void |
| 883 |
ExecIndexOnlyScanRetrieveInstrumentation(IndexOnlyScanState *node) |
| 884 |
{ |
| 885 |
SharedIndexScanInstrumentation *SharedInfo = node->ioss_SharedInfo; |
| 886 |
size_t size; |
| 887 |
|
| 888 |
if (SharedInfo == NULL) |
| 889 |
return; |
| 890 |
|
| 891 |
/* Create a copy of SharedInfo in backend-local memory */ |
| 892 |
size = offsetof(SharedIndexScanInstrumentation, winstrument) + |
| 893 |
SharedInfo->num_workers * sizeof(IndexScanInstrumentation); |
| 894 |
node->ioss_SharedInfo = palloc(size); |
| 895 |
memcpy(node->ioss_SharedInfo, SharedInfo, size); |
| 896 |
} |