Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * nodeHashjoin.c
4 : : * Routines to handle hash join nodes
5 : : *
6 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/executor/nodeHashjoin.c
12 : : *
13 : : * HASH JOIN
14 : : *
15 : : * This is based on the "hybrid hash join" algorithm described shortly in the
16 : : * following page
17 : : *
18 : : * https://en.wikipedia.org/wiki/Hash_join#Hybrid_hash_join
19 : : *
20 : : * and in detail in the referenced paper:
21 : : *
22 : : * "An Adaptive Hash Join Algorithm for Multiuser Environments"
23 : : * Hansjörg Zeller; Jim Gray (1990). Proceedings of the 16th VLDB conference.
24 : : * Brisbane: 186–197.
25 : : *
26 : : * If the inner side tuples of a hash join do not fit in memory, the hash join
27 : : * can be executed in multiple batches.
28 : : *
29 : : * If the statistics on the inner side relation are accurate, planner chooses a
30 : : * multi-batch strategy and estimates the number of batches.
31 : : *
32 : : * The query executor measures the real size of the hashtable and increases the
33 : : * number of batches if the hashtable grows too large.
34 : : *
35 : : * The number of batches is always a power of two, so an increase in the number
36 : : * of batches doubles it.
37 : : *
38 : : * Serial hash join measures batch size lazily -- waiting until it is loading a
39 : : * batch to determine if it will fit in memory. While inserting tuples into the
40 : : * hashtable, serial hash join will, if that tuple were to exceed work_mem,
41 : : * dump out the hashtable and reassign them either to other batch files or the
42 : : * current batch resident in the hashtable.
43 : : *
44 : : * Parallel hash join, on the other hand, completes all changes to the number
45 : : * of batches during the build phase. If it increases the number of batches, it
46 : : * dumps out all the tuples from all batches and reassigns them to entirely new
47 : : * batch files. Then it checks every batch to ensure it will fit in the space
48 : : * budget for the query.
49 : : *
50 : : * In both parallel and serial hash join, the executor currently makes a best
51 : : * effort. If a particular batch will not fit in memory, it tries doubling the
52 : : * number of batches. If after a batch increase, there is a batch which
53 : : * retained all or none of its tuples, the executor disables growth in the
54 : : * number of batches globally. After growth is disabled, all batches that would
55 : : * have previously triggered an increase in the number of batches instead
56 : : * exceed the space allowed.
57 : : *
58 : : * PARALLELISM
59 : : *
60 : : * Hash joins can participate in parallel query execution in several ways. A
61 : : * parallel-oblivious hash join is one where the node is unaware that it is
62 : : * part of a parallel plan. In this case, a copy of the inner plan is used to
63 : : * build a copy of the hash table in every backend, and the outer plan could
64 : : * either be built from a partial or complete path, so that the results of the
65 : : * hash join are correspondingly either partial or complete. A parallel-aware
66 : : * hash join is one that behaves differently, coordinating work between
67 : : * backends, and appears as Parallel Hash Join in EXPLAIN output. A Parallel
68 : : * Hash Join always appears with a Parallel Hash node.
69 : : *
70 : : * Parallel-aware hash joins use the same per-backend state machine to track
71 : : * progress through the hash join algorithm as parallel-oblivious hash joins.
72 : : * In a parallel-aware hash join, there is also a shared state machine that
73 : : * co-operating backends use to synchronize their local state machines and
74 : : * program counters. The shared state machine is managed with a Barrier IPC
75 : : * primitive. When all attached participants arrive at a barrier, the phase
76 : : * advances and all waiting participants are released.
77 : : *
78 : : * When a participant begins working on a parallel hash join, it must first
79 : : * figure out how much progress has already been made, because participants
80 : : * don't wait for each other to begin. For this reason there are switch
81 : : * statements at key points in the code where we have to synchronize our local
82 : : * state machine with the phase, and then jump to the correct part of the
83 : : * algorithm so that we can get started.
84 : : *
85 : : * One barrier called build_barrier is used to coordinate the hashing phases.
86 : : * The phase is represented by an integer which begins at zero and increments
87 : : * one by one, but in the code it is referred to by symbolic names as follows.
88 : : * An asterisk indicates a phase that is performed by a single arbitrarily
89 : : * chosen process.
90 : : *
91 : : * PHJ_BUILD_ELECT -- initial state
92 : : * PHJ_BUILD_ALLOCATE* -- one sets up the batches and table 0
93 : : * PHJ_BUILD_HASH_INNER -- all hash the inner rel
94 : : * PHJ_BUILD_HASH_OUTER -- (multi-batch only) all hash the outer
95 : : * PHJ_BUILD_RUN -- building done, probing can begin
96 : : * PHJ_BUILD_FREE* -- all work complete, one frees batches
97 : : *
98 : : * While in the phase PHJ_BUILD_HASH_INNER a separate pair of barriers may
99 : : * be used repeatedly as required to coordinate expansions in the number of
100 : : * batches or buckets. Their phases are as follows:
101 : : *
102 : : * PHJ_GROW_BATCHES_ELECT -- initial state
103 : : * PHJ_GROW_BATCHES_REALLOCATE* -- one allocates new batches
104 : : * PHJ_GROW_BATCHES_REPARTITION -- all repartition
105 : : * PHJ_GROW_BATCHES_DECIDE* -- one detects skew and cleans up
106 : : * PHJ_GROW_BATCHES_FINISH -- finished one growth cycle
107 : : *
108 : : * PHJ_GROW_BUCKETS_ELECT -- initial state
109 : : * PHJ_GROW_BUCKETS_REALLOCATE* -- one allocates new buckets
110 : : * PHJ_GROW_BUCKETS_REINSERT -- all insert tuples
111 : : *
112 : : * If the planner got the number of batches and buckets right, those won't be
113 : : * necessary, but on the other hand we might finish up needing to expand the
114 : : * buckets or batches multiple times while hashing the inner relation to stay
115 : : * within our memory budget and load factor target. For that reason it's a
116 : : * separate pair of barriers using circular phases.
117 : : *
118 : : * The PHJ_BUILD_HASH_OUTER phase is required only for multi-batch joins,
119 : : * because we need to divide the outer relation into batches up front in order
120 : : * to be able to process batches entirely independently. In contrast, the
121 : : * parallel-oblivious algorithm simply throws tuples 'forward' to 'later'
122 : : * batches whenever it encounters them while scanning and probing, which it
123 : : * can do because it processes batches in serial order.
124 : : *
125 : : * Once PHJ_BUILD_RUN is reached, backends then split up and process
126 : : * different batches, or gang up and work together on probing batches if there
127 : : * aren't enough to go around. For each batch there is a separate barrier
128 : : * with the following phases:
129 : : *
130 : : * PHJ_BATCH_ELECT -- initial state
131 : : * PHJ_BATCH_ALLOCATE* -- one allocates buckets
132 : : * PHJ_BATCH_LOAD -- all load the hash table from disk
133 : : * PHJ_BATCH_PROBE -- all probe
134 : : * PHJ_BATCH_SCAN* -- one does right/right-anti/full unmatched scan
135 : : * PHJ_BATCH_FREE* -- one frees memory
136 : : *
137 : : * Batch 0 is a special case, because it starts out in phase
138 : : * PHJ_BATCH_PROBE; populating batch 0's hash table is done during
139 : : * PHJ_BUILD_HASH_INNER so we can skip loading.
140 : : *
141 : : * Initially we try to plan for a single-batch hash join using the combined
142 : : * hash_mem of all participants to create a large shared hash table. If that
143 : : * turns out either at planning or execution time to be impossible then we
144 : : * fall back to regular hash_mem sized hash tables.
145 : : *
146 : : * To avoid deadlocks, we never wait for any barrier unless it is known that
147 : : * all other backends attached to it are actively executing the node or have
148 : : * finished. Practically, that means that we never emit a tuple while attached
149 : : * to a barrier, unless the barrier has reached a phase that means that no
150 : : * process will wait on it again. We emit tuples while attached to the build
151 : : * barrier in phase PHJ_BUILD_RUN, and to a per-batch barrier in phase
152 : : * PHJ_BATCH_PROBE. These are advanced to PHJ_BUILD_FREE and PHJ_BATCH_SCAN
153 : : * respectively without waiting, using BarrierArriveAndDetach() and
154 : : * BarrierArriveAndDetachExceptLast() respectively. The last to detach
155 : : * receives a different return value so that it knows that it's safe to
156 : : * clean up. Any straggler process that attaches after that phase is reached
157 : : * will see that it's too late to participate or access the relevant shared
158 : : * memory objects.
159 : : *
160 : : *-------------------------------------------------------------------------
161 : : */
162 : :
163 : : #include "postgres.h"
164 : :
165 : : #include "access/htup_details.h"
166 : : #include "access/parallel.h"
167 : : #include "executor/executor.h"
168 : : #include "executor/hashjoin.h"
169 : : #include "executor/instrument.h"
170 : : #include "executor/nodeHash.h"
171 : : #include "executor/nodeHashjoin.h"
172 : : #include "miscadmin.h"
173 : : #include "utils/lsyscache.h"
174 : : #include "utils/sharedtuplestore.h"
175 : : #include "utils/tuplestore.h"
176 : : #include "utils/wait_event.h"
177 : :
178 : :
179 : : /*
180 : : * States of the ExecHashJoin state machine
181 : : */
182 : : #define HJ_BUILD_HASHTABLE 1
183 : : #define HJ_NEED_NEW_OUTER 2
184 : : #define HJ_SCAN_BUCKET 3
185 : : #define HJ_FILL_OUTER_TUPLE 4
186 : : #define HJ_FILL_INNER_TUPLES 5
187 : : #define HJ_FILL_OUTER_NULL_TUPLES 6
188 : : #define HJ_FILL_INNER_NULL_TUPLES 7
189 : : #define HJ_NEED_NEW_BATCH 8
190 : :
191 : : /* Returns true if doing null-fill on outer relation */
192 : : #define HJ_FILL_OUTER(hjstate) ((hjstate)->hj_NullInnerTupleSlot != NULL)
193 : : /* Returns true if doing null-fill on inner relation */
194 : : #define HJ_FILL_INNER(hjstate) ((hjstate)->hj_NullOuterTupleSlot != NULL)
195 : :
196 : : static TupleTableSlot *ExecHashJoinOuterGetTuple(PlanState *outerNode,
197 : : HashJoinState *hjstate,
198 : : uint32 *hashvalue);
199 : : static TupleTableSlot *ExecParallelHashJoinOuterGetTuple(PlanState *outerNode,
200 : : HashJoinState *hjstate,
201 : : uint32 *hashvalue);
202 : : static TupleTableSlot *ExecHashJoinGetSavedTuple(HashJoinState *hjstate,
203 : : BufFile *file,
204 : : uint32 *hashvalue,
205 : : TupleTableSlot *tupleSlot);
206 : : static bool ExecHashJoinNewBatch(HashJoinState *hjstate);
207 : : static bool ExecParallelHashJoinNewBatch(HashJoinState *hjstate);
208 : : static void ExecParallelHashJoinPartitionOuter(HashJoinState *hjstate);
209 : :
210 : :
211 : : /* ----------------------------------------------------------------
212 : : * ExecHashJoinImpl
213 : : *
214 : : * This function implements the Hybrid Hashjoin algorithm. It is marked
215 : : * with an always-inline attribute so that ExecHashJoin() and
216 : : * ExecParallelHashJoin() can inline it. Compilers that respect the
217 : : * attribute should create versions specialized for parallel == true and
218 : : * parallel == false with unnecessary branches removed.
219 : : *
220 : : * Note: the relation we build hash table on is the "inner"
221 : : * the other one is "outer".
222 : : * ----------------------------------------------------------------
223 : : */
224 : : static pg_attribute_always_inline TupleTableSlot *
225 : 8498890 : ExecHashJoinImpl(PlanState *pstate, bool parallel)
226 : : {
227 : 8498890 : HashJoinState *node = castNode(HashJoinState, pstate);
228 : : PlanState *outerNode;
229 : : HashState *hashNode;
230 : : ExprState *joinqual;
231 : : ExprState *otherqual;
232 : : ExprContext *econtext;
233 : : HashJoinTable hashtable;
234 : : TupleTableSlot *outerTupleSlot;
235 : : uint32 hashvalue;
236 : : int batchno;
237 : : ParallelHashJoinState *parallel_state;
238 : :
239 : : /*
240 : : * get information from HashJoin node
241 : : */
242 : 8498890 : joinqual = node->js.joinqual;
243 : 8498890 : otherqual = node->js.ps.qual;
244 : 8498890 : hashNode = (HashState *) innerPlanState(node);
245 : 8498890 : outerNode = outerPlanState(node);
246 : 8498890 : hashtable = node->hj_HashTable;
247 : 8498890 : econtext = node->js.ps.ps_ExprContext;
248 : 8498890 : parallel_state = hashNode->parallel_state;
249 : :
250 : : /*
251 : : * Reset per-tuple memory context to free any expression evaluation
252 : : * storage allocated in the previous tuple cycle.
253 : : */
254 : 8498890 : ResetExprContext(econtext);
255 : :
256 : : /*
257 : : * run the hash join state machine
258 : : */
259 : : for (;;)
260 : : {
261 : : /*
262 : : * It's possible to iterate this loop many times before returning a
263 : : * tuple, in some pathological cases such as needing to move much of
264 : : * the current batch to a later batch. So let's check for interrupts
265 : : * each time through.
266 : : */
267 [ + + ]: 28431114 : CHECK_FOR_INTERRUPTS();
268 : :
269 [ + + + + : 28431114 : switch (node->hj_JoinState)
+ + + +
- ]
270 : : {
271 : 22712 : case HJ_BUILD_HASHTABLE:
272 : :
273 : : /*
274 : : * First time through: build hash table for inner relation.
275 : : */
276 : : Assert(hashtable == NULL);
277 : :
278 : : /*
279 : : * If the outer relation is completely empty, and it's not
280 : : * right/right-anti/full join, we can quit without building
281 : : * the hash table. However, for an inner join it is only a
282 : : * win to check this when the outer relation's startup cost is
283 : : * less than the projected cost of building the hash table.
284 : : * Otherwise it's best to build the hash table first and see
285 : : * if the inner relation is empty. (When it's a left join, we
286 : : * should always make this check, since we aren't going to be
287 : : * able to skip the join on the strength of an empty inner
288 : : * relation anyway.)
289 : : *
290 : : * If we are rescanning the join, we make use of information
291 : : * gained on the previous scan: don't bother to try the
292 : : * prefetch if the previous scan found the outer relation
293 : : * nonempty. This is not 100% reliable since with new
294 : : * parameters the outer relation might yield different
295 : : * results, but it's a good heuristic.
296 : : *
297 : : * The only way to make the check is to try to fetch a tuple
298 : : * from the outer plan node. If we succeed, we have to stash
299 : : * it away for later consumption by ExecHashJoinOuterGetTuple.
300 : : */
301 [ + + ]: 22712 : if (HJ_FILL_INNER(node))
302 : : {
303 : : /* no chance to not build the hash table */
304 : 2718 : node->hj_FirstOuterTupleSlot = NULL;
305 : : }
306 [ + + ]: 19994 : else if (parallel)
307 : : {
308 : : /*
309 : : * The empty-outer optimization is not implemented for
310 : : * shared hash tables, because no one participant can
311 : : * determine that there are no outer tuples, and it's not
312 : : * yet clear that it's worth the synchronization overhead
313 : : * of reaching consensus to figure that out. So we have
314 : : * to build the hash table.
315 : : */
316 : 228 : node->hj_FirstOuterTupleSlot = NULL;
317 : : }
318 [ + + ]: 19766 : else if (HJ_FILL_OUTER(node) ||
319 [ + + ]: 15537 : (outerNode->plan->startup_cost < hashNode->ps.plan->total_cost &&
320 [ + + ]: 13688 : !node->hj_OuterNotEmpty))
321 : : {
322 : 17135 : node->hj_FirstOuterTupleSlot = ExecProcNode(outerNode);
323 [ + + + + ]: 17135 : if (TupIsNull(node->hj_FirstOuterTupleSlot))
324 : : {
325 : 3637 : node->hj_OuterNotEmpty = false;
326 : 3637 : return NULL;
327 : : }
328 : : else
329 : 13498 : node->hj_OuterNotEmpty = true;
330 : : }
331 : : else
332 : 2631 : node->hj_FirstOuterTupleSlot = NULL;
333 : :
334 : : /*
335 : : * Create the hash table. If using Parallel Hash, then
336 : : * whoever gets here first will create the hash table and any
337 : : * later arrivals will merely attach to it.
338 : : */
339 : 19075 : hashtable = ExecHashTableCreate(hashNode);
340 : 19075 : node->hj_HashTable = hashtable;
341 : :
342 : : /*
343 : : * Execute the Hash node, to build the hash table. If using
344 : : * Parallel Hash, then we'll try to help hashing unless we
345 : : * arrived too late.
346 : : */
347 : 19075 : hashNode->hashtable = hashtable;
348 : 19075 : (void) MultiExecProcNode((PlanState *) hashNode);
349 : :
350 : : /*
351 : : * If the inner relation is completely empty, and we're not
352 : : * doing a left outer join, we can quit without scanning the
353 : : * outer relation. (If the inner relation contains only
354 : : * null-keyed tuples that we need to emit, we'll fall through
355 : : * and do the outer-relation scan. In principle we could go
356 : : * emit those tuples then quit, but it would complicate the
357 : : * state machine logic. The case seems rare enough to not be
358 : : * worth optimizing.)
359 : : */
360 [ + + ]: 19075 : if (hashtable->totalTuples == 0 &&
361 [ + + ]: 3318 : hashNode->null_tuple_store == NULL &&
362 [ + + ]: 3290 : !HJ_FILL_OUTER(node))
363 : : {
364 [ - + ]: 3096 : if (parallel)
365 : : {
366 : : /*
367 : : * Advance the build barrier to PHJ_BUILD_RUN before
368 : : * proceeding so we can negotiate resource cleanup.
369 : : */
370 : 0 : Barrier *build_barrier = ¶llel_state->build_barrier;
371 : :
372 [ # # ]: 0 : while (BarrierPhase(build_barrier) < PHJ_BUILD_RUN)
373 : 0 : BarrierArriveAndWait(build_barrier, 0);
374 : : }
375 : 3096 : return NULL;
376 : : }
377 : :
378 : : /*
379 : : * need to remember whether nbatch has increased since we
380 : : * began scanning the outer relation
381 : : */
382 : 15979 : hashtable->nbatch_outstart = hashtable->nbatch;
383 : :
384 : : /*
385 : : * Reset OuterNotEmpty for scan. (It's OK if we fetched a
386 : : * tuple above, because ExecHashJoinOuterGetTuple will
387 : : * immediately set it again.)
388 : : */
389 : 15979 : node->hj_OuterNotEmpty = false;
390 : :
391 [ + + ]: 15979 : if (parallel)
392 : 276 : {
393 : : Barrier *build_barrier;
394 : :
395 : 276 : build_barrier = ¶llel_state->build_barrier;
396 : : Assert(BarrierPhase(build_barrier) == PHJ_BUILD_HASH_OUTER ||
397 : : BarrierPhase(build_barrier) == PHJ_BUILD_RUN ||
398 : : BarrierPhase(build_barrier) == PHJ_BUILD_FREE);
399 [ + + ]: 276 : if (BarrierPhase(build_barrier) == PHJ_BUILD_HASH_OUTER)
400 : : {
401 : : /*
402 : : * If multi-batch, we need to hash the outer relation
403 : : * up front.
404 : : */
405 [ + + ]: 225 : if (hashtable->nbatch > 1)
406 : 121 : ExecParallelHashJoinPartitionOuter(node);
407 : 225 : BarrierArriveAndWait(build_barrier,
408 : : WAIT_EVENT_HASH_BUILD_HASH_OUTER);
409 : : Assert(BarrierPhase(build_barrier) == PHJ_BUILD_RUN);
410 : : }
411 : :
412 : : /*
413 : : * Each backend should now select a batch to work on.
414 : : * However, if we've already collected some null-keyed
415 : : * tuples, dump them first. (That is critical when we
416 : : * arrive late enough that no more batches are available;
417 : : * otherwise we'd fail to dump those tuples at all.)
418 : : */
419 : 276 : hashtable->curbatch = -1;
420 : :
421 [ + + ]: 276 : if (node->hj_NullOuterTupleStore)
422 : 4 : node->hj_JoinState = HJ_FILL_OUTER_NULL_TUPLES;
423 [ + + ]: 272 : else if (hashNode->null_tuple_store)
424 : 9 : node->hj_JoinState = HJ_FILL_INNER_NULL_TUPLES;
425 : : else
426 : 263 : node->hj_JoinState = HJ_NEED_NEW_BATCH;
427 : :
428 : 276 : continue;
429 : : }
430 : : else
431 : 15703 : node->hj_JoinState = HJ_NEED_NEW_OUTER;
432 : :
433 : : pg_fallthrough;
434 : :
435 : 14025178 : case HJ_NEED_NEW_OUTER:
436 : :
437 : : /*
438 : : * We don't have an outer tuple, try to get the next one
439 : : */
440 [ + + ]: 14025178 : if (parallel)
441 : : outerTupleSlot =
442 : 1444668 : ExecParallelHashJoinOuterGetTuple(outerNode, node,
443 : : &hashvalue);
444 : : else
445 : : outerTupleSlot =
446 : 12580510 : ExecHashJoinOuterGetTuple(outerNode, node, &hashvalue);
447 : :
448 [ + + - + ]: 14025178 : if (TupIsNull(outerTupleSlot))
449 : : {
450 : : /* end of batch, or maybe whole join */
451 [ + + ]: 17918 : if (HJ_FILL_INNER(node))
452 : : {
453 : : /* set up to scan for unmatched inner tuples */
454 [ + + ]: 2564 : if (parallel)
455 : : {
456 : : /*
457 : : * Only one process is currently allowed to handle
458 : : * each batch's unmatched tuples, in a parallel
459 : : * join. However, each process must deal with any
460 : : * null-keyed tuples it found.
461 : : */
462 [ + + ]: 72 : if (ExecParallelPrepHashTableForUnmatched(node))
463 : 44 : node->hj_JoinState = HJ_FILL_INNER_TUPLES;
464 [ + + ]: 28 : else if (node->hj_NullOuterTupleStore)
465 : 6 : node->hj_JoinState = HJ_FILL_OUTER_NULL_TUPLES;
466 [ + + ]: 22 : else if (hashNode->null_tuple_store)
467 : 8 : node->hj_JoinState = HJ_FILL_INNER_NULL_TUPLES;
468 : : else
469 : 14 : node->hj_JoinState = HJ_NEED_NEW_BATCH;
470 : : }
471 : : else
472 : : {
473 : 2492 : ExecPrepHashTableForUnmatched(node);
474 : 2492 : node->hj_JoinState = HJ_FILL_INNER_TUPLES;
475 : : }
476 : : }
477 : : else
478 : : {
479 : : /* might have outer null-keyed tuples to fill */
480 : : Assert(hashNode->null_tuple_store == NULL);
481 [ + + ]: 15354 : if (node->hj_NullOuterTupleStore)
482 : 33 : node->hj_JoinState = HJ_FILL_OUTER_NULL_TUPLES;
483 : : else
484 : 15321 : node->hj_JoinState = HJ_NEED_NEW_BATCH;
485 : : }
486 : 17918 : continue;
487 : : }
488 : :
489 : 14007260 : econtext->ecxt_outertuple = outerTupleSlot;
490 : 14007260 : node->hj_MatchedOuter = false;
491 : :
492 : : /*
493 : : * Find the corresponding bucket for this tuple in the main
494 : : * hash table or skew hash table.
495 : : */
496 : 14007260 : node->hj_CurHashValue = hashvalue;
497 : 14007260 : ExecHashGetBucketAndBatch(hashtable, hashvalue,
498 : : &node->hj_CurBucketNo, &batchno);
499 : 14007260 : node->hj_CurSkewBucketNo = ExecHashGetSkewBucket(hashtable,
500 : : hashvalue);
501 : 14007260 : node->hj_CurTuple = NULL;
502 : :
503 : : /*
504 : : * The tuple might not belong to the current batch (where
505 : : * "current batch" includes the skew buckets if any).
506 : : */
507 [ + + ]: 14007260 : if (batchno != hashtable->curbatch &&
508 [ + + ]: 959656 : node->hj_CurSkewBucketNo == INVALID_SKEW_BUCKET_NO)
509 : 958856 : {
510 : : bool shouldFree;
511 : 958856 : MinimalTuple mintuple = ExecFetchSlotMinimalTuple(outerTupleSlot,
512 : : &shouldFree);
513 : :
514 : : /*
515 : : * Need to postpone this outer tuple to a later batch.
516 : : * Save it in the corresponding outer-batch file.
517 : : */
518 : : Assert(parallel_state == NULL);
519 : : Assert(batchno > hashtable->curbatch);
520 : 958856 : ExecHashJoinSaveTuple(mintuple, hashvalue,
521 : 958856 : &hashtable->outerBatchFile[batchno],
522 : : hashtable);
523 : :
524 [ + - ]: 958856 : if (shouldFree)
525 : 958856 : heap_free_minimal_tuple(mintuple);
526 : :
527 : : /* Loop around, staying in HJ_NEED_NEW_OUTER state */
528 : 958856 : continue;
529 : : }
530 : :
531 : : /* OK, let's scan the bucket for matches */
532 : 13048404 : node->hj_JoinState = HJ_SCAN_BUCKET;
533 : :
534 : : pg_fallthrough;
535 : :
536 : 17856128 : case HJ_SCAN_BUCKET:
537 : :
538 : : /*
539 : : * Scan the selected hash bucket for matches to current outer
540 : : */
541 [ + + ]: 17856128 : if (parallel)
542 : : {
543 [ + + ]: 2804080 : if (!ExecParallelScanHashBucket(node, econtext))
544 : : {
545 : : /* out of matches; check for possible outer-join fill */
546 : 1444028 : node->hj_JoinState = HJ_FILL_OUTER_TUPLE;
547 : 1444028 : continue;
548 : : }
549 : : }
550 : : else
551 : : {
552 [ + + ]: 15052048 : if (!ExecScanHashBucket(node, econtext))
553 : : {
554 : : /* out of matches; check for possible outer-join fill */
555 : 7796667 : node->hj_JoinState = HJ_FILL_OUTER_TUPLE;
556 : 7796667 : continue;
557 : : }
558 : : }
559 : :
560 : : /*
561 : : * In a right-semijoin, we only need the first match for each
562 : : * inner tuple.
563 : : */
564 [ + + + + ]: 8615773 : if (node->js.jointype == JOIN_RIGHT_SEMI &&
565 : 340 : HeapTupleHeaderHasMatch(HJTUPLE_MINTUPLE(node->hj_CurTuple)))
566 : 48 : continue;
567 : :
568 : : /*
569 : : * We've got a match, but still need to test non-hashed quals.
570 : : * ExecScanHashBucket already set up all the state needed to
571 : : * call ExecQual.
572 : : *
573 : : * If we pass the qual, then save state for next call and have
574 : : * ExecProject form the projection, store it in the tuple
575 : : * table, and return the slot.
576 : : *
577 : : * Only the joinquals determine tuple match status, but all
578 : : * quals must pass to actually return the tuple.
579 : : */
580 [ + + + + ]: 8615385 : if (joinqual == NULL || ExecQual(joinqual, econtext))
581 : : {
582 : 8504767 : node->hj_MatchedOuter = true;
583 : :
584 : : /*
585 : : * This is really only needed if HJ_FILL_INNER(node) or if
586 : : * we are in a right-semijoin, but we'll avoid the branch
587 : : * and just set it always.
588 : : */
589 [ + + ]: 8504767 : if (!HeapTupleHeaderHasMatch(HJTUPLE_MINTUPLE(node->hj_CurTuple)))
590 : 4054149 : HeapTupleHeaderSetMatch(HJTUPLE_MINTUPLE(node->hj_CurTuple));
591 : :
592 : : /* In an antijoin, we never return a matched tuple */
593 [ + + ]: 8504767 : if (node->js.jointype == JOIN_ANTI)
594 : : {
595 : 875571 : node->hj_JoinState = HJ_NEED_NEW_OUTER;
596 : 875571 : continue;
597 : : }
598 : :
599 : : /*
600 : : * If we only need to consider the first matching inner
601 : : * tuple, then advance to next outer tuple after we've
602 : : * processed this one.
603 : : */
604 [ + + ]: 7629196 : if (node->js.single_match)
605 : 2932060 : node->hj_JoinState = HJ_NEED_NEW_OUTER;
606 : :
607 : : /*
608 : : * In a right-antijoin, we never return a matched tuple.
609 : : * If it's not an inner_unique join, we need to stay on
610 : : * the current outer tuple to continue scanning the inner
611 : : * side for matches.
612 : : */
613 [ + + ]: 7629196 : if (node->js.jointype == JOIN_RIGHT_ANTI)
614 : 17590 : continue;
615 : :
616 [ + + + + ]: 7611606 : if (otherqual == NULL || ExecQual(otherqual, econtext))
617 : 7488908 : return ExecProject(node->js.ps.ps_ProjInfo);
618 : : else
619 [ - + ]: 122698 : InstrCountFiltered2(node, 1);
620 : : }
621 : : else
622 [ - + ]: 110618 : InstrCountFiltered1(node, 1);
623 : 233316 : break;
624 : :
625 : 9240695 : case HJ_FILL_OUTER_TUPLE:
626 : :
627 : : /*
628 : : * The current outer tuple has run out of matches, so check
629 : : * whether to emit a dummy outer-join tuple. Whether we emit
630 : : * one or not, the next state is NEED_NEW_OUTER.
631 : : */
632 : 9240695 : node->hj_JoinState = HJ_NEED_NEW_OUTER;
633 : :
634 [ + + ]: 9240695 : if (!node->hj_MatchedOuter &&
635 [ + + ]: 5949432 : HJ_FILL_OUTER(node))
636 : : {
637 : : /*
638 : : * Generate a fake join tuple with nulls for the inner
639 : : * tuple, and return it if it passes the non-join quals.
640 : : */
641 : 1348595 : econtext->ecxt_innertuple = node->hj_NullInnerTupleSlot;
642 : :
643 [ + + + + ]: 1348595 : if (otherqual == NULL || ExecQual(otherqual, econtext))
644 : 661968 : return ExecProject(node->js.ps.ps_ProjInfo);
645 : : else
646 [ - + ]: 686627 : InstrCountFiltered2(node, 1);
647 : : }
648 : 8578727 : break;
649 : :
650 : 331947 : case HJ_FILL_INNER_TUPLES:
651 : :
652 : : /*
653 : : * We have finished a batch, but we are doing
654 : : * right/right-anti/full join, so any unmatched inner tuples
655 : : * in the hashtable have to be emitted before we continue to
656 : : * the next batch.
657 : : */
658 [ + + + + ]: 583846 : if (!(parallel ? ExecParallelScanHashTableForUnmatched(node, econtext)
659 : 251899 : : ExecScanHashTableForUnmatched(node, econtext)))
660 : : {
661 : : /* no more unmatched tuples, but maybe there are nulls */
662 [ + + ]: 2532 : if (node->hj_NullOuterTupleStore)
663 : 43 : node->hj_JoinState = HJ_FILL_OUTER_NULL_TUPLES;
664 [ + + ]: 2489 : else if (hashNode->null_tuple_store)
665 : 60 : node->hj_JoinState = HJ_FILL_INNER_NULL_TUPLES;
666 : : else
667 : 2429 : node->hj_JoinState = HJ_NEED_NEW_BATCH;
668 : 2532 : continue;
669 : : }
670 : :
671 : : /*
672 : : * Generate a fake join tuple with nulls for the outer tuple,
673 : : * and return it if it passes the non-join quals.
674 : : */
675 : 329415 : econtext->ecxt_outertuple = node->hj_NullOuterTupleSlot;
676 : :
677 [ + + + + ]: 329415 : if (otherqual == NULL || ExecQual(otherqual, econtext))
678 : 324336 : return ExecProject(node->js.ps.ps_ProjInfo);
679 : : else
680 [ - + ]: 5079 : InstrCountFiltered2(node, 1);
681 : 5079 : break;
682 : :
683 : 207 : case HJ_FILL_OUTER_NULL_TUPLES:
684 : :
685 : : /*
686 : : * We have finished a batch, but we are doing left/full join,
687 : : * so any null-keyed outer tuples have to be emitted before we
688 : : * continue to the next batch.
689 : : *
690 : : * (We could delay this till the end of the join, but there
691 : : * seems little percentage in that.)
692 : : *
693 : : * We have to use tuplestore_gettupleslot_force because
694 : : * hj_OuterTupleSlot may not be able to store a MinimalTuple.
695 : : */
696 [ + + ]: 426 : while (tuplestore_gettupleslot_force(node->hj_NullOuterTupleStore,
697 : : true, false,
698 : : node->hj_OuterTupleSlot))
699 : : {
700 : : /*
701 : : * Generate a fake join tuple with nulls for the inner
702 : : * tuple, and return it if it passes the non-join quals.
703 : : */
704 : 133 : econtext->ecxt_outertuple = node->hj_OuterTupleSlot;
705 : 133 : econtext->ecxt_innertuple = node->hj_NullInnerTupleSlot;
706 : :
707 [ + + + + ]: 133 : if (otherqual == NULL || ExecQual(otherqual, econtext))
708 : 121 : return ExecProject(node->js.ps.ps_ProjInfo);
709 : : else
710 [ - + ]: 12 : InstrCountFiltered2(node, 1);
711 : :
712 : 12 : ResetExprContext(econtext);
713 : :
714 : : /* allow this loop to be cancellable */
715 [ - + ]: 12 : CHECK_FOR_INTERRUPTS();
716 : : }
717 : :
718 : : /* We don't need the tuplestore any more, so discard it. */
719 : 86 : tuplestore_end(node->hj_NullOuterTupleStore);
720 : 86 : node->hj_NullOuterTupleStore = NULL;
721 : :
722 : : /* Fill inner tuples too if it's a full join, else advance. */
723 [ + + ]: 86 : if (hashNode->null_tuple_store)
724 : 24 : node->hj_JoinState = HJ_FILL_INNER_NULL_TUPLES;
725 : : else
726 : 62 : node->hj_JoinState = HJ_NEED_NEW_BATCH;
727 : 86 : break;
728 : :
729 : 168 : case HJ_FILL_INNER_NULL_TUPLES:
730 : :
731 : : /*
732 : : * We have finished a batch, but we are doing
733 : : * right/right-anti/full join, so any null-keyed inner tuples
734 : : * have to be emitted before we continue to the next batch.
735 : : *
736 : : * (We could delay this till the end of the join, but there
737 : : * seems little percentage in that.)
738 : : */
739 [ + + ]: 340 : while (tuplestore_gettupleslot(hashNode->null_tuple_store,
740 : : true, false,
741 : : node->hj_HashTupleSlot))
742 : : {
743 : : /*
744 : : * Generate a fake join tuple with nulls for the outer
745 : : * tuple, and return it if it passes the non-join quals.
746 : : */
747 : 75 : econtext->ecxt_outertuple = node->hj_NullOuterTupleSlot;
748 : 75 : econtext->ecxt_innertuple = node->hj_HashTupleSlot;
749 : :
750 [ + + + + ]: 75 : if (otherqual == NULL || ExecQual(otherqual, econtext))
751 : 71 : return ExecProject(node->js.ps.ps_ProjInfo);
752 : : else
753 [ - + ]: 4 : InstrCountFiltered2(node, 1);
754 : :
755 : 4 : ResetExprContext(econtext);
756 : :
757 : : /* allow this loop to be cancellable */
758 [ - + ]: 4 : CHECK_FOR_INTERRUPTS();
759 : : }
760 : :
761 : : /*
762 : : * Ideally we'd discard the tuplestore now, but we can't
763 : : * because we might need it for rescans.
764 : : */
765 : :
766 : : /* Now we can advance to the next batch. */
767 : 97 : node->hj_JoinState = HJ_NEED_NEW_BATCH;
768 : 97 : break;
769 : :
770 : 18186 : case HJ_NEED_NEW_BATCH:
771 : :
772 : : /*
773 : : * Try to advance to next batch. Done if there are no more.
774 : : */
775 [ + + ]: 18186 : if (parallel)
776 : : {
777 [ + + ]: 916 : if (!ExecParallelHashJoinNewBatch(node))
778 : 276 : return NULL; /* end of parallel-aware join */
779 : : }
780 : : else
781 : : {
782 [ + + ]: 17270 : if (!ExecHashJoinNewBatch(node))
783 : 16477 : return NULL; /* end of parallel-oblivious join */
784 : : }
785 : 1433 : node->hj_JoinState = HJ_NEED_NEW_OUTER;
786 : 1433 : break;
787 : :
788 : 0 : default:
789 [ # # ]: 0 : elog(ERROR, "unrecognized hashjoin state: %d",
790 : : (int) node->hj_JoinState);
791 : : }
792 : : }
793 : : }
794 : :
795 : : /* ----------------------------------------------------------------
796 : : * ExecHashJoin
797 : : *
798 : : * Parallel-oblivious version.
799 : : * ----------------------------------------------------------------
800 : : */
801 : : static TupleTableSlot * /* return: a tuple or NULL */
802 : 6978522 : ExecHashJoin(PlanState *pstate)
803 : : {
804 : : /*
805 : : * On sufficiently smart compilers this should be inlined with the
806 : : * parallel-aware branches removed.
807 : : */
808 : 6978522 : return ExecHashJoinImpl(pstate, false);
809 : : }
810 : :
811 : : /* ----------------------------------------------------------------
812 : : * ExecParallelHashJoin
813 : : *
814 : : * Parallel-aware version.
815 : : * ----------------------------------------------------------------
816 : : */
817 : : static TupleTableSlot * /* return: a tuple or NULL */
818 : 1520368 : ExecParallelHashJoin(PlanState *pstate)
819 : : {
820 : : /*
821 : : * On sufficiently smart compilers this should be inlined with the
822 : : * parallel-oblivious branches removed.
823 : : */
824 : 1520368 : return ExecHashJoinImpl(pstate, true);
825 : : }
826 : :
827 : : /* ----------------------------------------------------------------
828 : : * ExecInitHashJoin
829 : : *
830 : : * Init routine for HashJoin node.
831 : : * ----------------------------------------------------------------
832 : : */
833 : : HashJoinState *
834 : 27778 : ExecInitHashJoin(HashJoin *node, EState *estate, int eflags)
835 : : {
836 : : HashJoinState *hjstate;
837 : : Plan *outerNode;
838 : : Hash *hashNode;
839 : : TupleDesc outerDesc,
840 : : innerDesc;
841 : : const TupleTableSlotOps *ops;
842 : :
843 : : /* check for unsupported flags */
844 : : Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
845 : :
846 : : /*
847 : : * create state structure
848 : : */
849 : 27778 : hjstate = makeNode(HashJoinState);
850 : 27778 : hjstate->js.ps.plan = (Plan *) node;
851 : 27778 : hjstate->js.ps.state = estate;
852 : :
853 : : /*
854 : : * See ExecHashJoinInitializeDSM() and ExecHashJoinInitializeWorker()
855 : : * where this function may be replaced with a parallel version, if we
856 : : * managed to launch a parallel query.
857 : : */
858 : 27778 : hjstate->js.ps.ExecProcNode = ExecHashJoin;
859 : 27778 : hjstate->js.jointype = node->join.jointype;
860 : :
861 : : /*
862 : : * Miscellaneous initialization
863 : : *
864 : : * create expression context for node
865 : : */
866 : 27778 : ExecAssignExprContext(estate, &hjstate->js.ps);
867 : :
868 : : /*
869 : : * initialize child nodes
870 : : *
871 : : * Note: we could suppress the REWIND flag for the inner input, which
872 : : * would amount to betting that the hash will be a single batch. Not
873 : : * clear if this would be a win or not.
874 : : */
875 : 27778 : outerNode = outerPlan(node);
876 : 27778 : hashNode = (Hash *) innerPlan(node);
877 : :
878 : 27778 : outerPlanState(hjstate) = ExecInitNode(outerNode, estate, eflags);
879 : 27778 : outerDesc = ExecGetResultType(outerPlanState(hjstate));
880 : 27778 : innerPlanState(hjstate) = ExecInitNode((Plan *) hashNode, estate, eflags);
881 : 27778 : innerDesc = ExecGetResultType(innerPlanState(hjstate));
882 : :
883 : : /*
884 : : * Initialize result slot, type and projection.
885 : : */
886 : 27778 : ExecInitResultTupleSlotTL(&hjstate->js.ps, &TTSOpsVirtual);
887 : 27778 : ExecAssignProjectionInfo(&hjstate->js.ps, NULL);
888 : :
889 : : /*
890 : : * tuple table initialization
891 : : */
892 : 27778 : ops = ExecGetResultSlotOps(outerPlanState(hjstate), NULL);
893 : 27778 : hjstate->hj_OuterTupleSlot = ExecInitExtraTupleSlot(estate, outerDesc,
894 : : ops);
895 : :
896 : : /*
897 : : * detect whether we need only consider the first matching inner tuple
898 : : */
899 [ + + ]: 39429 : hjstate->js.single_match = (node->join.inner_unique ||
900 [ + + ]: 11651 : node->join.jointype == JOIN_SEMI);
901 : :
902 : : /* set up null tuples for outer joins, if needed */
903 [ + + + + : 27778 : switch (node->join.jointype)
- ]
904 : : {
905 : 19514 : case JOIN_INNER:
906 : : case JOIN_SEMI:
907 : : case JOIN_RIGHT_SEMI:
908 : 19514 : break;
909 : 4760 : case JOIN_LEFT:
910 : : case JOIN_ANTI:
911 : 4760 : hjstate->hj_NullInnerTupleSlot =
912 : 4760 : ExecInitNullTupleSlot(estate, innerDesc, &TTSOpsVirtual);
913 : 4760 : break;
914 : 2770 : case JOIN_RIGHT:
915 : : case JOIN_RIGHT_ANTI:
916 : 2770 : hjstate->hj_NullOuterTupleSlot =
917 : 2770 : ExecInitNullTupleSlot(estate, outerDesc, &TTSOpsVirtual);
918 : 2770 : break;
919 : 734 : case JOIN_FULL:
920 : 734 : hjstate->hj_NullOuterTupleSlot =
921 : 734 : ExecInitNullTupleSlot(estate, outerDesc, &TTSOpsVirtual);
922 : 734 : hjstate->hj_NullInnerTupleSlot =
923 : 734 : ExecInitNullTupleSlot(estate, innerDesc, &TTSOpsVirtual);
924 : 734 : break;
925 : 0 : default:
926 [ # # ]: 0 : elog(ERROR, "unrecognized join type: %d",
927 : : (int) node->join.jointype);
928 : : }
929 : :
930 : : /*
931 : : * now for some voodoo. our temporary tuple slot is actually the result
932 : : * tuple slot of the Hash node (which is our inner plan). we can do this
933 : : * because Hash nodes don't return tuples via ExecProcNode() -- instead
934 : : * the hash join node uses ExecScanHashBucket() to get at the contents of
935 : : * the hash table. -cim 6/9/91
936 : : */
937 : : {
938 : 27778 : HashState *hashstate = (HashState *) innerPlanState(hjstate);
939 : 27778 : Hash *hash = (Hash *) hashstate->ps.plan;
940 : 27778 : TupleTableSlot *slot = hashstate->ps.ps_ResultTupleSlot;
941 : : Oid *outer_hashfuncid;
942 : : Oid *inner_hashfuncid;
943 : : bool *hash_strict;
944 : : ListCell *lc;
945 : : int nkeys;
946 : :
947 : :
948 : 27778 : hjstate->hj_HashTupleSlot = slot;
949 : :
950 : : /*
951 : : * Build ExprStates to obtain hash values for either side of the join.
952 : : * Note: must build the ExprStates before ExecHashTableCreate() so we
953 : : * properly attribute any SubPlans that exist in the hash expressions
954 : : * to the correct PlanState.
955 : : */
956 : 27778 : nkeys = list_length(node->hashoperators);
957 : :
958 : 27778 : outer_hashfuncid = palloc_array(Oid, nkeys);
959 : 27778 : inner_hashfuncid = palloc_array(Oid, nkeys);
960 : 27778 : hash_strict = palloc_array(bool, nkeys);
961 : :
962 : : /*
963 : : * Determine the hash function for each side of the join for the given
964 : : * join operator, and detect whether the join operator is strict.
965 : : */
966 [ + - + + : 58490 : foreach(lc, node->hashoperators)
+ + ]
967 : : {
968 : 30712 : Oid hashop = lfirst_oid(lc);
969 : 30712 : int i = foreach_current_index(lc);
970 : :
971 [ - + ]: 30712 : if (!get_op_hash_functions(hashop,
972 : 30712 : &outer_hashfuncid[i],
973 : 30712 : &inner_hashfuncid[i]))
974 [ # # ]: 0 : elog(ERROR,
975 : : "could not find hash function for hash operator %u",
976 : : hashop);
977 : 30712 : hash_strict[i] = op_strict(hashop);
978 : : }
979 : :
980 : : /*
981 : : * Build an ExprState to generate the hash value for the expressions
982 : : * on the outer side of the join.
983 : : */
984 : 27778 : hjstate->hj_OuterHash =
985 : 27778 : ExecBuildHash32Expr(hjstate->js.ps.ps_ResultTupleDesc,
986 : : hjstate->js.ps.resultops,
987 : : outer_hashfuncid,
988 : 27778 : node->hashcollations,
989 : 27778 : node->hashkeys,
990 : : hash_strict,
991 : : &hjstate->js.ps,
992 : : 0);
993 : :
994 : : /* As above, but for the inner side of the join */
995 : 27778 : hashstate->hash_expr =
996 : 27778 : ExecBuildHash32Expr(hashstate->ps.ps_ResultTupleDesc,
997 : : hashstate->ps.resultops,
998 : : inner_hashfuncid,
999 : 27778 : node->hashcollations,
1000 : 27778 : hash->hashkeys,
1001 : : hash_strict,
1002 : : &hashstate->ps,
1003 : : 0);
1004 : :
1005 : : /* Remember whether we need to save tuples with null join keys */
1006 : 27778 : hjstate->hj_KeepNullTuples = HJ_FILL_OUTER(hjstate);
1007 : 27778 : hashstate->keep_null_tuples = HJ_FILL_INNER(hjstate);
1008 : :
1009 : : /*
1010 : : * Set up the skew table hash function while we have a record of the
1011 : : * first key's hash function Oid.
1012 : : */
1013 [ + + ]: 27778 : if (OidIsValid(hash->skewTable))
1014 : : {
1015 : 20490 : hashstate->skew_hashfunction = palloc0_object(FmgrInfo);
1016 : 20490 : hashstate->skew_collation = linitial_oid(node->hashcollations);
1017 : 20490 : fmgr_info(outer_hashfuncid[0], hashstate->skew_hashfunction);
1018 : : }
1019 : :
1020 : : /* no need to keep these */
1021 : 27778 : pfree(outer_hashfuncid);
1022 : 27778 : pfree(inner_hashfuncid);
1023 : 27778 : pfree(hash_strict);
1024 : : }
1025 : :
1026 : : /*
1027 : : * initialize child expressions
1028 : : */
1029 : 27778 : hjstate->js.ps.qual =
1030 : 27778 : ExecInitQual(node->join.plan.qual, (PlanState *) hjstate);
1031 : 27778 : hjstate->js.joinqual =
1032 : 27778 : ExecInitQual(node->join.joinqual, (PlanState *) hjstate);
1033 : 27778 : hjstate->hashclauses =
1034 : 27778 : ExecInitQual(node->hashclauses, (PlanState *) hjstate);
1035 : :
1036 : : /*
1037 : : * initialize hash-specific info
1038 : : */
1039 : 27778 : hjstate->hj_HashTable = NULL;
1040 : 27778 : hjstate->hj_NullOuterTupleStore = NULL;
1041 : 27778 : hjstate->hj_FirstOuterTupleSlot = NULL;
1042 : :
1043 : 27778 : hjstate->hj_CurHashValue = 0;
1044 : 27778 : hjstate->hj_CurBucketNo = 0;
1045 : 27778 : hjstate->hj_CurSkewBucketNo = INVALID_SKEW_BUCKET_NO;
1046 : 27778 : hjstate->hj_CurTuple = NULL;
1047 : :
1048 : 27778 : hjstate->hj_JoinState = HJ_BUILD_HASHTABLE;
1049 : 27778 : hjstate->hj_MatchedOuter = false;
1050 : 27778 : hjstate->hj_OuterNotEmpty = false;
1051 : :
1052 : 27778 : return hjstate;
1053 : : }
1054 : :
1055 : : /* ----------------------------------------------------------------
1056 : : * ExecEndHashJoin
1057 : : *
1058 : : * clean up routine for HashJoin node
1059 : : * ----------------------------------------------------------------
1060 : : */
1061 : : void
1062 : 27706 : ExecEndHashJoin(HashJoinState *node)
1063 : : {
1064 : 27706 : HashState *hashNode = castNode(HashState, innerPlanState(node));
1065 : :
1066 : : /*
1067 : : * Free tuple stores if we made them (must do this before
1068 : : * ExecHashTableDestroy deletes hashCxt)
1069 : : */
1070 [ - + ]: 27706 : if (node->hj_NullOuterTupleStore)
1071 : : {
1072 : 0 : tuplestore_end(node->hj_NullOuterTupleStore);
1073 : 0 : node->hj_NullOuterTupleStore = NULL;
1074 : : }
1075 [ + + ]: 27706 : if (hashNode->null_tuple_store)
1076 : : {
1077 : 65 : tuplestore_end(hashNode->null_tuple_store);
1078 : 65 : hashNode->null_tuple_store = NULL;
1079 : : }
1080 : :
1081 : : /*
1082 : : * Free hash table
1083 : : */
1084 [ + + ]: 27706 : if (node->hj_HashTable)
1085 : : {
1086 : 18088 : ExecHashTableDestroy(node->hj_HashTable);
1087 : 18088 : node->hj_HashTable = NULL;
1088 : : }
1089 : :
1090 : : /*
1091 : : * clean up subtrees
1092 : : */
1093 : 27706 : ExecEndNode(outerPlanState(node));
1094 : 27706 : ExecEndNode(innerPlanState(node));
1095 : 27706 : }
1096 : :
1097 : : /*
1098 : : * ExecHashJoinOuterGetTuple
1099 : : *
1100 : : * get the next outer tuple for a parallel oblivious hashjoin: either by
1101 : : * executing the outer plan node in the first pass, or from the temp
1102 : : * files for the hashjoin batches.
1103 : : *
1104 : : * Returns a null slot if no more outer tuples (within the current batch).
1105 : : *
1106 : : * On success, the tuple's hash value is stored at *hashvalue --- this is
1107 : : * either originally computed, or re-read from the temp file.
1108 : : */
1109 : : static TupleTableSlot *
1110 : 12580510 : ExecHashJoinOuterGetTuple(PlanState *outerNode,
1111 : : HashJoinState *hjstate,
1112 : : uint32 *hashvalue)
1113 : : {
1114 : 12580510 : HashJoinTable hashtable = hjstate->hj_HashTable;
1115 : 12580510 : int curbatch = hashtable->curbatch;
1116 : : TupleTableSlot *slot;
1117 : :
1118 [ + + ]: 12580510 : if (curbatch == 0) /* if it is the first pass */
1119 : : {
1120 : : /*
1121 : : * Check to see if first outer tuple was already fetched by
1122 : : * ExecHashJoin() and not used yet.
1123 : : */
1124 : 11620861 : slot = hjstate->hj_FirstOuterTupleSlot;
1125 [ + + + - ]: 11620861 : if (!TupIsNull(slot))
1126 : 11437 : hjstate->hj_FirstOuterTupleSlot = NULL;
1127 : : else
1128 : 11609424 : slot = ExecProcNode(outerNode);
1129 : :
1130 [ + + + + ]: 11621597 : while (!TupIsNull(slot))
1131 : : {
1132 : : bool isnull;
1133 : :
1134 : : /*
1135 : : * We have to compute the tuple's hash value.
1136 : : */
1137 : 11605112 : ExprContext *econtext = hjstate->js.ps.ps_ExprContext;
1138 : :
1139 : 11605112 : econtext->ecxt_outertuple = slot;
1140 : :
1141 : 11605112 : ResetExprContext(econtext);
1142 : :
1143 : 11605112 : *hashvalue = DatumGetUInt32(ExecEvalExprSwitchContext(hjstate->hj_OuterHash,
1144 : : econtext,
1145 : : &isnull));
1146 : :
1147 [ + + ]: 11605112 : if (!isnull)
1148 : : {
1149 : : /* normal case with a non-null join key */
1150 : : /* remember outer relation is not empty for possible rescan */
1151 : 11604376 : hjstate->hj_OuterNotEmpty = true;
1152 : :
1153 : 11604376 : return slot;
1154 : : }
1155 [ + + ]: 736 : else if (hjstate->hj_KeepNullTuples)
1156 : : {
1157 : : /* null join key, but we must save tuple to be emitted later */
1158 [ + + ]: 121 : if (hjstate->hj_NullOuterTupleStore == NULL)
1159 : 74 : hjstate->hj_NullOuterTupleStore = ExecHashBuildNullTupleStore(hashtable);
1160 : 121 : tuplestore_puttupleslot(hjstate->hj_NullOuterTupleStore, slot);
1161 : : }
1162 : :
1163 : : /*
1164 : : * That tuple couldn't match because of a NULL, so discard it and
1165 : : * continue with the next one.
1166 : : */
1167 : 736 : slot = ExecProcNode(outerNode);
1168 : : }
1169 : : }
1170 [ + - ]: 959649 : else if (curbatch < hashtable->nbatch)
1171 : : {
1172 : 959649 : BufFile *file = hashtable->outerBatchFile[curbatch];
1173 : :
1174 : : /*
1175 : : * In outer-join cases, we could get here even though the batch file
1176 : : * is empty.
1177 : : */
1178 [ - + ]: 959649 : if (file == NULL)
1179 : 0 : return NULL;
1180 : :
1181 : 959649 : slot = ExecHashJoinGetSavedTuple(hjstate,
1182 : : file,
1183 : : hashvalue,
1184 : : hjstate->hj_OuterTupleSlot);
1185 [ + + + - ]: 959649 : if (!TupIsNull(slot))
1186 : 958856 : return slot;
1187 : : }
1188 : :
1189 : : /* End of this batch */
1190 : 17278 : return NULL;
1191 : : }
1192 : :
1193 : : /*
1194 : : * ExecHashJoinOuterGetTuple variant for the parallel case.
1195 : : */
1196 : : static TupleTableSlot *
1197 : 1444668 : ExecParallelHashJoinOuterGetTuple(PlanState *outerNode,
1198 : : HashJoinState *hjstate,
1199 : : uint32 *hashvalue)
1200 : : {
1201 : 1444668 : HashJoinTable hashtable = hjstate->hj_HashTable;
1202 : 1444668 : int curbatch = hashtable->curbatch;
1203 : : TupleTableSlot *slot;
1204 : :
1205 : : /*
1206 : : * In the Parallel Hash case we only run the outer plan directly for
1207 : : * single-batch hash joins. Otherwise we have to go to batch files, even
1208 : : * for batch 0.
1209 : : */
1210 [ + + + + ]: 1444668 : if (curbatch == 0 && hashtable->nbatch == 1)
1211 : : {
1212 : 644119 : slot = ExecProcNode(outerNode);
1213 : :
1214 [ + + + - ]: 644139 : while (!TupIsNull(slot))
1215 : : {
1216 : : bool isnull;
1217 : :
1218 : 644024 : ExprContext *econtext = hjstate->js.ps.ps_ExprContext;
1219 : :
1220 : 644024 : econtext->ecxt_outertuple = slot;
1221 : :
1222 : 644024 : ResetExprContext(econtext);
1223 : :
1224 : 644024 : *hashvalue = DatumGetUInt32(ExecEvalExprSwitchContext(hjstate->hj_OuterHash,
1225 : : econtext,
1226 : : &isnull));
1227 : :
1228 [ + + ]: 644024 : if (!isnull)
1229 : : {
1230 : : /* normal case with a non-null join key */
1231 : 644004 : return slot;
1232 : : }
1233 [ + + ]: 20 : else if (hjstate->hj_KeepNullTuples)
1234 : : {
1235 : : /* null join key, but we must save tuple to be emitted later */
1236 [ + - ]: 8 : if (hjstate->hj_NullOuterTupleStore == NULL)
1237 : 8 : hjstate->hj_NullOuterTupleStore = ExecHashBuildNullTupleStore(hashtable);
1238 : 8 : tuplestore_puttupleslot(hjstate->hj_NullOuterTupleStore, slot);
1239 : : }
1240 : :
1241 : : /*
1242 : : * That tuple couldn't match because of a NULL, so discard it and
1243 : : * continue with the next one.
1244 : : */
1245 : 20 : slot = ExecProcNode(outerNode);
1246 : : }
1247 : : }
1248 [ + - ]: 800549 : else if (curbatch < hashtable->nbatch)
1249 : : {
1250 : : MinimalTuple tuple;
1251 : :
1252 : 800549 : tuple = sts_parallel_scan_next(hashtable->batches[curbatch].outer_tuples,
1253 : : hashvalue);
1254 [ + + ]: 800549 : if (tuple != NULL)
1255 : : {
1256 : 800024 : ExecForceStoreMinimalTuple(tuple,
1257 : : hjstate->hj_OuterTupleSlot,
1258 : : false);
1259 : 800024 : slot = hjstate->hj_OuterTupleSlot;
1260 : 800024 : return slot;
1261 : : }
1262 : : else
1263 : 525 : ExecClearTuple(hjstate->hj_OuterTupleSlot);
1264 : : }
1265 : :
1266 : : /* End of this batch */
1267 : 640 : hashtable->batches[curbatch].outer_eof = true;
1268 : :
1269 : 640 : return NULL;
1270 : : }
1271 : :
1272 : : /*
1273 : : * ExecHashJoinNewBatch
1274 : : * switch to a new hashjoin batch
1275 : : *
1276 : : * Returns true if successful, false if there are no more batches.
1277 : : */
1278 : : static bool
1279 : 17270 : ExecHashJoinNewBatch(HashJoinState *hjstate)
1280 : : {
1281 : 17270 : HashJoinTable hashtable = hjstate->hj_HashTable;
1282 : : int nbatch;
1283 : : int curbatch;
1284 : : BufFile *innerFile;
1285 : : TupleTableSlot *slot;
1286 : : uint32 hashvalue;
1287 : :
1288 : 17270 : nbatch = hashtable->nbatch;
1289 : 17270 : curbatch = hashtable->curbatch;
1290 : :
1291 [ + + ]: 17270 : if (curbatch > 0)
1292 : : {
1293 : : /*
1294 : : * We no longer need the previous outer batch file; close it right
1295 : : * away to free disk space.
1296 : : */
1297 [ + - ]: 793 : if (hashtable->outerBatchFile[curbatch])
1298 : 793 : BufFileClose(hashtable->outerBatchFile[curbatch]);
1299 : 793 : hashtable->outerBatchFile[curbatch] = NULL;
1300 : : }
1301 : : else /* we just finished the first batch */
1302 : : {
1303 : : /*
1304 : : * Reset some of the skew optimization state variables, since we no
1305 : : * longer need to consider skew tuples after the first batch. The
1306 : : * memory context reset we are about to do will release the skew
1307 : : * hashtable itself.
1308 : : */
1309 : 16477 : hashtable->skewEnabled = false;
1310 : 16477 : hashtable->skewBucket = NULL;
1311 : 16477 : hashtable->skewBucketNums = NULL;
1312 : 16477 : hashtable->nSkewBuckets = 0;
1313 : 16477 : hashtable->spaceUsedSkew = 0;
1314 : : }
1315 : :
1316 : : /*
1317 : : * We can always skip over any batches that are completely empty on both
1318 : : * sides. We can sometimes skip over batches that are empty on only one
1319 : : * side, but there are exceptions:
1320 : : *
1321 : : * 1. In a left/full outer join, we have to process outer batches even if
1322 : : * the inner batch is empty. Similarly, in a right/right-anti/full outer
1323 : : * join, we have to process inner batches even if the outer batch is
1324 : : * empty.
1325 : : *
1326 : : * 2. If we have increased nbatch since the initial estimate, we have to
1327 : : * scan inner batches since they might contain tuples that need to be
1328 : : * reassigned to later inner batches.
1329 : : *
1330 : : * 3. Similarly, if we have increased nbatch since starting the outer
1331 : : * scan, we have to rescan outer batches in case they contain tuples that
1332 : : * need to be reassigned.
1333 : : */
1334 : 17270 : curbatch++;
1335 [ + + ]: 17270 : while (curbatch < nbatch &&
1336 [ - + ]: 793 : (hashtable->outerBatchFile[curbatch] == NULL ||
1337 [ - + ]: 793 : hashtable->innerBatchFile[curbatch] == NULL))
1338 : : {
1339 [ # # ]: 0 : if (hashtable->outerBatchFile[curbatch] &&
1340 [ # # ]: 0 : HJ_FILL_OUTER(hjstate))
1341 : 0 : break; /* must process due to rule 1 */
1342 [ # # ]: 0 : if (hashtable->innerBatchFile[curbatch] &&
1343 [ # # ]: 0 : HJ_FILL_INNER(hjstate))
1344 : 0 : break; /* must process due to rule 1 */
1345 [ # # ]: 0 : if (hashtable->innerBatchFile[curbatch] &&
1346 [ # # ]: 0 : nbatch != hashtable->nbatch_original)
1347 : 0 : break; /* must process due to rule 2 */
1348 [ # # ]: 0 : if (hashtable->outerBatchFile[curbatch] &&
1349 [ # # ]: 0 : nbatch != hashtable->nbatch_outstart)
1350 : 0 : break; /* must process due to rule 3 */
1351 : : /* We can ignore this batch. */
1352 : : /* Release associated temp files right away. */
1353 [ # # ]: 0 : if (hashtable->innerBatchFile[curbatch])
1354 : 0 : BufFileClose(hashtable->innerBatchFile[curbatch]);
1355 : 0 : hashtable->innerBatchFile[curbatch] = NULL;
1356 [ # # ]: 0 : if (hashtable->outerBatchFile[curbatch])
1357 : 0 : BufFileClose(hashtable->outerBatchFile[curbatch]);
1358 : 0 : hashtable->outerBatchFile[curbatch] = NULL;
1359 : 0 : curbatch++;
1360 : : }
1361 : :
1362 [ + + ]: 17270 : if (curbatch >= nbatch)
1363 : 16477 : return false; /* no more batches */
1364 : :
1365 : 793 : hashtable->curbatch = curbatch;
1366 : :
1367 : : /*
1368 : : * Reload the hash table with the new inner batch (which could be empty)
1369 : : */
1370 : 793 : ExecHashTableReset(hashtable);
1371 : :
1372 : 793 : innerFile = hashtable->innerBatchFile[curbatch];
1373 : :
1374 [ + - ]: 793 : if (innerFile != NULL)
1375 : : {
1376 [ - + ]: 793 : if (BufFileSeek(innerFile, 0, 0, SEEK_SET))
1377 [ # # ]: 0 : ereport(ERROR,
1378 : : (errcode_for_file_access(),
1379 : : errmsg("could not rewind hash-join temporary file")));
1380 : :
1381 [ + + ]: 2286313 : while ((slot = ExecHashJoinGetSavedTuple(hjstate,
1382 : : innerFile,
1383 : : &hashvalue,
1384 : : hjstate->hj_HashTupleSlot)))
1385 : : {
1386 : : /*
1387 : : * NOTE: some tuples may be sent to future batches. Also, it is
1388 : : * possible for hashtable->nbatch to be increased here!
1389 : : */
1390 : 2285520 : ExecHashTableInsert(hashtable, slot, hashvalue);
1391 : : }
1392 : :
1393 : : /*
1394 : : * after we build the hash table, the inner batch file is no longer
1395 : : * needed
1396 : : */
1397 : 793 : BufFileClose(innerFile);
1398 : 793 : hashtable->innerBatchFile[curbatch] = NULL;
1399 : : }
1400 : :
1401 : : /*
1402 : : * Rewind outer batch file (if present), so that we can start reading it.
1403 : : */
1404 [ + - ]: 793 : if (hashtable->outerBatchFile[curbatch] != NULL)
1405 : : {
1406 [ - + ]: 793 : if (BufFileSeek(hashtable->outerBatchFile[curbatch], 0, 0, SEEK_SET))
1407 [ # # ]: 0 : ereport(ERROR,
1408 : : (errcode_for_file_access(),
1409 : : errmsg("could not rewind hash-join temporary file")));
1410 : : }
1411 : :
1412 : 793 : return true;
1413 : : }
1414 : :
1415 : : /*
1416 : : * Choose a batch to work on, and attach to it. Returns true if successful,
1417 : : * false if there are no more batches.
1418 : : */
1419 : : static bool
1420 : 916 : ExecParallelHashJoinNewBatch(HashJoinState *hjstate)
1421 : : {
1422 : 916 : HashJoinTable hashtable = hjstate->hj_HashTable;
1423 : : int start_batchno;
1424 : : int batchno;
1425 : :
1426 : : /*
1427 : : * If we are a very slow worker, MultiExecParallelHash could have observed
1428 : : * build_barrier phase PHJ_BUILD_FREE and not bothered to set up batch
1429 : : * accessors. In that case we must be done.
1430 : : */
1431 [ - + ]: 916 : if (hashtable->batches == NULL)
1432 : 0 : return false;
1433 : :
1434 : : /*
1435 : : * If we were already attached to a batch, remember not to bother checking
1436 : : * it again, and detach from it (possibly freeing the hash table if we are
1437 : : * last to detach).
1438 : : */
1439 [ + + ]: 916 : if (hashtable->curbatch >= 0)
1440 : : {
1441 : 612 : hashtable->batches[hashtable->curbatch].done = true;
1442 : 612 : ExecHashTableDetachBatch(hashtable);
1443 : : }
1444 : :
1445 : : /*
1446 : : * Search for a batch that isn't done. We use an atomic counter to start
1447 : : * our search at a different batch in every participant when there are
1448 : : * more batches than participants.
1449 : : */
1450 : 916 : batchno = start_batchno =
1451 : 916 : pg_atomic_fetch_add_u32(&hashtable->parallel_state->distributor, 1) %
1452 : 916 : hashtable->nbatch;
1453 : : do
1454 : : {
1455 : : uint32 hashvalue;
1456 : : MinimalTuple tuple;
1457 : : TupleTableSlot *slot;
1458 : :
1459 [ + + ]: 2233 : if (!hashtable->batches[batchno].done)
1460 : : {
1461 : : SharedTuplestoreAccessor *inner_tuples;
1462 : 1192 : Barrier *batch_barrier =
1463 : 1192 : &hashtable->batches[batchno].shared->batch_barrier;
1464 : :
1465 [ + - + + : 1192 : switch (BarrierAttach(batch_barrier))
+ + - ]
1466 : : {
1467 : 388 : case PHJ_BATCH_ELECT:
1468 : :
1469 : : /* One backend allocates the hash table. */
1470 [ + - ]: 388 : if (BarrierArriveAndWait(batch_barrier,
1471 : : WAIT_EVENT_HASH_BATCH_ELECT))
1472 : 388 : ExecParallelHashTableAlloc(hashtable, batchno);
1473 : : pg_fallthrough;
1474 : :
1475 : : case PHJ_BATCH_ALLOCATE:
1476 : : /* Wait for allocation to complete. */
1477 : 388 : BarrierArriveAndWait(batch_barrier,
1478 : : WAIT_EVENT_HASH_BATCH_ALLOCATE);
1479 : : pg_fallthrough;
1480 : :
1481 : 409 : case PHJ_BATCH_LOAD:
1482 : : /* Start (or join in) loading tuples. */
1483 : 409 : ExecParallelHashTableSetCurrentBatch(hashtable, batchno);
1484 : 409 : inner_tuples = hashtable->batches[batchno].inner_tuples;
1485 : 409 : sts_begin_parallel_scan(inner_tuples);
1486 [ + + ]: 721521 : while ((tuple = sts_parallel_scan_next(inner_tuples,
1487 : : &hashvalue)))
1488 : : {
1489 : 721112 : ExecForceStoreMinimalTuple(tuple,
1490 : : hjstate->hj_HashTupleSlot,
1491 : : false);
1492 : 721112 : slot = hjstate->hj_HashTupleSlot;
1493 : 721112 : ExecParallelHashTableInsertCurrentBatch(hashtable, slot,
1494 : : hashvalue);
1495 : : }
1496 : 409 : sts_end_parallel_scan(inner_tuples);
1497 : 409 : BarrierArriveAndWait(batch_barrier,
1498 : : WAIT_EVENT_HASH_BATCH_LOAD);
1499 : : pg_fallthrough;
1500 : :
1501 : 640 : case PHJ_BATCH_PROBE:
1502 : :
1503 : : /*
1504 : : * This batch is ready to probe. Return control to
1505 : : * caller. We stay attached to batch_barrier so that the
1506 : : * hash table stays alive until everyone's finished
1507 : : * probing it, but no participant is allowed to wait at
1508 : : * this barrier again (or else a deadlock could occur).
1509 : : * All attached participants must eventually detach from
1510 : : * the barrier and one worker must advance the phase so
1511 : : * that the final phase is reached.
1512 : : */
1513 : 640 : ExecParallelHashTableSetCurrentBatch(hashtable, batchno);
1514 : 640 : sts_begin_parallel_scan(hashtable->batches[batchno].outer_tuples);
1515 : :
1516 : 640 : return true;
1517 : 1 : case PHJ_BATCH_SCAN:
1518 : :
1519 : : /*
1520 : : * In principle, we could help scan for unmatched tuples,
1521 : : * since that phase is already underway (the thing we
1522 : : * can't do under current deadlock-avoidance rules is wait
1523 : : * for others to arrive at PHJ_BATCH_SCAN, because
1524 : : * PHJ_BATCH_PROBE emits tuples, but in this case we just
1525 : : * got here without waiting). That is not yet done. For
1526 : : * now, we just detach and go around again. We have to
1527 : : * use ExecHashTableDetachBatch() because there's a small
1528 : : * chance we'll be the last to detach, and then we're
1529 : : * responsible for freeing memory.
1530 : : */
1531 : 1 : ExecParallelHashTableSetCurrentBatch(hashtable, batchno);
1532 : 1 : hashtable->batches[batchno].done = true;
1533 : 1 : ExecHashTableDetachBatch(hashtable);
1534 : 1 : break;
1535 : :
1536 : 551 : case PHJ_BATCH_FREE:
1537 : :
1538 : : /*
1539 : : * Already done. Detach and go around again (if any
1540 : : * remain).
1541 : : */
1542 : 551 : BarrierDetach(batch_barrier);
1543 : 551 : hashtable->batches[batchno].done = true;
1544 : 551 : hashtable->curbatch = -1;
1545 : 551 : break;
1546 : :
1547 : 0 : default:
1548 [ # # ]: 0 : elog(ERROR, "unexpected batch phase %d",
1549 : : BarrierPhase(batch_barrier));
1550 : : }
1551 : : }
1552 : 1593 : batchno = (batchno + 1) % hashtable->nbatch;
1553 [ + + ]: 1593 : } while (batchno != start_batchno);
1554 : :
1555 : 276 : return false;
1556 : : }
1557 : :
1558 : : /*
1559 : : * ExecHashJoinSaveTuple
1560 : : * save a tuple to a batch file.
1561 : : *
1562 : : * The data recorded in the file for each tuple is its hash value,
1563 : : * then the tuple in MinimalTuple format.
1564 : : *
1565 : : * fileptr points to a batch file in one of the hashtable arrays.
1566 : : *
1567 : : * The batch files (and their buffers) are allocated in the spill context
1568 : : * created for the hashtable.
1569 : : */
1570 : : void
1571 : 3244376 : ExecHashJoinSaveTuple(MinimalTuple tuple, uint32 hashvalue,
1572 : : BufFile **fileptr, HashJoinTable hashtable)
1573 : : {
1574 : 3244376 : BufFile *file = *fileptr;
1575 : :
1576 : : /*
1577 : : * The batch file is lazily created. If this is the first tuple written to
1578 : : * this batch, the batch file is created and its buffer is allocated in
1579 : : * the spillCxt context, NOT in the batchCxt.
1580 : : *
1581 : : * During the build phase, buffered files are created for inner batches.
1582 : : * Each batch's buffered file is closed (and its buffer freed) after the
1583 : : * batch is loaded into memory during the outer side scan. Therefore, it
1584 : : * is necessary to allocate the batch file buffer in a memory context
1585 : : * which outlives the batch itself.
1586 : : *
1587 : : * Also, we use spillCxt instead of hashCxt for a better accounting of the
1588 : : * spilling memory consumption.
1589 : : */
1590 [ + + ]: 3244376 : if (file == NULL)
1591 : : {
1592 : 1586 : MemoryContext oldctx = MemoryContextSwitchTo(hashtable->spillCxt);
1593 : :
1594 : 1586 : file = BufFileCreateTemp(false);
1595 : 1586 : *fileptr = file;
1596 : :
1597 : 1586 : MemoryContextSwitchTo(oldctx);
1598 : : }
1599 : :
1600 : 3244376 : BufFileWrite(file, &hashvalue, sizeof(uint32));
1601 : 3244376 : BufFileWrite(file, tuple, tuple->t_len);
1602 : 3244376 : }
1603 : :
1604 : : /*
1605 : : * ExecHashJoinGetSavedTuple
1606 : : * read the next tuple from a batch file. Return NULL if no more.
1607 : : *
1608 : : * On success, *hashvalue is set to the tuple's hash value, and the tuple
1609 : : * itself is stored in the given slot.
1610 : : */
1611 : : static TupleTableSlot *
1612 : 3245962 : ExecHashJoinGetSavedTuple(HashJoinState *hjstate,
1613 : : BufFile *file,
1614 : : uint32 *hashvalue,
1615 : : TupleTableSlot *tupleSlot)
1616 : : {
1617 : : uint32 header[2];
1618 : : size_t nread;
1619 : : MinimalTuple tuple;
1620 : :
1621 : : /*
1622 : : * We check for interrupts here because this is typically taken as an
1623 : : * alternative code path to an ExecProcNode() call, which would include
1624 : : * such a check.
1625 : : */
1626 [ + + ]: 3245962 : CHECK_FOR_INTERRUPTS();
1627 : :
1628 : : /*
1629 : : * Since both the hash value and the MinimalTuple length word are uint32,
1630 : : * we can read them both in one BufFileRead() call without any type
1631 : : * cheating.
1632 : : */
1633 : 3245962 : nread = BufFileReadMaybeEOF(file, header, sizeof(header), true);
1634 [ + + ]: 3245962 : if (nread == 0) /* end of file */
1635 : : {
1636 : 1586 : ExecClearTuple(tupleSlot);
1637 : 1586 : return NULL;
1638 : : }
1639 : 3244376 : *hashvalue = header[0];
1640 : 3244376 : tuple = (MinimalTuple) palloc(header[1]);
1641 : 3244376 : tuple->t_len = header[1];
1642 : 3244376 : BufFileReadExact(file,
1643 : : (char *) tuple + sizeof(uint32),
1644 : 3244376 : header[1] - sizeof(uint32));
1645 : 3244376 : ExecForceStoreMinimalTuple(tuple, tupleSlot, true);
1646 : 3244376 : return tupleSlot;
1647 : : }
1648 : :
1649 : :
1650 : : void
1651 : 2285 : ExecReScanHashJoin(HashJoinState *node)
1652 : : {
1653 : 2285 : PlanState *outerPlan = outerPlanState(node);
1654 : 2285 : PlanState *innerPlan = innerPlanState(node);
1655 : :
1656 : : /*
1657 : : * We're always going to rescan the outer rel, so drop the associated
1658 : : * null-keys tuplestore; we'll rebuild it during the rescan. (Must do
1659 : : * this before ExecHashTableDestroy deletes hashCxt.)
1660 : : */
1661 [ - + ]: 2285 : if (node->hj_NullOuterTupleStore)
1662 : : {
1663 : 0 : tuplestore_end(node->hj_NullOuterTupleStore);
1664 : 0 : node->hj_NullOuterTupleStore = NULL;
1665 : : }
1666 : :
1667 : : /*
1668 : : * In a multi-batch join, we currently have to do rescans the hard way,
1669 : : * primarily because batch temp files may have already been released. But
1670 : : * if it's a single-batch join, and there is no parameter change for the
1671 : : * inner subnode, then we can just re-use the existing hash table without
1672 : : * rebuilding it.
1673 : : */
1674 [ + + ]: 2285 : if (node->hj_HashTable != NULL)
1675 : : {
1676 : 1876 : HashState *hashNode = castNode(HashState, innerPlan);
1677 : :
1678 : : Assert(hashNode->hashtable == node->hj_HashTable);
1679 : :
1680 [ + - ]: 1876 : if (node->hj_HashTable->nbatch == 1 &&
1681 [ + + ]: 1876 : innerPlan->chgParam == NULL)
1682 : : {
1683 : : /*
1684 : : * Okay to reuse the hash table; needn't rescan inner, either.
1685 : : *
1686 : : * However, if it's a right/right-anti/right-semi/full join, we'd
1687 : : * better reset the inner-tuple match flags contained in the
1688 : : * table.
1689 : : */
1690 [ + + + + ]: 960 : if (HJ_FILL_INNER(node) || node->js.jointype == JOIN_RIGHT_SEMI)
1691 : 46 : ExecHashTableResetMatchFlags(node->hj_HashTable);
1692 : :
1693 : : /*
1694 : : * Also, we need to reset our state about the emptiness of the
1695 : : * outer relation, so that the new scan of the outer will update
1696 : : * it correctly if it turns out to be empty this time. (There's no
1697 : : * harm in clearing it now because ExecHashJoin won't need the
1698 : : * info. In the other cases, where the hash table doesn't exist
1699 : : * or we are destroying it, we leave this state alone because
1700 : : * ExecHashJoin will need it the first time through.)
1701 : : */
1702 : 960 : node->hj_OuterNotEmpty = false;
1703 : :
1704 : : /*
1705 : : * Also, rewind inner null-key tuplestore so that we can return
1706 : : * those tuples again.
1707 : : */
1708 [ + + ]: 960 : if (hashNode->null_tuple_store)
1709 : 4 : tuplestore_rescan(hashNode->null_tuple_store);
1710 : :
1711 : : /* ExecHashJoin can skip the BUILD_HASHTABLE step */
1712 : 960 : node->hj_JoinState = HJ_NEED_NEW_OUTER;
1713 : : }
1714 : : else
1715 : : {
1716 : : /* must destroy and rebuild hash table */
1717 : :
1718 : : /* accumulate stats from old hash table, if wanted */
1719 : : /* (this should match ExecShutdownHash) */
1720 [ - + - - ]: 916 : if (hashNode->ps.instrument && !hashNode->hinstrument)
1721 : 0 : hashNode->hinstrument = palloc0_object(HashInstrumentation);
1722 [ - + ]: 916 : if (hashNode->hinstrument)
1723 : 0 : ExecHashAccumInstrumentation(hashNode->hinstrument,
1724 : : hashNode->hashtable);
1725 : :
1726 : : /* free inner null-key tuplestore before ExecHashTableDestroy */
1727 [ - + ]: 916 : if (hashNode->null_tuple_store)
1728 : : {
1729 : 0 : tuplestore_end(hashNode->null_tuple_store);
1730 : 0 : hashNode->null_tuple_store = NULL;
1731 : : }
1732 : :
1733 : : /* for safety, be sure to clear child plan node's pointer too */
1734 : 916 : hashNode->hashtable = NULL;
1735 : :
1736 : 916 : ExecHashTableDestroy(node->hj_HashTable);
1737 : 916 : node->hj_HashTable = NULL;
1738 : 916 : node->hj_JoinState = HJ_BUILD_HASHTABLE;
1739 : :
1740 : : /*
1741 : : * if chgParam of subnode is not null then plan will be re-scanned
1742 : : * by first ExecProcNode.
1743 : : */
1744 [ - + ]: 916 : if (innerPlan->chgParam == NULL)
1745 : 0 : ExecReScan(innerPlan);
1746 : : }
1747 : : }
1748 : :
1749 : : /* Always reset intra-tuple state */
1750 : 2285 : node->hj_CurHashValue = 0;
1751 : 2285 : node->hj_CurBucketNo = 0;
1752 : 2285 : node->hj_CurSkewBucketNo = INVALID_SKEW_BUCKET_NO;
1753 : 2285 : node->hj_CurTuple = NULL;
1754 : :
1755 : 2285 : node->hj_MatchedOuter = false;
1756 : 2285 : node->hj_FirstOuterTupleSlot = NULL;
1757 : :
1758 : : /*
1759 : : * if chgParam of subnode is not null then plan will be re-scanned by
1760 : : * first ExecProcNode.
1761 : : */
1762 [ + + ]: 2285 : if (outerPlan->chgParam == NULL)
1763 : 1498 : ExecReScan(outerPlan);
1764 : 2285 : }
1765 : :
1766 : : void
1767 : 24875 : ExecShutdownHashJoin(HashJoinState *node)
1768 : : {
1769 [ + + ]: 24875 : if (node->hj_HashTable)
1770 : : {
1771 : : /*
1772 : : * Detach from shared state before DSM memory goes away. This makes
1773 : : * sure that we don't have any pointers into DSM memory by the time
1774 : : * ExecEndHashJoin runs.
1775 : : */
1776 : 18076 : ExecHashTableDetachBatch(node->hj_HashTable);
1777 : 18076 : ExecHashTableDetach(node->hj_HashTable);
1778 : : }
1779 : 24875 : }
1780 : :
1781 : : static void
1782 : 121 : ExecParallelHashJoinPartitionOuter(HashJoinState *hjstate)
1783 : : {
1784 : 121 : PlanState *outerState = outerPlanState(hjstate);
1785 : 121 : ExprContext *econtext = hjstate->js.ps.ps_ExprContext;
1786 : 121 : HashJoinTable hashtable = hjstate->hj_HashTable;
1787 : : TupleTableSlot *slot;
1788 : : int i;
1789 : :
1790 : : Assert(hjstate->hj_FirstOuterTupleSlot == NULL);
1791 : :
1792 : : /* Execute outer plan, writing all tuples to shared tuplestores. */
1793 : : for (;;)
1794 : 800052 : {
1795 : : bool isnull;
1796 : : uint32 hashvalue;
1797 : :
1798 : 800173 : slot = ExecProcNode(outerState);
1799 [ + + + + ]: 800173 : if (TupIsNull(slot))
1800 : : break;
1801 : 800052 : econtext->ecxt_outertuple = slot;
1802 : :
1803 : 800052 : ResetExprContext(econtext);
1804 : :
1805 : 800052 : hashvalue = DatumGetUInt32(ExecEvalExprSwitchContext(hjstate->hj_OuterHash,
1806 : : econtext,
1807 : : &isnull));
1808 : :
1809 [ + + ]: 800052 : if (!isnull)
1810 : : {
1811 : : /* normal case with a non-null join key */
1812 : : int batchno;
1813 : : int bucketno;
1814 : : bool shouldFree;
1815 : 800024 : MinimalTuple mintup = ExecFetchSlotMinimalTuple(slot, &shouldFree);
1816 : :
1817 : 800024 : ExecHashGetBucketAndBatch(hashtable, hashvalue, &bucketno,
1818 : : &batchno);
1819 : 800024 : sts_puttuple(hashtable->batches[batchno].outer_tuples,
1820 : : &hashvalue, mintup);
1821 : :
1822 [ + - ]: 800024 : if (shouldFree)
1823 : 800024 : heap_free_minimal_tuple(mintup);
1824 : : }
1825 [ + + ]: 28 : else if (hjstate->hj_KeepNullTuples)
1826 : : {
1827 : : /* null join key, but we must save tuple to be emitted later */
1828 [ + - ]: 4 : if (hjstate->hj_NullOuterTupleStore == NULL)
1829 : 4 : hjstate->hj_NullOuterTupleStore = ExecHashBuildNullTupleStore(hashtable);
1830 : 4 : tuplestore_puttupleslot(hjstate->hj_NullOuterTupleStore, slot);
1831 : : }
1832 : : /* else we can just discard the tuple immediately */
1833 : :
1834 [ - + ]: 800052 : CHECK_FOR_INTERRUPTS();
1835 : : }
1836 : :
1837 : : /* Make sure all outer partitions are readable by any backend. */
1838 [ + + ]: 1025 : for (i = 0; i < hashtable->nbatch; ++i)
1839 : 904 : sts_end_write(hashtable->batches[i].outer_tuples);
1840 : 121 : }
1841 : :
1842 : : void
1843 : 84 : ExecHashJoinEstimate(HashJoinState *state, ParallelContext *pcxt)
1844 : : {
1845 : 84 : shm_toc_estimate_chunk(&pcxt->estimator, sizeof(ParallelHashJoinState));
1846 : 84 : shm_toc_estimate_keys(&pcxt->estimator, 1);
1847 : 84 : }
1848 : :
1849 : : void
1850 : 84 : ExecHashJoinInitializeDSM(HashJoinState *state, ParallelContext *pcxt)
1851 : : {
1852 : 84 : int plan_node_id = state->js.ps.plan->plan_node_id;
1853 : : HashState *hashNode;
1854 : : ParallelHashJoinState *pstate;
1855 : :
1856 : : /*
1857 : : * Disable shared hash table mode if we failed to create a real DSM
1858 : : * segment, because that means that we don't have a DSA area to work with.
1859 : : */
1860 [ - + ]: 84 : if (pcxt->seg == NULL)
1861 : 0 : return;
1862 : :
1863 : 84 : ExecSetExecProcNode(&state->js.ps, ExecParallelHashJoin);
1864 : :
1865 : : /*
1866 : : * Set up the state needed to coordinate access to the shared hash
1867 : : * table(s), using the plan node ID as the toc key.
1868 : : */
1869 : 84 : pstate = shm_toc_allocate(pcxt->toc, sizeof(ParallelHashJoinState));
1870 : 84 : shm_toc_insert(pcxt->toc, plan_node_id, pstate);
1871 : :
1872 : : /*
1873 : : * Set up the shared hash join state with no batches initially.
1874 : : * ExecHashTableCreate() will prepare at least one later and set nbatch
1875 : : * and space_allowed.
1876 : : */
1877 : 84 : pstate->nbatch = 0;
1878 : 84 : pstate->space_allowed = 0;
1879 : 84 : pstate->batches = InvalidDsaPointer;
1880 : 84 : pstate->old_batches = InvalidDsaPointer;
1881 : 84 : pstate->nbuckets = 0;
1882 : 84 : pstate->growth = PHJ_GROWTH_OK;
1883 : 84 : pstate->chunk_work_queue = InvalidDsaPointer;
1884 : 84 : pg_atomic_init_u32(&pstate->distributor, 0);
1885 : 84 : pstate->nparticipants = pcxt->nworkers + 1;
1886 : 84 : pstate->total_tuples = 0;
1887 : 84 : LWLockInitialize(&pstate->lock,
1888 : : LWTRANCHE_PARALLEL_HASH_JOIN);
1889 : 84 : BarrierInit(&pstate->build_barrier, 0);
1890 : 84 : BarrierInit(&pstate->grow_batches_barrier, 0);
1891 : 84 : BarrierInit(&pstate->grow_buckets_barrier, 0);
1892 : :
1893 : : /* Set up the space we'll use for shared temporary files. */
1894 : 84 : SharedFileSetInit(&pstate->fileset, pcxt->seg);
1895 : :
1896 : : /* Initialize the shared state in the hash node. */
1897 : 84 : hashNode = (HashState *) innerPlanState(state);
1898 : 84 : hashNode->parallel_state = pstate;
1899 : : }
1900 : :
1901 : : /* ----------------------------------------------------------------
1902 : : * ExecHashJoinReInitializeDSM
1903 : : *
1904 : : * Reset shared state before beginning a fresh scan.
1905 : : * ----------------------------------------------------------------
1906 : : */
1907 : : void
1908 : 32 : ExecHashJoinReInitializeDSM(HashJoinState *state, ParallelContext *pcxt)
1909 : : {
1910 : 32 : int plan_node_id = state->js.ps.plan->plan_node_id;
1911 : : ParallelHashJoinState *pstate;
1912 : : HashState *hashNode;
1913 : :
1914 : : /* Nothing to do if we failed to create a DSM segment. */
1915 [ - + ]: 32 : if (pcxt->seg == NULL)
1916 : 0 : return;
1917 : :
1918 : 32 : pstate = shm_toc_lookup(pcxt->toc, plan_node_id, false);
1919 : :
1920 : : /*
1921 : : * It would be possible to reuse the shared hash table in single-batch
1922 : : * cases by resetting and then fast-forwarding build_barrier to
1923 : : * PHJ_BUILD_FREE and batch 0's batch_barrier to PHJ_BATCH_PROBE, but
1924 : : * currently shared hash tables are already freed by now (by the last
1925 : : * participant to detach from the batch). We could consider keeping it
1926 : : * around for single-batch joins. We'd also need to adjust
1927 : : * finalize_plan() so that it doesn't record a dummy dependency for
1928 : : * Parallel Hash nodes, preventing the rescan optimization. For now we
1929 : : * don't try.
1930 : : */
1931 : :
1932 : : /* Detach, freeing any remaining shared memory. */
1933 [ - + ]: 32 : if (state->hj_HashTable != NULL)
1934 : : {
1935 : 0 : ExecHashTableDetachBatch(state->hj_HashTable);
1936 : 0 : ExecHashTableDetach(state->hj_HashTable);
1937 : : }
1938 : :
1939 : : /* Clear any shared batch files. */
1940 : 32 : SharedFileSetDeleteAll(&pstate->fileset);
1941 : :
1942 : : /* We'd better clear our local null-key tuplestores, too. */
1943 [ - + ]: 32 : if (state->hj_NullOuterTupleStore)
1944 : : {
1945 : 0 : tuplestore_end(state->hj_NullOuterTupleStore);
1946 : 0 : state->hj_NullOuterTupleStore = NULL;
1947 : : }
1948 : 32 : hashNode = (HashState *) innerPlanState(state);
1949 [ - + ]: 32 : if (hashNode->null_tuple_store)
1950 : : {
1951 : 0 : tuplestore_end(hashNode->null_tuple_store);
1952 : 0 : hashNode->null_tuple_store = NULL;
1953 : : }
1954 : :
1955 : :
1956 : : /* Reset build_barrier to PHJ_BUILD_ELECT so we can go around again. */
1957 : 32 : BarrierInit(&pstate->build_barrier, 0);
1958 : : }
1959 : :
1960 : : void
1961 : 212 : ExecHashJoinInitializeWorker(HashJoinState *state,
1962 : : ParallelWorkerContext *pwcxt)
1963 : : {
1964 : : HashState *hashNode;
1965 : 212 : int plan_node_id = state->js.ps.plan->plan_node_id;
1966 : : ParallelHashJoinState *pstate =
1967 : 212 : shm_toc_lookup(pwcxt->toc, plan_node_id, false);
1968 : :
1969 : : /* Attach to the space for shared temporary files. */
1970 : 212 : SharedFileSetAttach(&pstate->fileset, pwcxt->seg);
1971 : :
1972 : : /* Attach to the shared state in the hash node. */
1973 : 212 : hashNode = (HashState *) innerPlanState(state);
1974 : 212 : hashNode->parallel_state = pstate;
1975 : :
1976 : 212 : ExecSetExecProcNode(&state->js.ps, ExecParallelHashJoin);
1977 : 212 : }
|
