| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* execProcnode.c |
| 4 |
* contains dispatch functions which call the appropriate "initialize", |
| 5 |
* "get a tuple", and "cleanup" routines for the given node type. |
| 6 |
* If the node has children, then it will presumably call ExecInitNode, |
| 7 |
* ExecProcNode, or ExecEndNode on its subnodes and do the appropriate |
| 8 |
* processing. |
| 9 |
* |
| 10 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 11 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 12 |
* |
| 13 |
* |
| 14 |
* IDENTIFICATION |
| 15 |
* src/backend/executor/execProcnode.c |
| 16 |
* |
| 17 |
*------------------------------------------------------------------------- |
| 18 |
*/ |
| 19 |
/* |
| 20 |
* NOTES |
| 21 |
* This used to be three files. It is now all combined into |
| 22 |
* one file so that it is easier to keep the dispatch routines |
| 23 |
* in sync when new nodes are added. |
| 24 |
* |
| 25 |
* EXAMPLE |
| 26 |
* Suppose we want the age of the manager of the shoe department and |
| 27 |
* the number of employees in that department. So we have the query: |
| 28 |
* |
| 29 |
* select DEPT.no_emps, EMP.age |
| 30 |
* from DEPT, EMP |
| 31 |
* where EMP.name = DEPT.mgr and |
| 32 |
* DEPT.name = "shoe" |
| 33 |
* |
| 34 |
* Suppose the planner gives us the following plan: |
| 35 |
* |
| 36 |
* Nest Loop (DEPT.mgr = EMP.name) |
| 37 |
* / \ |
| 38 |
* / \ |
| 39 |
* Seq Scan Seq Scan |
| 40 |
* DEPT EMP |
| 41 |
* (name = "shoe") |
| 42 |
* |
| 43 |
* ExecutorStart() is called first. |
| 44 |
* It calls InitPlan() which calls ExecInitNode() on |
| 45 |
* the root of the plan -- the nest loop node. |
| 46 |
* |
| 47 |
* * ExecInitNode() notices that it is looking at a nest loop and |
| 48 |
* as the code below demonstrates, it calls ExecInitNestLoop(). |
| 49 |
* Eventually this calls ExecInitNode() on the right and left subplans |
| 50 |
* and so forth until the entire plan is initialized. The result |
| 51 |
* of ExecInitNode() is a plan state tree built with the same structure |
| 52 |
* as the underlying plan tree. |
| 53 |
* |
| 54 |
* * Then when ExecutorRun() is called, it calls ExecutePlan() which calls |
| 55 |
* ExecProcNode() repeatedly on the top node of the plan state tree. |
| 56 |
* Each time this happens, ExecProcNode() will end up calling |
| 57 |
* ExecNestLoop(), which calls ExecProcNode() on its subplans. |
| 58 |
* Each of these subplans is a sequential scan so ExecSeqScan() is |
| 59 |
* called. The slots returned by ExecSeqScan() may contain |
| 60 |
* tuples which contain the attributes ExecNestLoop() uses to |
| 61 |
* form the tuples it returns. |
| 62 |
* |
| 63 |
* * Eventually ExecSeqScan() stops returning tuples and the nest |
| 64 |
* loop join ends. Lastly, ExecutorEnd() calls ExecEndNode() which |
| 65 |
* calls ExecEndNestLoop() which in turn calls ExecEndNode() on |
| 66 |
* its subplans which result in ExecEndSeqScan(). |
| 67 |
* |
| 68 |
* This should show how the executor works by having |
| 69 |
* ExecInitNode(), ExecProcNode() and ExecEndNode() dispatch |
| 70 |
* their work to the appropriate node support routines which may |
| 71 |
* in turn call these routines themselves on their subplans. |
| 72 |
*/ |
| 73 |
#include "postgres.h" |
| 74 |
|
| 75 |
#include "executor/executor.h" |
| 76 |
#include "executor/nodeAgg.h" |
| 77 |
#include "executor/nodeAppend.h" |
| 78 |
#include "executor/nodeBitmapAnd.h" |
| 79 |
#include "executor/nodeBitmapHeapscan.h" |
| 80 |
#include "executor/nodeBitmapIndexscan.h" |
| 81 |
#include "executor/nodeBitmapOr.h" |
| 82 |
#include "executor/nodeCtescan.h" |
| 83 |
#include "executor/nodeCustom.h" |
| 84 |
#include "executor/nodeForeignscan.h" |
| 85 |
#include "executor/nodeFunctionscan.h" |
| 86 |
#include "executor/nodeGather.h" |
| 87 |
#include "executor/nodeGatherMerge.h" |
| 88 |
#include "executor/nodeGroup.h" |
| 89 |
#include "executor/nodeHash.h" |
| 90 |
#include "executor/nodeHashjoin.h" |
| 91 |
#include "executor/nodeIncrementalSort.h" |
| 92 |
#include "executor/nodeIndexonlyscan.h" |
| 93 |
#include "executor/nodeIndexscan.h" |
| 94 |
#include "executor/nodeLimit.h" |
| 95 |
#include "executor/nodeLockRows.h" |
| 96 |
#include "executor/nodeMaterial.h" |
| 97 |
#include "executor/nodeMemoize.h" |
| 98 |
#include "executor/nodeMergeAppend.h" |
| 99 |
#include "executor/nodeMergejoin.h" |
| 100 |
#include "executor/nodeModifyTable.h" |
| 101 |
#include "executor/nodeNamedtuplestorescan.h" |
| 102 |
#include "executor/nodeNestloop.h" |
| 103 |
#include "executor/nodeProjectSet.h" |
| 104 |
#include "executor/nodeRecursiveunion.h" |
| 105 |
#include "executor/nodeResult.h" |
| 106 |
#include "executor/nodeSamplescan.h" |
| 107 |
#include "executor/nodeSeqscan.h" |
| 108 |
#include "executor/nodeSetOp.h" |
| 109 |
#include "executor/nodeSort.h" |
| 110 |
#include "executor/nodeSubplan.h" |
| 111 |
#include "executor/nodeSubqueryscan.h" |
| 112 |
#include "executor/nodeTableFuncscan.h" |
| 113 |
#include "executor/nodeTidrangescan.h" |
| 114 |
#include "executor/nodeTidscan.h" |
| 115 |
#include "executor/nodeUnique.h" |
| 116 |
#include "executor/nodeValuesscan.h" |
| 117 |
#include "executor/nodeWindowAgg.h" |
| 118 |
#include "executor/nodeWorktablescan.h" |
| 119 |
#include "miscadmin.h" |
| 120 |
#include "nodes/nodeFuncs.h" |
| 121 |
|
| 122 |
static TupleTableSlot *ExecProcNodeFirst(PlanState *node); |
| 123 |
static TupleTableSlot *ExecProcNodeInstr(PlanState *node); |
| 124 |
static bool ExecShutdownNode_walker(PlanState *node, void *context); |
| 125 |
|
| 126 |
|
| 127 |
/* ------------------------------------------------------------------------ |
| 128 |
* ExecInitNode |
| 129 |
* |
| 130 |
* Recursively initializes all the nodes in the plan tree rooted |
| 131 |
* at 'node'. |
| 132 |
* |
| 133 |
* Inputs: |
| 134 |
* 'node' is the current node of the plan produced by the query planner |
| 135 |
* 'estate' is the shared execution state for the plan tree |
| 136 |
* 'eflags' is a bitwise OR of flag bits described in executor.h |
| 137 |
* |
| 138 |
* Returns a PlanState node corresponding to the given Plan node. |
| 139 |
* ------------------------------------------------------------------------ |
| 140 |
*/ |
| 141 |
PlanState * |
| 142 |
ExecInitNode(Plan *node, EState *estate, int eflags) |
| 143 |
{ |
| 144 |
PlanState *result; |
| 145 |
List *subps; |
| 146 |
ListCell *l; |
| 147 |
|
| 148 |
/* |
| 149 |
* do nothing when we get to the end of a leaf on tree. |
| 150 |
*/ |
| 151 |
if (node == NULL) |
| 152 |
return NULL; |
| 153 |
|
| 154 |
/* |
| 155 |
* Make sure there's enough stack available. Need to check here, in |
| 156 |
* addition to ExecProcNode() (via ExecProcNodeFirst()), to ensure the |
| 157 |
* stack isn't overrun while initializing the node tree. |
| 158 |
*/ |
| 159 |
check_stack_depth(); |
| 160 |
|
| 161 |
switch (nodeTag(node)) |
| 162 |
{ |
| 163 |
/* |
| 164 |
* control nodes |
| 165 |
*/ |
| 166 |
case T_Result: |
| 167 |
result = (PlanState *) ExecInitResult((Result *) node, |
| 168 |
estate, eflags); |
| 169 |
break; |
| 170 |
|
| 171 |
case T_ProjectSet: |
| 172 |
result = (PlanState *) ExecInitProjectSet((ProjectSet *) node, |
| 173 |
estate, eflags); |
| 174 |
break; |
| 175 |
|
| 176 |
case T_ModifyTable: |
| 177 |
result = (PlanState *) ExecInitModifyTable((ModifyTable *) node, |
| 178 |
estate, eflags); |
| 179 |
break; |
| 180 |
|
| 181 |
case T_Append: |
| 182 |
result = (PlanState *) ExecInitAppend((Append *) node, |
| 183 |
estate, eflags); |
| 184 |
break; |
| 185 |
|
| 186 |
case T_MergeAppend: |
| 187 |
result = (PlanState *) ExecInitMergeAppend((MergeAppend *) node, |
| 188 |
estate, eflags); |
| 189 |
break; |
| 190 |
|
| 191 |
case T_RecursiveUnion: |
| 192 |
result = (PlanState *) ExecInitRecursiveUnion((RecursiveUnion *) node, |
| 193 |
estate, eflags); |
| 194 |
break; |
| 195 |
|
| 196 |
case T_BitmapAnd: |
| 197 |
result = (PlanState *) ExecInitBitmapAnd((BitmapAnd *) node, |
| 198 |
estate, eflags); |
| 199 |
break; |
| 200 |
|
| 201 |
case T_BitmapOr: |
| 202 |
result = (PlanState *) ExecInitBitmapOr((BitmapOr *) node, |
| 203 |
estate, eflags); |
| 204 |
break; |
| 205 |
|
| 206 |
/* |
| 207 |
* scan nodes |
| 208 |
*/ |
| 209 |
case T_SeqScan: |
| 210 |
result = (PlanState *) ExecInitSeqScan((SeqScan *) node, |
| 211 |
estate, eflags); |
| 212 |
break; |
| 213 |
|
| 214 |
case T_SampleScan: |
| 215 |
result = (PlanState *) ExecInitSampleScan((SampleScan *) node, |
| 216 |
estate, eflags); |
| 217 |
break; |
| 218 |
|
| 219 |
case T_IndexScan: |
| 220 |
result = (PlanState *) ExecInitIndexScan((IndexScan *) node, |
| 221 |
estate, eflags); |
| 222 |
break; |
| 223 |
|
| 224 |
case T_IndexOnlyScan: |
| 225 |
result = (PlanState *) ExecInitIndexOnlyScan((IndexOnlyScan *) node, |
| 226 |
estate, eflags); |
| 227 |
break; |
| 228 |
|
| 229 |
case T_BitmapIndexScan: |
| 230 |
result = (PlanState *) ExecInitBitmapIndexScan((BitmapIndexScan *) node, |
| 231 |
estate, eflags); |
| 232 |
break; |
| 233 |
|
| 234 |
case T_BitmapHeapScan: |
| 235 |
result = (PlanState *) ExecInitBitmapHeapScan((BitmapHeapScan *) node, |
| 236 |
estate, eflags); |
| 237 |
break; |
| 238 |
|
| 239 |
case T_TidScan: |
| 240 |
result = (PlanState *) ExecInitTidScan((TidScan *) node, |
| 241 |
estate, eflags); |
| 242 |
break; |
| 243 |
|
| 244 |
case T_TidRangeScan: |
| 245 |
result = (PlanState *) ExecInitTidRangeScan((TidRangeScan *) node, |
| 246 |
estate, eflags); |
| 247 |
break; |
| 248 |
|
| 249 |
case T_SubqueryScan: |
| 250 |
result = (PlanState *) ExecInitSubqueryScan((SubqueryScan *) node, |
| 251 |
estate, eflags); |
| 252 |
break; |
| 253 |
|
| 254 |
case T_FunctionScan: |
| 255 |
result = (PlanState *) ExecInitFunctionScan((FunctionScan *) node, |
| 256 |
estate, eflags); |
| 257 |
break; |
| 258 |
|
| 259 |
case T_TableFuncScan: |
| 260 |
result = (PlanState *) ExecInitTableFuncScan((TableFuncScan *) node, |
| 261 |
estate, eflags); |
| 262 |
break; |
| 263 |
|
| 264 |
case T_ValuesScan: |
| 265 |
result = (PlanState *) ExecInitValuesScan((ValuesScan *) node, |
| 266 |
estate, eflags); |
| 267 |
break; |
| 268 |
|
| 269 |
case T_CteScan: |
| 270 |
result = (PlanState *) ExecInitCteScan((CteScan *) node, |
| 271 |
estate, eflags); |
| 272 |
break; |
| 273 |
|
| 274 |
case T_NamedTuplestoreScan: |
| 275 |
result = (PlanState *) ExecInitNamedTuplestoreScan((NamedTuplestoreScan *) node, |
| 276 |
estate, eflags); |
| 277 |
break; |
| 278 |
|
| 279 |
case T_WorkTableScan: |
| 280 |
result = (PlanState *) ExecInitWorkTableScan((WorkTableScan *) node, |
| 281 |
estate, eflags); |
| 282 |
break; |
| 283 |
|
| 284 |
case T_ForeignScan: |
| 285 |
result = (PlanState *) ExecInitForeignScan((ForeignScan *) node, |
| 286 |
estate, eflags); |
| 287 |
break; |
| 288 |
|
| 289 |
case T_CustomScan: |
| 290 |
result = (PlanState *) ExecInitCustomScan((CustomScan *) node, |
| 291 |
estate, eflags); |
| 292 |
break; |
| 293 |
|
| 294 |
/* |
| 295 |
* join nodes |
| 296 |
*/ |
| 297 |
case T_NestLoop: |
| 298 |
result = (PlanState *) ExecInitNestLoop((NestLoop *) node, |
| 299 |
estate, eflags); |
| 300 |
break; |
| 301 |
|
| 302 |
case T_MergeJoin: |
| 303 |
result = (PlanState *) ExecInitMergeJoin((MergeJoin *) node, |
| 304 |
estate, eflags); |
| 305 |
break; |
| 306 |
|
| 307 |
case T_HashJoin: |
| 308 |
result = (PlanState *) ExecInitHashJoin((HashJoin *) node, |
| 309 |
estate, eflags); |
| 310 |
break; |
| 311 |
|
| 312 |
/* |
| 313 |
* materialization nodes |
| 314 |
*/ |
| 315 |
case T_Material: |
| 316 |
result = (PlanState *) ExecInitMaterial((Material *) node, |
| 317 |
estate, eflags); |
| 318 |
break; |
| 319 |
|
| 320 |
case T_Sort: |
| 321 |
result = (PlanState *) ExecInitSort((Sort *) node, |
| 322 |
estate, eflags); |
| 323 |
break; |
| 324 |
|
| 325 |
case T_IncrementalSort: |
| 326 |
result = (PlanState *) ExecInitIncrementalSort((IncrementalSort *) node, |
| 327 |
estate, eflags); |
| 328 |
break; |
| 329 |
|
| 330 |
case T_Memoize: |
| 331 |
result = (PlanState *) ExecInitMemoize((Memoize *) node, estate, |
| 332 |
eflags); |
| 333 |
break; |
| 334 |
|
| 335 |
case T_Group: |
| 336 |
result = (PlanState *) ExecInitGroup((Group *) node, |
| 337 |
estate, eflags); |
| 338 |
break; |
| 339 |
|
| 340 |
case T_Agg: |
| 341 |
result = (PlanState *) ExecInitAgg((Agg *) node, |
| 342 |
estate, eflags); |
| 343 |
break; |
| 344 |
|
| 345 |
case T_WindowAgg: |
| 346 |
result = (PlanState *) ExecInitWindowAgg((WindowAgg *) node, |
| 347 |
estate, eflags); |
| 348 |
break; |
| 349 |
|
| 350 |
case T_Unique: |
| 351 |
result = (PlanState *) ExecInitUnique((Unique *) node, |
| 352 |
estate, eflags); |
| 353 |
break; |
| 354 |
|
| 355 |
case T_Gather: |
| 356 |
result = (PlanState *) ExecInitGather((Gather *) node, |
| 357 |
estate, eflags); |
| 358 |
break; |
| 359 |
|
| 360 |
case T_GatherMerge: |
| 361 |
result = (PlanState *) ExecInitGatherMerge((GatherMerge *) node, |
| 362 |
estate, eflags); |
| 363 |
break; |
| 364 |
|
| 365 |
case T_Hash: |
| 366 |
result = (PlanState *) ExecInitHash((Hash *) node, |
| 367 |
estate, eflags); |
| 368 |
break; |
| 369 |
|
| 370 |
case T_SetOp: |
| 371 |
result = (PlanState *) ExecInitSetOp((SetOp *) node, |
| 372 |
estate, eflags); |
| 373 |
break; |
| 374 |
|
| 375 |
case T_LockRows: |
| 376 |
result = (PlanState *) ExecInitLockRows((LockRows *) node, |
| 377 |
estate, eflags); |
| 378 |
break; |
| 379 |
|
| 380 |
case T_Limit: |
| 381 |
result = (PlanState *) ExecInitLimit((Limit *) node, |
| 382 |
estate, eflags); |
| 383 |
break; |
| 384 |
|
| 385 |
default: |
| 386 |
elog(ERROR, "unrecognized node type: %d", (int) nodeTag(node)); |
| 387 |
result = NULL; /* keep compiler quiet */ |
| 388 |
break; |
| 389 |
} |
| 390 |
|
| 391 |
ExecSetExecProcNode(result, result->ExecProcNode); |
| 392 |
|
| 393 |
/* |
| 394 |
* Initialize any initPlans present in this node. The planner put them in |
| 395 |
* a separate list for us. |
| 396 |
* |
| 397 |
* The defining characteristic of initplans is that they don't have |
| 398 |
* arguments, so we don't need to evaluate them (in contrast to |
| 399 |
* ExecInitSubPlanExpr()). |
| 400 |
*/ |
| 401 |
subps = NIL; |
| 402 |
foreach(l, node->initPlan) |
| 403 |
{ |
| 404 |
SubPlan *subplan = (SubPlan *) lfirst(l); |
| 405 |
SubPlanState *sstate; |
| 406 |
|
| 407 |
Assert(IsA(subplan, SubPlan)); |
| 408 |
Assert(subplan->args == NIL); |
| 409 |
sstate = ExecInitSubPlan(subplan, result); |
| 410 |
subps = lappend(subps, sstate); |
| 411 |
} |
| 412 |
result->initPlan = subps; |
| 413 |
|
| 414 |
/* Set up instrumentation for this node if requested */ |
| 415 |
if (estate->es_instrument) |
| 416 |
result->instrument = InstrAlloc(1, estate->es_instrument, |
| 417 |
result->async_capable); |
| 418 |
|
| 419 |
return result; |
| 420 |
} |
| 421 |
|
| 422 |
|
| 423 |
/* |
| 424 |
* If a node wants to change its ExecProcNode function after ExecInitNode() |
| 425 |
* has finished, it should do so with this function. That way any wrapper |
| 426 |
* functions can be reinstalled, without the node having to know how that |
| 427 |
* works. |
| 428 |
*/ |
| 429 |
void |
| 430 |
ExecSetExecProcNode(PlanState *node, ExecProcNodeMtd function) |
| 431 |
{ |
| 432 |
/* |
| 433 |
* Add a wrapper around the ExecProcNode callback that checks stack depth |
| 434 |
* during the first execution and maybe adds an instrumentation wrapper. |
| 435 |
* When the callback is changed after execution has already begun that |
| 436 |
* means we'll superfluously execute ExecProcNodeFirst, but that seems ok. |
| 437 |
*/ |
| 438 |
node->ExecProcNodeReal = function; |
| 439 |
node->ExecProcNode = ExecProcNodeFirst; |
| 440 |
} |
| 441 |
|
| 442 |
|
| 443 |
/* |
| 444 |
* ExecProcNode wrapper that performs some one-time checks, before calling |
| 445 |
* the relevant node method (possibly via an instrumentation wrapper). |
| 446 |
*/ |
| 447 |
static TupleTableSlot * |
| 448 |
ExecProcNodeFirst(PlanState *node) |
| 449 |
{ |
| 450 |
/* |
| 451 |
* Perform stack depth check during the first execution of the node. We |
| 452 |
* only do so the first time round because it turns out to not be cheap on |
| 453 |
* some common architectures (eg. x86). This relies on the assumption |
| 454 |
* that ExecProcNode calls for a given plan node will always be made at |
| 455 |
* roughly the same stack depth. |
| 456 |
*/ |
| 457 |
check_stack_depth(); |
| 458 |
|
| 459 |
/* |
| 460 |
* If instrumentation is required, change the wrapper to one that just |
| 461 |
* does instrumentation. Otherwise we can dispense with all wrappers and |
| 462 |
* have ExecProcNode() directly call the relevant function from now on. |
| 463 |
*/ |
| 464 |
if (node->instrument) |
| 465 |
node->ExecProcNode = ExecProcNodeInstr; |
| 466 |
else |
| 467 |
node->ExecProcNode = node->ExecProcNodeReal; |
| 468 |
|
| 469 |
return node->ExecProcNode(node); |
| 470 |
} |
| 471 |
|
| 472 |
|
| 473 |
/* |
| 474 |
* ExecProcNode wrapper that performs instrumentation calls. By keeping |
| 475 |
* this a separate function, we avoid overhead in the normal case where |
| 476 |
* no instrumentation is wanted. |
| 477 |
*/ |
| 478 |
static TupleTableSlot * |
| 479 |
ExecProcNodeInstr(PlanState *node) |
| 480 |
{ |
| 481 |
TupleTableSlot *result; |
| 482 |
|
| 483 |
InstrStartNode(node->instrument); |
| 484 |
|
| 485 |
result = node->ExecProcNodeReal(node); |
| 486 |
|
| 487 |
InstrStopNode(node->instrument, TupIsNull(result) ? 0.0 : 1.0); |
| 488 |
|
| 489 |
return result; |
| 490 |
} |
| 491 |
|
| 492 |
|
| 493 |
/* ---------------------------------------------------------------- |
| 494 |
* MultiExecProcNode |
| 495 |
* |
| 496 |
* Execute a node that doesn't return individual tuples |
| 497 |
* (it might return a hashtable, bitmap, etc). Caller should |
| 498 |
* check it got back the expected kind of Node. |
| 499 |
* |
| 500 |
* This has essentially the same responsibilities as ExecProcNode, |
| 501 |
* but it does not do InstrStartNode/InstrStopNode (mainly because |
| 502 |
* it can't tell how many returned tuples to count). Each per-node |
| 503 |
* function must provide its own instrumentation support. |
| 504 |
* ---------------------------------------------------------------- |
| 505 |
*/ |
| 506 |
Node * |
| 507 |
MultiExecProcNode(PlanState *node) |
| 508 |
{ |
| 509 |
Node *result; |
| 510 |
|
| 511 |
check_stack_depth(); |
| 512 |
|
| 513 |
CHECK_FOR_INTERRUPTS(); |
| 514 |
|
| 515 |
if (node->chgParam != NULL) /* something changed */ |
| 516 |
ExecReScan(node); /* let ReScan handle this */ |
| 517 |
|
| 518 |
switch (nodeTag(node)) |
| 519 |
{ |
| 520 |
/* |
| 521 |
* Only node types that actually support multiexec will be listed |
| 522 |
*/ |
| 523 |
|
| 524 |
case T_HashState: |
| 525 |
result = MultiExecHash((HashState *) node); |
| 526 |
break; |
| 527 |
|
| 528 |
case T_BitmapIndexScanState: |
| 529 |
result = MultiExecBitmapIndexScan((BitmapIndexScanState *) node); |
| 530 |
break; |
| 531 |
|
| 532 |
case T_BitmapAndState: |
| 533 |
result = MultiExecBitmapAnd((BitmapAndState *) node); |
| 534 |
break; |
| 535 |
|
| 536 |
case T_BitmapOrState: |
| 537 |
result = MultiExecBitmapOr((BitmapOrState *) node); |
| 538 |
break; |
| 539 |
|
| 540 |
default: |
| 541 |
elog(ERROR, "unrecognized node type: %d", (int) nodeTag(node)); |
| 542 |
result = NULL; |
| 543 |
break; |
| 544 |
} |
| 545 |
|
| 546 |
return result; |
| 547 |
} |
| 548 |
|
| 549 |
|
| 550 |
/* ---------------------------------------------------------------- |
| 551 |
* ExecEndNode |
| 552 |
* |
| 553 |
* Recursively cleans up all the nodes in the plan rooted |
| 554 |
* at 'node'. |
| 555 |
* |
| 556 |
* After this operation, the query plan will not be able to be |
| 557 |
* processed any further. This should be called only after |
| 558 |
* the query plan has been fully executed. |
| 559 |
* ---------------------------------------------------------------- |
| 560 |
*/ |
| 561 |
void |
| 562 |
ExecEndNode(PlanState *node) |
| 563 |
{ |
| 564 |
/* |
| 565 |
* do nothing when we get to the end of a leaf on tree. |
| 566 |
*/ |
| 567 |
if (node == NULL) |
| 568 |
return; |
| 569 |
|
| 570 |
/* |
| 571 |
* Make sure there's enough stack available. Need to check here, in |
| 572 |
* addition to ExecProcNode() (via ExecProcNodeFirst()), because it's not |
| 573 |
* guaranteed that ExecProcNode() is reached for all nodes. |
| 574 |
*/ |
| 575 |
check_stack_depth(); |
| 576 |
|
| 577 |
if (node->chgParam != NULL) |
| 578 |
{ |
| 579 |
bms_free(node->chgParam); |
| 580 |
node->chgParam = NULL; |
| 581 |
} |
| 582 |
|
| 583 |
switch (nodeTag(node)) |
| 584 |
{ |
| 585 |
/* |
| 586 |
* control nodes |
| 587 |
*/ |
| 588 |
case T_ResultState: |
| 589 |
ExecEndResult((ResultState *) node); |
| 590 |
break; |
| 591 |
|
| 592 |
case T_ProjectSetState: |
| 593 |
ExecEndProjectSet((ProjectSetState *) node); |
| 594 |
break; |
| 595 |
|
| 596 |
case T_ModifyTableState: |
| 597 |
ExecEndModifyTable((ModifyTableState *) node); |
| 598 |
break; |
| 599 |
|
| 600 |
case T_AppendState: |
| 601 |
ExecEndAppend((AppendState *) node); |
| 602 |
break; |
| 603 |
|
| 604 |
case T_MergeAppendState: |
| 605 |
ExecEndMergeAppend((MergeAppendState *) node); |
| 606 |
break; |
| 607 |
|
| 608 |
case T_RecursiveUnionState: |
| 609 |
ExecEndRecursiveUnion((RecursiveUnionState *) node); |
| 610 |
break; |
| 611 |
|
| 612 |
case T_BitmapAndState: |
| 613 |
ExecEndBitmapAnd((BitmapAndState *) node); |
| 614 |
break; |
| 615 |
|
| 616 |
case T_BitmapOrState: |
| 617 |
ExecEndBitmapOr((BitmapOrState *) node); |
| 618 |
break; |
| 619 |
|
| 620 |
/* |
| 621 |
* scan nodes |
| 622 |
*/ |
| 623 |
case T_SeqScanState: |
| 624 |
ExecEndSeqScan((SeqScanState *) node); |
| 625 |
break; |
| 626 |
|
| 627 |
case T_SampleScanState: |
| 628 |
ExecEndSampleScan((SampleScanState *) node); |
| 629 |
break; |
| 630 |
|
| 631 |
case T_GatherState: |
| 632 |
ExecEndGather((GatherState *) node); |
| 633 |
break; |
| 634 |
|
| 635 |
case T_GatherMergeState: |
| 636 |
ExecEndGatherMerge((GatherMergeState *) node); |
| 637 |
break; |
| 638 |
|
| 639 |
case T_IndexScanState: |
| 640 |
ExecEndIndexScan((IndexScanState *) node); |
| 641 |
break; |
| 642 |
|
| 643 |
case T_IndexOnlyScanState: |
| 644 |
ExecEndIndexOnlyScan((IndexOnlyScanState *) node); |
| 645 |
break; |
| 646 |
|
| 647 |
case T_BitmapIndexScanState: |
| 648 |
ExecEndBitmapIndexScan((BitmapIndexScanState *) node); |
| 649 |
break; |
| 650 |
|
| 651 |
case T_BitmapHeapScanState: |
| 652 |
ExecEndBitmapHeapScan((BitmapHeapScanState *) node); |
| 653 |
break; |
| 654 |
|
| 655 |
case T_TidScanState: |
| 656 |
ExecEndTidScan((TidScanState *) node); |
| 657 |
break; |
| 658 |
|
| 659 |
case T_TidRangeScanState: |
| 660 |
ExecEndTidRangeScan((TidRangeScanState *) node); |
| 661 |
break; |
| 662 |
|
| 663 |
case T_SubqueryScanState: |
| 664 |
ExecEndSubqueryScan((SubqueryScanState *) node); |
| 665 |
break; |
| 666 |
|
| 667 |
case T_FunctionScanState: |
| 668 |
ExecEndFunctionScan((FunctionScanState *) node); |
| 669 |
break; |
| 670 |
|
| 671 |
case T_TableFuncScanState: |
| 672 |
ExecEndTableFuncScan((TableFuncScanState *) node); |
| 673 |
break; |
| 674 |
|
| 675 |
case T_CteScanState: |
| 676 |
ExecEndCteScan((CteScanState *) node); |
| 677 |
break; |
| 678 |
|
| 679 |
case T_ForeignScanState: |
| 680 |
ExecEndForeignScan((ForeignScanState *) node); |
| 681 |
break; |
| 682 |
|
| 683 |
case T_CustomScanState: |
| 684 |
ExecEndCustomScan((CustomScanState *) node); |
| 685 |
break; |
| 686 |
|
| 687 |
/* |
| 688 |
* join nodes |
| 689 |
*/ |
| 690 |
case T_NestLoopState: |
| 691 |
ExecEndNestLoop((NestLoopState *) node); |
| 692 |
break; |
| 693 |
|
| 694 |
case T_MergeJoinState: |
| 695 |
ExecEndMergeJoin((MergeJoinState *) node); |
| 696 |
break; |
| 697 |
|
| 698 |
case T_HashJoinState: |
| 699 |
ExecEndHashJoin((HashJoinState *) node); |
| 700 |
break; |
| 701 |
|
| 702 |
/* |
| 703 |
* materialization nodes |
| 704 |
*/ |
| 705 |
case T_MaterialState: |
| 706 |
ExecEndMaterial((MaterialState *) node); |
| 707 |
break; |
| 708 |
|
| 709 |
case T_SortState: |
| 710 |
ExecEndSort((SortState *) node); |
| 711 |
break; |
| 712 |
|
| 713 |
case T_IncrementalSortState: |
| 714 |
ExecEndIncrementalSort((IncrementalSortState *) node); |
| 715 |
break; |
| 716 |
|
| 717 |
case T_MemoizeState: |
| 718 |
ExecEndMemoize((MemoizeState *) node); |
| 719 |
break; |
| 720 |
|
| 721 |
case T_GroupState: |
| 722 |
ExecEndGroup((GroupState *) node); |
| 723 |
break; |
| 724 |
|
| 725 |
case T_AggState: |
| 726 |
ExecEndAgg((AggState *) node); |
| 727 |
break; |
| 728 |
|
| 729 |
case T_WindowAggState: |
| 730 |
ExecEndWindowAgg((WindowAggState *) node); |
| 731 |
break; |
| 732 |
|
| 733 |
case T_UniqueState: |
| 734 |
ExecEndUnique((UniqueState *) node); |
| 735 |
break; |
| 736 |
|
| 737 |
case T_HashState: |
| 738 |
ExecEndHash((HashState *) node); |
| 739 |
break; |
| 740 |
|
| 741 |
case T_SetOpState: |
| 742 |
ExecEndSetOp((SetOpState *) node); |
| 743 |
break; |
| 744 |
|
| 745 |
case T_LockRowsState: |
| 746 |
ExecEndLockRows((LockRowsState *) node); |
| 747 |
break; |
| 748 |
|
| 749 |
case T_LimitState: |
| 750 |
ExecEndLimit((LimitState *) node); |
| 751 |
break; |
| 752 |
|
| 753 |
/* No clean up actions for these nodes. */ |
| 754 |
case T_ValuesScanState: |
| 755 |
case T_NamedTuplestoreScanState: |
| 756 |
case T_WorkTableScanState: |
| 757 |
break; |
| 758 |
|
| 759 |
default: |
| 760 |
elog(ERROR, "unrecognized node type: %d", (int) nodeTag(node)); |
| 761 |
break; |
| 762 |
} |
| 763 |
} |
| 764 |
|
| 765 |
/* |
| 766 |
* ExecShutdownNode |
| 767 |
* |
| 768 |
* Give execution nodes a chance to stop asynchronous resource consumption |
| 769 |
* and release any resources still held. |
| 770 |
*/ |
| 771 |
void |
| 772 |
ExecShutdownNode(PlanState *node) |
| 773 |
{ |
| 774 |
(void) ExecShutdownNode_walker(node, NULL); |
| 775 |
} |
| 776 |
|
| 777 |
static bool |
| 778 |
ExecShutdownNode_walker(PlanState *node, void *context) |
| 779 |
{ |
| 780 |
if (node == NULL) |
| 781 |
return false; |
| 782 |
|
| 783 |
check_stack_depth(); |
| 784 |
|
| 785 |
/* |
| 786 |
* Treat the node as running while we shut it down, but only if it's run |
| 787 |
* at least once already. We don't expect much CPU consumption during |
| 788 |
* node shutdown, but in the case of Gather or Gather Merge, we may shut |
| 789 |
* down workers at this stage. If so, their buffer usage will get |
| 790 |
* propagated into pgBufferUsage at this point, and we want to make sure |
| 791 |
* that it gets associated with the Gather node. We skip this if the node |
| 792 |
* has never been executed, so as to avoid incorrectly making it appear |
| 793 |
* that it has. |
| 794 |
*/ |
| 795 |
if (node->instrument && node->instrument->running) |
| 796 |
InstrStartNode(node->instrument); |
| 797 |
|
| 798 |
planstate_tree_walker(node, ExecShutdownNode_walker, context); |
| 799 |
|
| 800 |
switch (nodeTag(node)) |
| 801 |
{ |
| 802 |
case T_GatherState: |
| 803 |
ExecShutdownGather((GatherState *) node); |
| 804 |
break; |
| 805 |
case T_ForeignScanState: |
| 806 |
ExecShutdownForeignScan((ForeignScanState *) node); |
| 807 |
break; |
| 808 |
case T_CustomScanState: |
| 809 |
ExecShutdownCustomScan((CustomScanState *) node); |
| 810 |
break; |
| 811 |
case T_GatherMergeState: |
| 812 |
ExecShutdownGatherMerge((GatherMergeState *) node); |
| 813 |
break; |
| 814 |
case T_HashState: |
| 815 |
ExecShutdownHash((HashState *) node); |
| 816 |
break; |
| 817 |
case T_HashJoinState: |
| 818 |
ExecShutdownHashJoin((HashJoinState *) node); |
| 819 |
break; |
| 820 |
default: |
| 821 |
break; |
| 822 |
} |
| 823 |
|
| 824 |
/* Stop the node if we started it above, reporting 0 tuples. */ |
| 825 |
if (node->instrument && node->instrument->running) |
| 826 |
InstrStopNode(node->instrument, 0); |
| 827 |
|
| 828 |
return false; |
| 829 |
} |
| 830 |
|
| 831 |
/* |
| 832 |
* ExecSetTupleBound |
| 833 |
* |
| 834 |
* Set a tuple bound for a planstate node. This lets child plan nodes |
| 835 |
* optimize based on the knowledge that the maximum number of tuples that |
| 836 |
* their parent will demand is limited. The tuple bound for a node may |
| 837 |
* only be changed between scans (i.e., after node initialization or just |
| 838 |
* before an ExecReScan call). |
| 839 |
* |
| 840 |
* Any negative tuples_needed value means "no limit", which should be the |
| 841 |
* default assumption when this is not called at all for a particular node. |
| 842 |
* |
| 843 |
* Note: if this is called repeatedly on a plan tree, the exact same set |
| 844 |
* of nodes must be updated with the new limit each time; be careful that |
| 845 |
* only unchanging conditions are tested here. |
| 846 |
*/ |
| 847 |
void |
| 848 |
ExecSetTupleBound(int64 tuples_needed, PlanState *child_node) |
| 849 |
{ |
| 850 |
/* |
| 851 |
* Since this function recurses, in principle we should check stack depth |
| 852 |
* here. In practice, it's probably pointless since the earlier node |
| 853 |
* initialization tree traversal would surely have consumed more stack. |
| 854 |
*/ |
| 855 |
|
| 856 |
if (IsA(child_node, SortState)) |
| 857 |
{ |
| 858 |
/* |
| 859 |
* If it is a Sort node, notify it that it can use bounded sort. |
| 860 |
* |
| 861 |
* Note: it is the responsibility of nodeSort.c to react properly to |
| 862 |
* changes of these parameters. If we ever redesign this, it'd be a |
| 863 |
* good idea to integrate this signaling with the parameter-change |
| 864 |
* mechanism. |
| 865 |
*/ |
| 866 |
SortState *sortState = (SortState *) child_node; |
| 867 |
|
| 868 |
if (tuples_needed < 0) |
| 869 |
{ |
| 870 |
/* make sure flag gets reset if needed upon rescan */ |
| 871 |
sortState->bounded = false; |
| 872 |
} |
| 873 |
else |
| 874 |
{ |
| 875 |
sortState->bounded = true; |
| 876 |
sortState->bound = tuples_needed; |
| 877 |
} |
| 878 |
} |
| 879 |
else if (IsA(child_node, IncrementalSortState)) |
| 880 |
{ |
| 881 |
/* |
| 882 |
* If it is an IncrementalSort node, notify it that it can use bounded |
| 883 |
* sort. |
| 884 |
* |
| 885 |
* Note: it is the responsibility of nodeIncrementalSort.c to react |
| 886 |
* properly to changes of these parameters. If we ever redesign this, |
| 887 |
* it'd be a good idea to integrate this signaling with the |
| 888 |
* parameter-change mechanism. |
| 889 |
*/ |
| 890 |
IncrementalSortState *sortState = (IncrementalSortState *) child_node; |
| 891 |
|
| 892 |
if (tuples_needed < 0) |
| 893 |
{ |
| 894 |
/* make sure flag gets reset if needed upon rescan */ |
| 895 |
sortState->bounded = false; |
| 896 |
} |
| 897 |
else |
| 898 |
{ |
| 899 |
sortState->bounded = true; |
| 900 |
sortState->bound = tuples_needed; |
| 901 |
} |
| 902 |
} |
| 903 |
else if (IsA(child_node, AppendState)) |
| 904 |
{ |
| 905 |
/* |
| 906 |
* If it is an Append, we can apply the bound to any nodes that are |
| 907 |
* children of the Append, since the Append surely need read no more |
| 908 |
* than that many tuples from any one input. |
| 909 |
*/ |
| 910 |
AppendState *aState = (AppendState *) child_node; |
| 911 |
int i; |
| 912 |
|
| 913 |
for (i = 0; i < aState->as_nplans; i++) |
| 914 |
ExecSetTupleBound(tuples_needed, aState->appendplans[i]); |
| 915 |
} |
| 916 |
else if (IsA(child_node, MergeAppendState)) |
| 917 |
{ |
| 918 |
/* |
| 919 |
* If it is a MergeAppend, we can apply the bound to any nodes that |
| 920 |
* are children of the MergeAppend, since the MergeAppend surely need |
| 921 |
* read no more than that many tuples from any one input. |
| 922 |
*/ |
| 923 |
MergeAppendState *maState = (MergeAppendState *) child_node; |
| 924 |
int i; |
| 925 |
|
| 926 |
for (i = 0; i < maState->ms_nplans; i++) |
| 927 |
ExecSetTupleBound(tuples_needed, maState->mergeplans[i]); |
| 928 |
} |
| 929 |
else if (IsA(child_node, ResultState)) |
| 930 |
{ |
| 931 |
/* |
| 932 |
* Similarly, for a projecting Result, we can apply the bound to its |
| 933 |
* child node. |
| 934 |
* |
| 935 |
* If Result supported qual checking, we'd have to punt on seeing a |
| 936 |
* qual. Note that having a resconstantqual is not a showstopper: if |
| 937 |
* that condition succeeds it affects nothing, while if it fails, no |
| 938 |
* rows will be demanded from the Result child anyway. |
| 939 |
*/ |
| 940 |
if (outerPlanState(child_node)) |
| 941 |
ExecSetTupleBound(tuples_needed, outerPlanState(child_node)); |
| 942 |
} |
| 943 |
else if (IsA(child_node, SubqueryScanState)) |
| 944 |
{ |
| 945 |
/* |
| 946 |
* We can also descend through SubqueryScan, but only if it has no |
| 947 |
* qual (otherwise it might discard rows). |
| 948 |
*/ |
| 949 |
SubqueryScanState *subqueryState = (SubqueryScanState *) child_node; |
| 950 |
|
| 951 |
if (subqueryState->ss.ps.qual == NULL) |
| 952 |
ExecSetTupleBound(tuples_needed, subqueryState->subplan); |
| 953 |
} |
| 954 |
else if (IsA(child_node, GatherState)) |
| 955 |
{ |
| 956 |
/* |
| 957 |
* A Gather node can propagate the bound to its workers. As with |
| 958 |
* MergeAppend, no one worker could possibly need to return more |
| 959 |
* tuples than the Gather itself needs to. |
| 960 |
* |
| 961 |
* Note: As with Sort, the Gather node is responsible for reacting |
| 962 |
* properly to changes to this parameter. |
| 963 |
*/ |
| 964 |
GatherState *gstate = (GatherState *) child_node; |
| 965 |
|
| 966 |
gstate->tuples_needed = tuples_needed; |
| 967 |
|
| 968 |
/* Also pass down the bound to our own copy of the child plan */ |
| 969 |
ExecSetTupleBound(tuples_needed, outerPlanState(child_node)); |
| 970 |
} |
| 971 |
else if (IsA(child_node, GatherMergeState)) |
| 972 |
{ |
| 973 |
/* Same comments as for Gather */ |
| 974 |
GatherMergeState *gstate = (GatherMergeState *) child_node; |
| 975 |
|
| 976 |
gstate->tuples_needed = tuples_needed; |
| 977 |
|
| 978 |
ExecSetTupleBound(tuples_needed, outerPlanState(child_node)); |
| 979 |
} |
| 980 |
|
| 981 |
/* |
| 982 |
* In principle we could descend through any plan node type that is |
| 983 |
* certain not to discard or combine input rows; but on seeing a node that |
| 984 |
* can do that, we can't propagate the bound any further. For the moment |
| 985 |
* it's unclear that any other cases are worth checking here. |
| 986 |
*/ |
| 987 |
} |