Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * nodeMemoize.c
4 : : * Routines to handle caching of results from parameterized nodes
5 : : *
6 : : * Portions Copyright (c) 2021-2025, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/executor/nodeMemoize.c
12 : : *
13 : : * Memoize nodes are intended to sit above parameterized nodes in the plan
14 : : * tree in order to cache results from them. The intention here is that a
15 : : * repeat scan with a parameter value that has already been seen by the node
16 : : * can fetch tuples from the cache rather than having to re-scan the inner
17 : : * node all over again. The query planner may choose to make use of one of
18 : : * these when it thinks rescans for previously seen values are likely enough
19 : : * to warrant adding the additional node.
20 : : *
21 : : * The method of cache we use is a hash table. When the cache fills, we never
22 : : * spill tuples to disk, instead, we choose to evict the least recently used
23 : : * cache entry from the cache. We remember the least recently used entry by
24 : : * always pushing new entries and entries we look for onto the tail of a
25 : : * doubly linked list. This means that older items always bubble to the top
26 : : * of this LRU list.
27 : : *
28 : : * Sometimes our callers won't run their scans to completion. For example a
29 : : * semi-join only needs to run until it finds a matching tuple, and once it
30 : : * does, the join operator skips to the next outer tuple and does not execute
31 : : * the inner side again on that scan. Because of this, we must keep track of
32 : : * when a cache entry is complete, and by default, we know it is when we run
33 : : * out of tuples to read during the scan. However, there are cases where we
34 : : * can mark the cache entry as complete without exhausting the scan of all
35 : : * tuples. One case is unique joins, where the join operator knows that there
36 : : * will only be at most one match for any given outer tuple. In order to
37 : : * support such cases we allow the "singlerow" option to be set for the cache.
38 : : * This option marks the cache entry as complete after we read the first tuple
39 : : * from the subnode.
40 : : *
41 : : * It's possible when we're filling the cache for a given set of parameters
42 : : * that we're unable to free enough memory to store any more tuples. If this
43 : : * happens then we'll have already evicted all other cache entries. When
44 : : * caching another tuple would cause us to exceed our memory budget, we must
45 : : * free the entry that we're currently populating and move the state machine
46 : : * into MEMO_CACHE_BYPASS_MODE. This means that we'll not attempt to cache
47 : : * any further tuples for this particular scan. We don't have the memory for
48 : : * it. The state machine will be reset again on the next rescan. If the
49 : : * memory requirements to cache the next parameter's tuples are less
50 : : * demanding, then that may allow us to start putting useful entries back into
51 : : * the cache again.
52 : : *
53 : : *
54 : : * INTERFACE ROUTINES
55 : : * ExecMemoize - lookup cache, exec subplan when not found
56 : : * ExecInitMemoize - initialize node and subnodes
57 : : * ExecEndMemoize - shutdown node and subnodes
58 : : * ExecReScanMemoize - rescan the memoize node
59 : : *
60 : : * ExecMemoizeEstimate estimates DSM space needed for parallel plan
61 : : * ExecMemoizeInitializeDSM initialize DSM for parallel plan
62 : : * ExecMemoizeInitializeWorker attach to DSM info in parallel worker
63 : : * ExecMemoizeRetrieveInstrumentation get instrumentation from worker
64 : : *-------------------------------------------------------------------------
65 : : */
66 : :
67 : : #include "postgres.h"
68 : :
69 : : #include "access/htup_details.h"
70 : : #include "common/hashfn.h"
71 : : #include "executor/executor.h"
72 : : #include "executor/nodeMemoize.h"
73 : : #include "lib/ilist.h"
74 : : #include "miscadmin.h"
75 : : #include "utils/datum.h"
76 : : #include "utils/lsyscache.h"
77 : :
78 : : /* States of the ExecMemoize state machine */
79 : : #define MEMO_CACHE_LOOKUP 1 /* Attempt to perform a cache lookup */
80 : : #define MEMO_CACHE_FETCH_NEXT_TUPLE 2 /* Get another tuple from the cache */
81 : : #define MEMO_FILLING_CACHE 3 /* Read outer node to fill cache */
82 : : #define MEMO_CACHE_BYPASS_MODE 4 /* Bypass mode. Just read from our
83 : : * subplan without caching anything */
84 : : #define MEMO_END_OF_SCAN 5 /* Ready for rescan */
85 : :
86 : :
87 : : /* Helper macros for memory accounting */
88 : : #define EMPTY_ENTRY_MEMORY_BYTES(e) (sizeof(MemoizeEntry) + \
89 : : sizeof(MemoizeKey) + \
90 : : (e)->key->params->t_len);
91 : : #define CACHE_TUPLE_BYTES(t) (sizeof(MemoizeTuple) + \
92 : : (t)->mintuple->t_len)
93 : :
94 : : /* MemoizeTuple Stores an individually cached tuple */
95 : : typedef struct MemoizeTuple
96 : : {
97 : : MinimalTuple mintuple; /* Cached tuple */
98 : : struct MemoizeTuple *next; /* The next tuple with the same parameter
99 : : * values or NULL if it's the last one */
100 : : } MemoizeTuple;
101 : :
102 : : /*
103 : : * MemoizeKey
104 : : * The hash table key for cached entries plus the LRU list link
105 : : */
106 : : typedef struct MemoizeKey
107 : : {
108 : : MinimalTuple params;
109 : : dlist_node lru_node; /* Pointer to next/prev key in LRU list */
110 : : } MemoizeKey;
111 : :
112 : : /*
113 : : * MemoizeEntry
114 : : * The data struct that the cache hash table stores
115 : : */
116 : : typedef struct MemoizeEntry
117 : : {
118 : : MemoizeKey *key; /* Hash key for hash table lookups */
119 : : MemoizeTuple *tuplehead; /* Pointer to the first tuple or NULL if no
120 : : * tuples are cached for this entry */
121 : : uint32 hash; /* Hash value (cached) */
122 : : char status; /* Hash status */
123 : : bool complete; /* Did we read the outer plan to completion? */
124 : : } MemoizeEntry;
125 : :
126 : :
127 : : #define SH_PREFIX memoize
128 : : #define SH_ELEMENT_TYPE MemoizeEntry
129 : : #define SH_KEY_TYPE MemoizeKey *
130 : : #define SH_SCOPE static inline
131 : : #define SH_DECLARE
132 : : #include "lib/simplehash.h"
133 : :
134 : : static uint32 MemoizeHash_hash(struct memoize_hash *tb,
135 : : const MemoizeKey *key);
136 : : static bool MemoizeHash_equal(struct memoize_hash *tb,
137 : : const MemoizeKey *key1,
138 : : const MemoizeKey *key2);
139 : :
140 : : #define SH_PREFIX memoize
141 : : #define SH_ELEMENT_TYPE MemoizeEntry
142 : : #define SH_KEY_TYPE MemoizeKey *
143 : : #define SH_KEY key
144 : : #define SH_HASH_KEY(tb, key) MemoizeHash_hash(tb, key)
145 : : #define SH_EQUAL(tb, a, b) MemoizeHash_equal(tb, a, b)
146 : : #define SH_SCOPE static inline
147 : : #define SH_STORE_HASH
148 : : #define SH_GET_HASH(tb, a) a->hash
149 : : #define SH_DEFINE
150 : : #include "lib/simplehash.h"
151 : :
152 : : /*
153 : : * MemoizeHash_hash
154 : : * Hash function for simplehash hashtable. 'key' is unused here as we
155 : : * require that all table lookups first populate the MemoizeState's
156 : : * probeslot with the key values to be looked up.
157 : : */
158 : : static uint32
1566 drowley@postgresql.o 159 :CBC 378687 : MemoizeHash_hash(struct memoize_hash *tb, const MemoizeKey *key)
160 : : {
161 : 378687 : MemoizeState *mstate = (MemoizeState *) tb->private_data;
753 162 : 378687 : ExprContext *econtext = mstate->ss.ps.ps_ExprContext;
163 : : MemoryContext oldcontext;
1566 164 : 378687 : TupleTableSlot *pslot = mstate->probeslot;
1669 165 : 378687 : uint32 hashkey = 0;
1566 166 : 378687 : int numkeys = mstate->nkeys;
167 : :
753 168 : 378687 : oldcontext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory);
169 : :
1433 170 [ + + ]: 378687 : if (mstate->binary_mode)
171 : : {
172 [ + + ]: 111785 : for (int i = 0; i < numkeys; i++)
173 : : {
174 : : /* combine successive hashkeys by rotating */
1345 john.naylor@postgres 175 : 60362 : hashkey = pg_rotate_left32(hashkey, 1);
176 : :
1433 drowley@postgresql.o 177 [ + - ]: 60362 : if (!pslot->tts_isnull[i]) /* treat nulls as having hash key 0 */
178 : : {
179 : : CompactAttribute *attr;
180 : : uint32 hkey;
181 : :
311 182 : 60362 : attr = TupleDescCompactAttr(pslot->tts_tupleDescriptor, i);
183 : :
1433 184 : 60362 : hkey = datum_image_hash(pslot->tts_values[i], attr->attbyval, attr->attlen);
185 : :
186 : 60362 : hashkey ^= hkey;
187 : : }
188 : : }
189 : : }
190 : : else
191 : : {
192 : 327264 : FmgrInfo *hashfunctions = mstate->hashfunctions;
193 : 327264 : Oid *collations = mstate->collations;
194 : :
195 [ + + ]: 655023 : for (int i = 0; i < numkeys; i++)
196 : : {
197 : : /* combine successive hashkeys by rotating */
1345 john.naylor@postgres 198 : 327759 : hashkey = pg_rotate_left32(hashkey, 1);
199 : :
1433 drowley@postgresql.o 200 [ + + ]: 327759 : if (!pslot->tts_isnull[i]) /* treat nulls as having hash key 0 */
201 : : {
202 : : uint32 hkey;
203 : :
204 : 327284 : hkey = DatumGetUInt32(FunctionCall1Coll(&hashfunctions[i],
205 : 327284 : collations[i], pslot->tts_values[i]));
206 : 327284 : hashkey ^= hkey;
207 : : }
208 : : }
209 : : }
210 : :
753 211 : 378687 : MemoryContextSwitchTo(oldcontext);
1669 212 : 378687 : return murmurhash32(hashkey);
213 : : }
214 : :
215 : : /*
216 : : * MemoizeHash_equal
217 : : * Equality function for confirming hash value matches during a hash
218 : : * table lookup. 'key2' is never used. Instead the MemoizeState's
219 : : * probeslot is always populated with details of what's being looked up.
220 : : */
221 : : static bool
1566 222 : 330927 : MemoizeHash_equal(struct memoize_hash *tb, const MemoizeKey *key1,
223 : : const MemoizeKey *key2)
224 : : {
225 : 330927 : MemoizeState *mstate = (MemoizeState *) tb->private_data;
226 : 330927 : ExprContext *econtext = mstate->ss.ps.ps_ExprContext;
227 : 330927 : TupleTableSlot *tslot = mstate->tableslot;
228 : 330927 : TupleTableSlot *pslot = mstate->probeslot;
229 : :
230 : : /* probeslot should have already been prepared by prepare_probe_slot() */
1669 231 : 330927 : ExecStoreMinimalTuple(key1->params, tslot, false);
232 : :
1433 233 [ + + ]: 330927 : if (mstate->binary_mode)
234 : : {
235 : : MemoryContext oldcontext;
236 : 50999 : int numkeys = mstate->nkeys;
753 237 : 50999 : bool match = true;
238 : :
239 : 50999 : oldcontext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory);
240 : :
1433 241 : 50999 : slot_getallattrs(tslot);
242 : 50999 : slot_getallattrs(pslot);
243 : :
244 [ + + ]: 110769 : for (int i = 0; i < numkeys; i++)
245 : : {
246 : : CompactAttribute *attr;
247 : :
248 [ - + ]: 59770 : if (tslot->tts_isnull[i] != pslot->tts_isnull[i])
249 : : {
753 drowley@postgresql.o 250 :UBC 0 : match = false;
251 : 0 : break;
252 : : }
253 : :
254 : : /* both NULL? they're equal */
1433 drowley@postgresql.o 255 [ - + ]:CBC 59770 : if (tslot->tts_isnull[i])
1433 drowley@postgresql.o 256 :UBC 0 : continue;
257 : :
258 : : /* perform binary comparison on the two datums */
311 drowley@postgresql.o 259 :CBC 59770 : attr = TupleDescCompactAttr(tslot->tts_tupleDescriptor, i);
1433 260 [ - + ]: 59770 : if (!datum_image_eq(tslot->tts_values[i], pslot->tts_values[i],
261 : 59770 : attr->attbyval, attr->attlen))
262 : : {
753 drowley@postgresql.o 263 :UBC 0 : match = false;
264 : 0 : break;
265 : : }
266 : : }
267 : :
753 drowley@postgresql.o 268 :CBC 50999 : MemoryContextSwitchTo(oldcontext);
269 : 50999 : return match;
270 : : }
271 : : else
272 : : {
1433 273 : 279928 : econtext->ecxt_innertuple = tslot;
274 : 279928 : econtext->ecxt_outertuple = pslot;
595 275 : 279928 : return ExecQual(mstate->cache_eq_expr, econtext);
276 : : }
277 : : }
278 : :
279 : : /*
280 : : * Initialize the hash table to empty. The MemoizeState's hashtable field
281 : : * must point to NULL.
282 : : */
283 : : static void
1566 284 : 833 : build_hash_table(MemoizeState *mstate, uint32 size)
285 : : {
636 286 [ - + ]: 833 : Assert(mstate->hashtable == NULL);
287 : :
288 : : /* Make a guess at a good size when we're not given a valid size. */
1669 289 [ - + ]: 833 : if (size == 0)
1669 drowley@postgresql.o 290 :UBC 0 : size = 1024;
291 : :
292 : : /* memoize_create will convert the size to a power of 2 */
1566 drowley@postgresql.o 293 :CBC 833 : mstate->hashtable = memoize_create(mstate->tableContext, size, mstate);
1669 294 : 833 : }
295 : :
296 : : /*
297 : : * prepare_probe_slot
298 : : * Populate mstate's probeslot with the values from the tuple stored
299 : : * in 'key'. If 'key' is NULL, then perform the population by evaluating
300 : : * mstate's param_exprs.
301 : : */
302 : : static inline void
1566 303 : 378687 : prepare_probe_slot(MemoizeState *mstate, MemoizeKey *key)
304 : : {
305 : 378687 : TupleTableSlot *pslot = mstate->probeslot;
306 : 378687 : TupleTableSlot *tslot = mstate->tableslot;
307 : 378687 : int numKeys = mstate->nkeys;
308 : :
1669 309 : 378687 : ExecClearTuple(pslot);
310 : :
311 [ + + ]: 378687 : if (key == NULL)
312 : : {
952 313 : 377487 : ExprContext *econtext = mstate->ss.ps.ps_ExprContext;
314 : : MemoryContext oldcontext;
315 : :
316 : 377487 : oldcontext = MemoryContextSwitchTo(econtext->ecxt_per_tuple_memory);
317 : :
318 : : /* Set the probeslot's values based on the current parameter values */
1669 319 [ + + ]: 764408 : for (int i = 0; i < numKeys; i++)
1566 320 : 386921 : pslot->tts_values[i] = ExecEvalExpr(mstate->param_exprs[i],
321 : : econtext,
1669 322 : 386921 : &pslot->tts_isnull[i]);
323 : :
952 324 : 377487 : MemoryContextSwitchTo(oldcontext);
325 : : }
326 : : else
327 : : {
328 : : /* Process the key's MinimalTuple and store the values in probeslot */
1669 329 : 1200 : ExecStoreMinimalTuple(key->params, tslot, false);
330 : 1200 : slot_getallattrs(tslot);
331 : 1200 : memcpy(pslot->tts_values, tslot->tts_values, sizeof(Datum) * numKeys);
332 : 1200 : memcpy(pslot->tts_isnull, tslot->tts_isnull, sizeof(bool) * numKeys);
333 : : }
334 : :
335 : 378687 : ExecStoreVirtualTuple(pslot);
336 : 378687 : }
337 : :
338 : : /*
339 : : * entry_purge_tuples
340 : : * Remove all tuples from the cache entry pointed to by 'entry'. This
341 : : * leaves an empty cache entry. Also, update the memory accounting to
342 : : * reflect the removal of the tuples.
343 : : */
344 : : static inline void
1566 345 : 1194 : entry_purge_tuples(MemoizeState *mstate, MemoizeEntry *entry)
346 : : {
347 : 1194 : MemoizeTuple *tuple = entry->tuplehead;
1669 348 : 1194 : uint64 freed_mem = 0;
349 : :
350 [ + + ]: 2388 : while (tuple != NULL)
351 : : {
1566 352 : 1194 : MemoizeTuple *next = tuple->next;
353 : :
1669 354 : 1194 : freed_mem += CACHE_TUPLE_BYTES(tuple);
355 : :
356 : : /* Free memory used for this tuple */
357 : 1194 : pfree(tuple->mintuple);
358 : 1194 : pfree(tuple);
359 : :
360 : 1194 : tuple = next;
361 : : }
362 : :
363 : 1194 : entry->complete = false;
364 : 1194 : entry->tuplehead = NULL;
365 : :
366 : : /* Update the memory accounting */
1566 367 : 1194 : mstate->mem_used -= freed_mem;
1669 368 : 1194 : }
369 : :
370 : : /*
371 : : * remove_cache_entry
372 : : * Remove 'entry' from the cache and free memory used by it.
373 : : */
374 : : static void
1566 375 : 1194 : remove_cache_entry(MemoizeState *mstate, MemoizeEntry *entry)
376 : : {
377 : 1194 : MemoizeKey *key = entry->key;
378 : :
1669 379 : 1194 : dlist_delete(&entry->key->lru_node);
380 : :
381 : : /* Remove all of the tuples from this entry */
1566 382 : 1194 : entry_purge_tuples(mstate, entry);
383 : :
384 : : /*
385 : : * Update memory accounting. entry_purge_tuples should have already
386 : : * subtracted the memory used for each cached tuple. Here we just update
387 : : * the amount used by the entry itself.
388 : : */
389 : 1194 : mstate->mem_used -= EMPTY_ENTRY_MEMORY_BYTES(entry);
390 : :
391 : : /* Remove the entry from the cache */
392 : 1194 : memoize_delete_item(mstate->hashtable, entry);
393 : :
1669 394 : 1194 : pfree(key->params);
395 : 1194 : pfree(key);
396 : 1194 : }
397 : :
398 : : /*
399 : : * cache_purge_all
400 : : * Remove all items from the cache
401 : : */
402 : : static void
1433 403 : 9 : cache_purge_all(MemoizeState *mstate)
404 : : {
636 405 : 9 : uint64 evictions = 0;
406 : :
407 [ + + ]: 9 : if (mstate->hashtable != NULL)
408 : 6 : evictions = mstate->hashtable->members;
409 : :
410 : : /*
411 : : * Likely the most efficient way to remove all items is to just reset the
412 : : * memory context for the cache and then rebuild a fresh hash table. This
413 : : * saves having to remove each item one by one and pfree each cached tuple
414 : : */
1433 415 : 9 : MemoryContextReset(mstate->tableContext);
416 : :
417 : : /* NULLify so we recreate the table on the next call */
636 418 : 9 : mstate->hashtable = NULL;
419 : :
420 : : /* reset the LRU list */
1433 421 : 9 : dlist_init(&mstate->lru_list);
422 : 9 : mstate->last_tuple = NULL;
423 : 9 : mstate->entry = NULL;
424 : :
425 : 9 : mstate->mem_used = 0;
426 : :
427 : : /* XXX should we add something new to track these purges? */
428 : 9 : mstate->stats.cache_evictions += evictions; /* Update Stats */
429 : 9 : }
430 : :
431 : : /*
432 : : * cache_reduce_memory
433 : : * Evict older and less recently used items from the cache in order to
434 : : * reduce the memory consumption back to something below the
435 : : * MemoizeState's mem_limit.
436 : : *
437 : : * 'specialkey', if not NULL, causes the function to return false if the entry
438 : : * which the key belongs to is removed from the cache.
439 : : */
440 : : static bool
1566 441 : 1194 : cache_reduce_memory(MemoizeState *mstate, MemoizeKey *specialkey)
442 : : {
1669 443 : 1194 : bool specialkey_intact = true; /* for now */
444 : : dlist_mutable_iter iter;
445 : 1194 : uint64 evictions = 0;
446 : :
447 : : /* Update peak memory usage */
1566 448 [ + + ]: 1194 : if (mstate->mem_used > mstate->stats.mem_peak)
449 : 3 : mstate->stats.mem_peak = mstate->mem_used;
450 : :
451 : : /* We expect only to be called when we've gone over budget on memory */
452 [ - + ]: 1194 : Assert(mstate->mem_used > mstate->mem_limit);
453 : :
454 : : /* Start the eviction process starting at the head of the LRU list. */
455 [ + - + - ]: 1194 : dlist_foreach_modify(iter, &mstate->lru_list)
456 : : {
457 : 1194 : MemoizeKey *key = dlist_container(MemoizeKey, lru_node, iter.cur);
458 : : MemoizeEntry *entry;
459 : :
460 : : /*
461 : : * Populate the hash probe slot in preparation for looking up this LRU
462 : : * entry.
463 : : */
464 : 1194 : prepare_probe_slot(mstate, key);
465 : :
466 : : /*
467 : : * Ideally the LRU list pointers would be stored in the entry itself
468 : : * rather than in the key. Unfortunately, we can't do that as the
469 : : * simplehash.h code may resize the table and allocate new memory for
470 : : * entries which would result in those pointers pointing to the old
471 : : * buckets. However, it's fine to use the key to store this as that's
472 : : * only referenced by a pointer in the entry, which of course follows
473 : : * the entry whenever the hash table is resized. Since we only have a
474 : : * pointer to the key here, we must perform a hash table lookup to
475 : : * find the entry that the key belongs to.
476 : : */
477 : 1194 : entry = memoize_lookup(mstate->hashtable, NULL);
478 : :
479 : : /*
480 : : * Sanity check that we found the entry belonging to the LRU list
481 : : * item. A misbehaving hash or equality function could cause the
482 : : * entry not to be found or the wrong entry to be found.
483 : : */
1237 484 [ + - - + : 1194 : if (unlikely(entry == NULL || entry->key != key))
- + ]
1237 drowley@postgresql.o 485 [ # # ]:UBC 0 : elog(ERROR, "could not find memoization table entry");
486 : :
487 : : /*
488 : : * If we're being called to free memory while the cache is being
489 : : * populated with new tuples, then we'd better take some care as we
490 : : * could end up freeing the entry which 'specialkey' belongs to.
491 : : * Generally callers will pass 'specialkey' as the key for the cache
492 : : * entry which is currently being populated, so we must set
493 : : * 'specialkey_intact' to false to inform the caller the specialkey
494 : : * entry has been removed.
495 : : */
1669 drowley@postgresql.o 496 [ - + ]:CBC 1194 : if (key == specialkey)
1669 drowley@postgresql.o 497 :UBC 0 : specialkey_intact = false;
498 : :
499 : : /*
500 : : * Finally remove the entry. This will remove from the LRU list too.
501 : : */
1566 drowley@postgresql.o 502 :CBC 1194 : remove_cache_entry(mstate, entry);
503 : :
1669 504 : 1194 : evictions++;
505 : :
506 : : /* Exit if we've freed enough memory */
1566 507 [ + - ]: 1194 : if (mstate->mem_used <= mstate->mem_limit)
1669 508 : 1194 : break;
509 : : }
510 : :
1566 511 : 1194 : mstate->stats.cache_evictions += evictions; /* Update Stats */
512 : :
1669 513 : 1194 : return specialkey_intact;
514 : : }
515 : :
516 : : /*
517 : : * cache_lookup
518 : : * Perform a lookup to see if we've already cached tuples based on the
519 : : * scan's current parameters. If we find an existing entry we move it to
520 : : * the end of the LRU list, set *found to true then return it. If we
521 : : * don't find an entry then we create a new one and add it to the end of
522 : : * the LRU list. We also update cache memory accounting and remove older
523 : : * entries if we go over the memory budget. If we managed to free enough
524 : : * memory we return the new entry, else we return NULL.
525 : : *
526 : : * Callers can assume we'll never return NULL when *found is true.
527 : : */
528 : : static MemoizeEntry *
1566 529 : 377487 : cache_lookup(MemoizeState *mstate, bool *found)
530 : : {
531 : : MemoizeKey *key;
532 : : MemoizeEntry *entry;
533 : : MemoryContext oldcontext;
534 : :
535 : : /* prepare the probe slot with the current scan parameters */
536 : 377487 : prepare_probe_slot(mstate, NULL);
537 : :
538 : : /*
539 : : * Add the new entry to the cache. No need to pass a valid key since the
540 : : * hash function uses mstate's probeslot, which we populated above.
541 : : */
542 : 377487 : entry = memoize_insert(mstate->hashtable, NULL, found);
543 : :
1669 544 [ + + ]: 377487 : if (*found)
545 : : {
546 : : /*
547 : : * Move existing entry to the tail of the LRU list to mark it as the
548 : : * most recently used item.
549 : : */
1566 550 : 329727 : dlist_move_tail(&mstate->lru_list, &entry->key->lru_node);
551 : :
1669 552 : 329727 : return entry;
553 : : }
554 : :
1566 555 : 47760 : oldcontext = MemoryContextSwitchTo(mstate->tableContext);
556 : :
557 : : /* Allocate a new key */
558 : 47760 : entry->key = key = (MemoizeKey *) palloc(sizeof(MemoizeKey));
559 : 47760 : key->params = ExecCopySlotMinimalTuple(mstate->probeslot);
560 : :
561 : : /* Update the total cache memory utilization */
562 : 47760 : mstate->mem_used += EMPTY_ENTRY_MEMORY_BYTES(entry);
563 : :
564 : : /* Initialize this entry */
1669 565 : 47760 : entry->complete = false;
566 : 47760 : entry->tuplehead = NULL;
567 : :
568 : : /*
569 : : * Since this is the most recently used entry, push this entry onto the
570 : : * end of the LRU list.
571 : : */
1566 572 : 47760 : dlist_push_tail(&mstate->lru_list, &entry->key->lru_node);
573 : :
574 : 47760 : mstate->last_tuple = NULL;
575 : :
1669 576 : 47760 : MemoryContextSwitchTo(oldcontext);
577 : :
578 : : /*
579 : : * If we've gone over our memory budget, then we'll free up some space in
580 : : * the cache.
581 : : */
1566 582 [ + + ]: 47760 : if (mstate->mem_used > mstate->mem_limit)
583 : : {
584 : : /*
585 : : * Try to free up some memory. It's highly unlikely that we'll fail
586 : : * to do so here since the entry we've just added is yet to contain
587 : : * any tuples and we're able to remove any other entry to reduce the
588 : : * memory consumption.
589 : : */
590 [ - + ]: 1194 : if (unlikely(!cache_reduce_memory(mstate, key)))
1669 drowley@postgresql.o 591 :UBC 0 : return NULL;
592 : :
593 : : /*
594 : : * The process of removing entries from the cache may have caused the
595 : : * code in simplehash.h to shuffle elements to earlier buckets in the
596 : : * hash table. If it has, we'll need to find the entry again by
597 : : * performing a lookup. Fortunately, we can detect if this has
598 : : * happened by seeing if the entry is still in use and that the key
599 : : * pointer matches our expected key.
600 : : */
1566 drowley@postgresql.o 601 [ + - + + ]:CBC 1194 : if (entry->status != memoize_SH_IN_USE || entry->key != key)
602 : : {
603 : : /*
604 : : * We need to repopulate the probeslot as lookups performed during
605 : : * the cache evictions above will have stored some other key.
606 : : */
607 : 6 : prepare_probe_slot(mstate, key);
608 : :
609 : : /* Re-find the newly added entry */
610 : 6 : entry = memoize_lookup(mstate->hashtable, NULL);
1669 611 [ - + ]: 6 : Assert(entry != NULL);
612 : : }
613 : : }
614 : :
615 : 47760 : return entry;
616 : : }
617 : :
618 : : /*
619 : : * cache_store_tuple
620 : : * Add the tuple stored in 'slot' to the mstate's current cache entry.
621 : : * The cache entry must have already been made with cache_lookup().
622 : : * mstate's last_tuple field must point to the tail of mstate->entry's
623 : : * list of tuples.
624 : : */
625 : : static bool
1566 626 : 44583 : cache_store_tuple(MemoizeState *mstate, TupleTableSlot *slot)
627 : : {
628 : : MemoizeTuple *tuple;
629 : 44583 : MemoizeEntry *entry = mstate->entry;
630 : : MemoryContext oldcontext;
631 : :
1669 632 [ - + ]: 44583 : Assert(slot != NULL);
633 [ - + ]: 44583 : Assert(entry != NULL);
634 : :
1566 635 : 44583 : oldcontext = MemoryContextSwitchTo(mstate->tableContext);
636 : :
637 : 44583 : tuple = (MemoizeTuple *) palloc(sizeof(MemoizeTuple));
1669 638 : 44583 : tuple->mintuple = ExecCopySlotMinimalTuple(slot);
639 : 44583 : tuple->next = NULL;
640 : :
641 : : /* Account for the memory we just consumed */
1566 642 : 44583 : mstate->mem_used += CACHE_TUPLE_BYTES(tuple);
643 : :
1669 644 [ + + ]: 44583 : if (entry->tuplehead == NULL)
645 : : {
646 : : /*
647 : : * This is the first tuple for this entry, so just point the list head
648 : : * to it.
649 : : */
650 : 44354 : entry->tuplehead = tuple;
651 : : }
652 : : else
653 : : {
654 : : /* push this tuple onto the tail of the list */
1566 655 : 229 : mstate->last_tuple->next = tuple;
656 : : }
657 : :
658 : 44583 : mstate->last_tuple = tuple;
1669 659 : 44583 : MemoryContextSwitchTo(oldcontext);
660 : :
661 : : /*
662 : : * If we've gone over our memory budget then free up some space in the
663 : : * cache.
664 : : */
1566 665 [ - + ]: 44583 : if (mstate->mem_used > mstate->mem_limit)
666 : : {
1566 drowley@postgresql.o 667 :UBC 0 : MemoizeKey *key = entry->key;
668 : :
669 [ # # ]: 0 : if (!cache_reduce_memory(mstate, key))
1669 670 : 0 : return false;
671 : :
672 : : /*
673 : : * The process of removing entries from the cache may have caused the
674 : : * code in simplehash.h to shuffle elements to earlier buckets in the
675 : : * hash table. If it has, we'll need to find the entry again by
676 : : * performing a lookup. Fortunately, we can detect if this has
677 : : * happened by seeing if the entry is still in use and that the key
678 : : * pointer matches our expected key.
679 : : */
1566 680 [ # # # # ]: 0 : if (entry->status != memoize_SH_IN_USE || entry->key != key)
681 : : {
682 : : /*
683 : : * We need to repopulate the probeslot as lookups performed during
684 : : * the cache evictions above will have stored some other key.
685 : : */
686 : 0 : prepare_probe_slot(mstate, key);
687 : :
688 : : /* Re-find the entry */
689 : 0 : mstate->entry = entry = memoize_lookup(mstate->hashtable, NULL);
1669 690 [ # # ]: 0 : Assert(entry != NULL);
691 : : }
692 : : }
693 : :
1669 drowley@postgresql.o 694 :CBC 44583 : return true;
695 : : }
696 : :
697 : : static TupleTableSlot *
1566 698 : 456565 : ExecMemoize(PlanState *pstate)
699 : : {
700 : 456565 : MemoizeState *node = castNode(MemoizeState, pstate);
595 701 : 456565 : ExprContext *econtext = node->ss.ps.ps_ExprContext;
702 : : PlanState *outerNode;
703 : : TupleTableSlot *slot;
704 : :
705 [ + + ]: 456565 : CHECK_FOR_INTERRUPTS();
706 : :
707 : : /*
708 : : * Reset per-tuple memory context to free any expression evaluation
709 : : * storage allocated in the previous tuple cycle.
710 : : */
711 : 456565 : ResetExprContext(econtext);
712 : :
1566 713 [ + + + - : 456565 : switch (node->mstatus)
- - ]
714 : : {
715 : 377487 : case MEMO_CACHE_LOOKUP:
716 : : {
717 : : MemoizeEntry *entry;
718 : : TupleTableSlot *outerslot;
719 : : bool found;
720 : :
1669 721 [ - + ]: 377487 : Assert(node->entry == NULL);
722 : :
723 : : /* first call? we'll need a hash table. */
636 724 [ + + ]: 377487 : if (unlikely(node->hashtable == NULL))
725 : 833 : build_hash_table(node, ((Memoize *) pstate->plan)->est_entries);
726 : :
727 : : /*
728 : : * We're only ever in this state for the first call of the
729 : : * scan. Here we have a look to see if we've already seen the
730 : : * current parameters before and if we have already cached a
731 : : * complete set of records that the outer plan will return for
732 : : * these parameters.
733 : : *
734 : : * When we find a valid cache entry, we'll return the first
735 : : * tuple from it. If not found, we'll create a cache entry and
736 : : * then try to fetch a tuple from the outer scan. If we find
737 : : * one there, we'll try to cache it.
738 : : */
739 : :
740 : : /* see if we've got anything cached for the current parameters */
1669 741 : 377487 : entry = cache_lookup(node, &found);
742 : :
743 [ + + + - ]: 377487 : if (found && entry->complete)
744 : : {
745 : 329727 : node->stats.cache_hits += 1; /* stats update */
746 : :
747 : : /*
748 : : * Set last_tuple and entry so that the state
749 : : * MEMO_CACHE_FETCH_NEXT_TUPLE can easily find the next
750 : : * tuple for these parameters.
751 : : */
752 : 329727 : node->last_tuple = entry->tuplehead;
753 : 329727 : node->entry = entry;
754 : :
755 : : /* Fetch the first cached tuple, if there is one */
756 [ + + ]: 329727 : if (entry->tuplehead)
757 : : {
1566 758 : 184432 : node->mstatus = MEMO_CACHE_FETCH_NEXT_TUPLE;
759 : :
1669 760 : 184432 : slot = node->ss.ps.ps_ResultTupleSlot;
761 : 184432 : ExecStoreMinimalTuple(entry->tuplehead->mintuple,
762 : : slot, false);
763 : :
764 : 184432 : return slot;
765 : : }
766 : :
767 : : /* The cache entry is void of any tuples. */
1566 768 : 145295 : node->mstatus = MEMO_END_OF_SCAN;
1669 769 : 145295 : return NULL;
770 : : }
771 : :
772 : : /* Handle cache miss */
773 : 47760 : node->stats.cache_misses += 1; /* stats update */
774 : :
775 [ - + ]: 47760 : if (found)
776 : : {
777 : : /*
778 : : * A cache entry was found, but the scan for that entry
779 : : * did not run to completion. We'll just remove all
780 : : * tuples and start again. It might be tempting to
781 : : * continue where we left off, but there's no guarantee
782 : : * the outer node will produce the tuples in the same
783 : : * order as it did last time.
784 : : */
1669 drowley@postgresql.o 785 :UBC 0 : entry_purge_tuples(node, entry);
786 : : }
787 : :
788 : : /* Scan the outer node for a tuple to cache */
1669 drowley@postgresql.o 789 :CBC 47760 : outerNode = outerPlanState(node);
790 : 47760 : outerslot = ExecProcNode(outerNode);
791 [ + - + + ]: 47760 : if (TupIsNull(outerslot))
792 : : {
793 : : /*
794 : : * cache_lookup may have returned NULL due to failure to
795 : : * free enough cache space, so ensure we don't do anything
796 : : * here that assumes it worked. There's no need to go into
797 : : * bypass mode here as we're setting mstatus to end of
798 : : * scan.
799 : : */
800 [ + - ]: 3406 : if (likely(entry))
801 : 3406 : entry->complete = true;
802 : :
1566 803 : 3406 : node->mstatus = MEMO_END_OF_SCAN;
1669 804 : 3406 : return NULL;
805 : : }
806 : :
807 : 44354 : node->entry = entry;
808 : :
809 : : /*
810 : : * If we failed to create the entry or failed to store the
811 : : * tuple in the entry, then go into bypass mode.
812 : : */
813 [ + - - + : 44354 : if (unlikely(entry == NULL ||
- + ]
814 : : !cache_store_tuple(node, outerslot)))
815 : : {
1669 drowley@postgresql.o 816 :UBC 0 : node->stats.cache_overflows += 1; /* stats update */
817 : :
1566 818 : 0 : node->mstatus = MEMO_CACHE_BYPASS_MODE;
819 : :
820 : : /*
821 : : * No need to clear out last_tuple as we'll stay in bypass
822 : : * mode until the end of the scan.
823 : : */
824 : : }
825 : : else
826 : : {
827 : : /*
828 : : * If we only expect a single row from this scan then we
829 : : * can mark that we're not expecting more. This allows
830 : : * cache lookups to work even when the scan has not been
831 : : * executed to completion.
832 : : */
1669 drowley@postgresql.o 833 :CBC 44354 : entry->complete = node->singlerow;
1566 834 : 44354 : node->mstatus = MEMO_FILLING_CACHE;
835 : : }
836 : :
1669 837 : 44354 : slot = node->ss.ps.ps_ResultTupleSlot;
838 : 44354 : ExecCopySlot(slot, outerslot);
839 : 44354 : return slot;
840 : : }
841 : :
1566 842 : 45118 : case MEMO_CACHE_FETCH_NEXT_TUPLE:
843 : : {
844 : : /* We shouldn't be in this state if these are not set */
1669 845 [ - + ]: 45118 : Assert(node->entry != NULL);
846 [ - + ]: 45118 : Assert(node->last_tuple != NULL);
847 : :
848 : : /* Skip to the next tuple to output */
849 : 45118 : node->last_tuple = node->last_tuple->next;
850 : :
851 : : /* No more tuples in the cache */
852 [ + + ]: 45118 : if (node->last_tuple == NULL)
853 : : {
1566 854 : 42520 : node->mstatus = MEMO_END_OF_SCAN;
1669 855 : 42520 : return NULL;
856 : : }
857 : :
858 : 2598 : slot = node->ss.ps.ps_ResultTupleSlot;
859 : 2598 : ExecStoreMinimalTuple(node->last_tuple->mintuple, slot,
860 : : false);
861 : :
862 : 2598 : return slot;
863 : : }
864 : :
1566 865 : 33960 : case MEMO_FILLING_CACHE:
866 : : {
867 : : TupleTableSlot *outerslot;
868 : 33960 : MemoizeEntry *entry = node->entry;
869 : :
870 : : /* entry should already have been set by MEMO_CACHE_LOOKUP */
1669 871 [ - + ]: 33960 : Assert(entry != NULL);
872 : :
873 : : /*
874 : : * When in the MEMO_FILLING_CACHE state, we've just had a
875 : : * cache miss and are populating the cache with the current
876 : : * scan tuples.
877 : : */
878 : 33960 : outerNode = outerPlanState(node);
879 : 33960 : outerslot = ExecProcNode(outerNode);
880 [ + + + + ]: 33960 : if (TupIsNull(outerslot))
881 : : {
882 : : /* No more tuples. Mark it as complete */
883 : 33731 : entry->complete = true;
1566 884 : 33731 : node->mstatus = MEMO_END_OF_SCAN;
1669 885 : 33731 : return NULL;
886 : : }
887 : :
888 : : /*
889 : : * Validate if the planner properly set the singlerow flag. It
890 : : * should only set that if each cache entry can, at most,
891 : : * return 1 row.
892 : : */
893 [ - + ]: 229 : if (unlikely(entry->complete))
1669 drowley@postgresql.o 894 [ # # ]:UBC 0 : elog(ERROR, "cache entry already complete");
895 : :
896 : : /* Record the tuple in the current cache entry */
1669 drowley@postgresql.o 897 [ - + ]:CBC 229 : if (unlikely(!cache_store_tuple(node, outerslot)))
898 : : {
899 : : /* Couldn't store it? Handle overflow */
1669 drowley@postgresql.o 900 :UBC 0 : node->stats.cache_overflows += 1; /* stats update */
901 : :
1566 902 : 0 : node->mstatus = MEMO_CACHE_BYPASS_MODE;
903 : :
904 : : /*
905 : : * No need to clear out entry or last_tuple as we'll stay
906 : : * in bypass mode until the end of the scan.
907 : : */
908 : : }
909 : :
1669 drowley@postgresql.o 910 :CBC 229 : slot = node->ss.ps.ps_ResultTupleSlot;
911 : 229 : ExecCopySlot(slot, outerslot);
912 : 229 : return slot;
913 : : }
914 : :
1566 drowley@postgresql.o 915 :UBC 0 : case MEMO_CACHE_BYPASS_MODE:
916 : : {
917 : : TupleTableSlot *outerslot;
918 : :
919 : : /*
920 : : * When in bypass mode we just continue to read tuples without
921 : : * caching. We need to wait until the next rescan before we
922 : : * can come out of this mode.
923 : : */
1669 924 : 0 : outerNode = outerPlanState(node);
925 : 0 : outerslot = ExecProcNode(outerNode);
926 [ # # # # ]: 0 : if (TupIsNull(outerslot))
927 : : {
1566 928 : 0 : node->mstatus = MEMO_END_OF_SCAN;
1669 929 : 0 : return NULL;
930 : : }
931 : :
932 : 0 : slot = node->ss.ps.ps_ResultTupleSlot;
933 : 0 : ExecCopySlot(slot, outerslot);
934 : 0 : return slot;
935 : : }
936 : :
1566 937 : 0 : case MEMO_END_OF_SCAN:
938 : :
939 : : /*
940 : : * We've already returned NULL for this scan, but just in case
941 : : * something calls us again by mistake.
942 : : */
1669 943 : 0 : return NULL;
944 : :
945 : 0 : default:
1566 946 [ # # ]: 0 : elog(ERROR, "unrecognized memoize state: %d",
947 : : (int) node->mstatus);
948 : : return NULL;
949 : : } /* switch */
950 : : }
951 : :
952 : : MemoizeState *
1566 drowley@postgresql.o 953 :CBC 1024 : ExecInitMemoize(Memoize *node, EState *estate, int eflags)
954 : : {
955 : 1024 : MemoizeState *mstate = makeNode(MemoizeState);
956 : : Plan *outerNode;
957 : : int i;
958 : : int nkeys;
959 : : Oid *eqfuncoids;
960 : :
961 : : /* check for unsupported flags */
1669 962 [ - + ]: 1024 : Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
963 : :
1566 964 : 1024 : mstate->ss.ps.plan = (Plan *) node;
965 : 1024 : mstate->ss.ps.state = estate;
966 : 1024 : mstate->ss.ps.ExecProcNode = ExecMemoize;
967 : :
968 : : /*
969 : : * Miscellaneous initialization
970 : : *
971 : : * create expression context for node
972 : : */
973 : 1024 : ExecAssignExprContext(estate, &mstate->ss.ps);
974 : :
1669 975 : 1024 : outerNode = outerPlan(node);
1566 976 : 1024 : outerPlanState(mstate) = ExecInitNode(outerNode, estate, eflags);
977 : :
978 : : /*
979 : : * Initialize return slot and type. No need to initialize projection info
980 : : * because this node doesn't do projections.
981 : : */
982 : 1024 : ExecInitResultTupleSlotTL(&mstate->ss.ps, &TTSOpsMinimalTuple);
983 : 1024 : mstate->ss.ps.ps_ProjInfo = NULL;
984 : :
985 : : /*
986 : : * Initialize scan slot and type.
987 : : */
988 : 1024 : ExecCreateScanSlotFromOuterPlan(estate, &mstate->ss, &TTSOpsMinimalTuple);
989 : :
990 : : /*
991 : : * Set the state machine to lookup the cache. We won't find anything
992 : : * until we cache something, but this saves a special case to create the
993 : : * first entry.
994 : : */
995 : 1024 : mstate->mstatus = MEMO_CACHE_LOOKUP;
996 : :
997 : 1024 : mstate->nkeys = nkeys = node->numKeys;
998 : 1024 : mstate->hashkeydesc = ExecTypeFromExprList(node->param_exprs);
999 : 1024 : mstate->tableslot = MakeSingleTupleTableSlot(mstate->hashkeydesc,
1000 : : &TTSOpsMinimalTuple);
1001 : 1024 : mstate->probeslot = MakeSingleTupleTableSlot(mstate->hashkeydesc,
1002 : : &TTSOpsVirtual);
1003 : :
1004 : 1024 : mstate->param_exprs = (ExprState **) palloc(nkeys * sizeof(ExprState *));
1005 : 1024 : mstate->collations = node->collations; /* Just point directly to the plan
1006 : : * data */
1007 : 1024 : mstate->hashfunctions = (FmgrInfo *) palloc(nkeys * sizeof(FmgrInfo));
1008 : :
1669 1009 : 1024 : eqfuncoids = palloc(nkeys * sizeof(Oid));
1010 : :
1011 [ + + ]: 2081 : for (i = 0; i < nkeys; i++)
1012 : : {
1013 : 1057 : Oid hashop = node->hashOperators[i];
1014 : : Oid left_hashfn;
1015 : : Oid right_hashfn;
1016 : 1057 : Expr *param_expr = (Expr *) list_nth(node->param_exprs, i);
1017 : :
1018 [ - + ]: 1057 : if (!get_op_hash_functions(hashop, &left_hashfn, &right_hashfn))
1669 drowley@postgresql.o 1019 [ # # ]:UBC 0 : elog(ERROR, "could not find hash function for hash operator %u",
1020 : : hashop);
1021 : :
1566 drowley@postgresql.o 1022 :CBC 1057 : fmgr_info(left_hashfn, &mstate->hashfunctions[i]);
1023 : :
1024 : 1057 : mstate->param_exprs[i] = ExecInitExpr(param_expr, (PlanState *) mstate);
1669 1025 : 1057 : eqfuncoids[i] = get_opcode(hashop);
1026 : : }
1027 : :
1566 1028 : 2048 : mstate->cache_eq_expr = ExecBuildParamSetEqual(mstate->hashkeydesc,
1029 : : &TTSOpsMinimalTuple,
1030 : : &TTSOpsVirtual,
1031 : : eqfuncoids,
1032 : 1024 : node->collations,
1033 : 1024 : node->param_exprs,
1034 : : (PlanState *) mstate);
1035 : :
1669 1036 : 1024 : pfree(eqfuncoids);
1566 1037 : 1024 : mstate->mem_used = 0;
1038 : :
1039 : : /* Limit the total memory consumed by the cache to this */
1555 tgl@sss.pgh.pa.us 1040 : 1024 : mstate->mem_limit = get_hash_memory_limit();
1041 : :
1042 : : /* A memory context dedicated for the cache */
1566 drowley@postgresql.o 1043 : 1024 : mstate->tableContext = AllocSetContextCreate(CurrentMemoryContext,
1044 : : "MemoizeHashTable",
1045 : : ALLOCSET_DEFAULT_SIZES);
1046 : :
1047 : 1024 : dlist_init(&mstate->lru_list);
1048 : 1024 : mstate->last_tuple = NULL;
1049 : 1024 : mstate->entry = NULL;
1050 : :
1051 : : /*
1052 : : * Mark if we can assume the cache entry is completed after we get the
1053 : : * first record for it. Some callers might not call us again after
1054 : : * getting the first match. e.g. A join operator performing a unique join
1055 : : * is able to skip to the next outer tuple after getting the first
1056 : : * matching inner tuple. In this case, the cache entry is complete after
1057 : : * getting the first tuple. This allows us to mark it as so.
1058 : : */
1059 : 1024 : mstate->singlerow = node->singlerow;
1433 1060 : 1024 : mstate->keyparamids = node->keyparamids;
1061 : :
1062 : : /*
1063 : : * Record if the cache keys should be compared bit by bit, or logically
1064 : : * using the type's hash equality operator
1065 : : */
1066 : 1024 : mstate->binary_mode = node->binary_mode;
1067 : :
1068 : : /* Zero the statistics counters */
1566 1069 : 1024 : memset(&mstate->stats, 0, sizeof(MemoizeInstrumentation));
1070 : :
1071 : : /*
1072 : : * Because it may require a large allocation, we delay building of the
1073 : : * hash table until executor run.
1074 : : */
636 1075 : 1024 : mstate->hashtable = NULL;
1076 : :
1566 1077 : 1024 : return mstate;
1078 : : }
1079 : :
1080 : : void
1081 : 1024 : ExecEndMemoize(MemoizeState *node)
1082 : : {
1083 : : #ifdef USE_ASSERT_CHECKING
1084 : : /* Validate the memory accounting code is correct in assert builds. */
636 1085 [ + + ]: 1024 : if (node->hashtable != NULL)
1086 : : {
1087 : : int count;
1632 1088 : 827 : uint64 mem = 0;
1089 : : memoize_iterator i;
1090 : : MemoizeEntry *entry;
1091 : :
1566 1092 : 827 : memoize_start_iterate(node->hashtable, &i);
1093 : :
1632 1094 : 827 : count = 0;
1566 1095 [ + + ]: 46943 : while ((entry = memoize_iterate(node->hashtable, &i)) != NULL)
1096 : : {
1097 : 46116 : MemoizeTuple *tuple = entry->tuplehead;
1098 : :
1632 1099 : 46116 : mem += EMPTY_ENTRY_MEMORY_BYTES(entry);
1100 [ + + ]: 89505 : while (tuple != NULL)
1101 : : {
1102 : 43389 : mem += CACHE_TUPLE_BYTES(tuple);
1103 : 43389 : tuple = tuple->next;
1104 : : }
1105 : 46116 : count++;
1106 : : }
1107 : :
1108 [ - + ]: 827 : Assert(count == node->hashtable->members);
1109 [ - + ]: 827 : Assert(mem == node->mem_used);
1110 : : }
1111 : : #endif
1112 : :
1113 : : /*
1114 : : * When ending a parallel worker, copy the statistics gathered by the
1115 : : * worker back into shared memory so that it can be picked up by the main
1116 : : * process to report in EXPLAIN ANALYZE.
1117 : : */
1669 1118 [ - + - - ]: 1024 : if (node->shared_info != NULL && IsParallelWorker())
1119 : : {
1120 : : MemoizeInstrumentation *si;
1121 : :
1122 : : /* Make mem_peak available for EXPLAIN */
1669 drowley@postgresql.o 1123 [ # # ]:UBC 0 : if (node->stats.mem_peak == 0)
1124 : 0 : node->stats.mem_peak = node->mem_used;
1125 : :
1126 [ # # ]: 0 : Assert(ParallelWorkerNumber <= node->shared_info->num_workers);
1127 : 0 : si = &node->shared_info->sinstrument[ParallelWorkerNumber];
1566 1128 : 0 : memcpy(si, &node->stats, sizeof(MemoizeInstrumentation));
1129 : : }
1130 : :
1131 : : /* Remove the cache context */
1669 drowley@postgresql.o 1132 :CBC 1024 : MemoryContextDelete(node->tableContext);
1133 : :
1134 : : /*
1135 : : * shut down the subplan
1136 : : */
1137 : 1024 : ExecEndNode(outerPlanState(node));
1138 : 1024 : }
1139 : :
1140 : : void
1566 1141 : 377487 : ExecReScanMemoize(MemoizeState *node)
1142 : : {
1669 1143 : 377487 : PlanState *outerPlan = outerPlanState(node);
1144 : :
1145 : : /* Mark that we must lookup the cache for a new set of parameters */
1566 1146 : 377487 : node->mstatus = MEMO_CACHE_LOOKUP;
1147 : :
1148 : : /* nullify pointers used for the last scan */
1669 1149 : 377487 : node->entry = NULL;
1150 : 377487 : node->last_tuple = NULL;
1151 : :
1152 : : /*
1153 : : * if chgParam of subnode is not null then plan will be re-scanned by
1154 : : * first ExecProcNode.
1155 : : */
1156 [ - + ]: 377487 : if (outerPlan->chgParam == NULL)
1669 drowley@postgresql.o 1157 :UBC 0 : ExecReScan(outerPlan);
1158 : :
1159 : : /*
1160 : : * Purge the entire cache if a parameter changed that is not part of the
1161 : : * cache key.
1162 : : */
1433 drowley@postgresql.o 1163 [ + + ]:CBC 377487 : if (bms_nonempty_difference(outerPlan->chgParam, node->keyparamids))
1164 : 9 : cache_purge_all(node);
1669 1165 : 377487 : }
1166 : :
1167 : : /*
1168 : : * ExecEstimateCacheEntryOverheadBytes
1169 : : * For use in the query planner to help it estimate the amount of memory
1170 : : * required to store a single entry in the cache.
1171 : : */
1172 : : double
1173 : 145743 : ExecEstimateCacheEntryOverheadBytes(double ntuples)
1174 : : {
1566 1175 : 145743 : return sizeof(MemoizeEntry) + sizeof(MemoizeKey) + sizeof(MemoizeTuple) *
1176 : : ntuples;
1177 : : }
1178 : :
1179 : : /* ----------------------------------------------------------------
1180 : : * Parallel Query Support
1181 : : * ----------------------------------------------------------------
1182 : : */
1183 : :
1184 : : /* ----------------------------------------------------------------
1185 : : * ExecMemoizeEstimate
1186 : : *
1187 : : * Estimate space required to propagate memoize statistics.
1188 : : * ----------------------------------------------------------------
1189 : : */
1190 : : void
1191 : 3 : ExecMemoizeEstimate(MemoizeState *node, ParallelContext *pcxt)
1192 : : {
1193 : : Size size;
1194 : :
1195 : : /* don't need this if not instrumenting or no workers */
1669 1196 [ - + - - ]: 3 : if (!node->ss.ps.instrument || pcxt->nworkers == 0)
1197 : 3 : return;
1198 : :
1566 drowley@postgresql.o 1199 :UBC 0 : size = mul_size(pcxt->nworkers, sizeof(MemoizeInstrumentation));
1200 : 0 : size = add_size(size, offsetof(SharedMemoizeInfo, sinstrument));
1669 1201 : 0 : shm_toc_estimate_chunk(&pcxt->estimator, size);
1202 : 0 : shm_toc_estimate_keys(&pcxt->estimator, 1);
1203 : : }
1204 : :
1205 : : /* ----------------------------------------------------------------
1206 : : * ExecMemoizeInitializeDSM
1207 : : *
1208 : : * Initialize DSM space for memoize statistics.
1209 : : * ----------------------------------------------------------------
1210 : : */
1211 : : void
1566 drowley@postgresql.o 1212 :CBC 3 : ExecMemoizeInitializeDSM(MemoizeState *node, ParallelContext *pcxt)
1213 : : {
1214 : : Size size;
1215 : :
1216 : : /* don't need this if not instrumenting or no workers */
1669 1217 [ - + - - ]: 3 : if (!node->ss.ps.instrument || pcxt->nworkers == 0)
1218 : 3 : return;
1219 : :
1566 drowley@postgresql.o 1220 :UBC 0 : size = offsetof(SharedMemoizeInfo, sinstrument)
1221 : 0 : + pcxt->nworkers * sizeof(MemoizeInstrumentation);
1669 1222 : 0 : node->shared_info = shm_toc_allocate(pcxt->toc, size);
1223 : : /* ensure any unfilled slots will contain zeroes */
1224 : 0 : memset(node->shared_info, 0, size);
1225 : 0 : node->shared_info->num_workers = pcxt->nworkers;
1226 : 0 : shm_toc_insert(pcxt->toc, node->ss.ps.plan->plan_node_id,
1227 : 0 : node->shared_info);
1228 : : }
1229 : :
1230 : : /* ----------------------------------------------------------------
1231 : : * ExecMemoizeInitializeWorker
1232 : : *
1233 : : * Attach worker to DSM space for memoize statistics.
1234 : : * ----------------------------------------------------------------
1235 : : */
1236 : : void
1566 drowley@postgresql.o 1237 :CBC 6 : ExecMemoizeInitializeWorker(MemoizeState *node, ParallelWorkerContext *pwcxt)
1238 : : {
1669 1239 : 6 : node->shared_info =
1240 : 6 : shm_toc_lookup(pwcxt->toc, node->ss.ps.plan->plan_node_id, true);
1241 : 6 : }
1242 : :
1243 : : /* ----------------------------------------------------------------
1244 : : * ExecMemoizeRetrieveInstrumentation
1245 : : *
1246 : : * Transfer memoize statistics from DSM to private memory.
1247 : : * ----------------------------------------------------------------
1248 : : */
1249 : : void
1566 drowley@postgresql.o 1250 :UBC 0 : ExecMemoizeRetrieveInstrumentation(MemoizeState *node)
1251 : : {
1252 : : Size size;
1253 : : SharedMemoizeInfo *si;
1254 : :
1669 1255 [ # # ]: 0 : if (node->shared_info == NULL)
1256 : 0 : return;
1257 : :
1566 1258 : 0 : size = offsetof(SharedMemoizeInfo, sinstrument)
1259 : 0 : + node->shared_info->num_workers * sizeof(MemoizeInstrumentation);
1669 1260 : 0 : si = palloc(size);
1261 : 0 : memcpy(si, node->shared_info, size);
1262 : 0 : node->shared_info = si;
1263 : : }
|
