Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * hashinsert.c
4 : * Item insertion in hash tables for Postgres.
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/access/hash/hashinsert.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/hash.h"
19 : #include "access/hash_xlog.h"
20 : #include "access/xloginsert.h"
21 : #include "miscadmin.h"
22 : #include "storage/predicate.h"
23 : #include "utils/rel.h"
24 :
25 : static void _hash_vacuum_one_page(Relation rel, Relation hrel,
26 : Buffer metabuf, Buffer buf);
27 :
28 : /*
29 : * _hash_doinsert() -- Handle insertion of a single index tuple.
30 : *
31 : * This routine is called by the public interface routines, hashbuild
32 : * and hashinsert. By here, itup is completely filled in.
33 : *
34 : * 'sorted' must only be passed as 'true' when inserts are done in hashkey
35 : * order.
36 : */
37 : void
38 490813 : _hash_doinsert(Relation rel, IndexTuple itup, Relation heapRel, bool sorted)
39 : {
40 490813 : Buffer buf = InvalidBuffer;
41 : Buffer bucket_buf;
42 : Buffer metabuf;
43 : HashMetaPage metap;
44 490813 : HashMetaPage usedmetap = NULL;
45 : Page metapage;
46 : Page page;
47 : HashPageOpaque pageopaque;
48 : Size itemsz;
49 : bool do_expand;
50 : uint32 hashkey;
51 : Bucket bucket;
52 : OffsetNumber itup_off;
53 : XLogRecPtr recptr;
54 :
55 : /*
56 : * Get the hash key for the item (it's stored in the index tuple itself).
57 : */
58 490813 : hashkey = _hash_get_indextuple_hashkey(itup);
59 :
60 : /* compute item size too */
61 490813 : itemsz = IndexTupleSize(itup);
62 490813 : itemsz = MAXALIGN(itemsz); /* be safe, PageAddItem will do this but we
63 : * need to be consistent */
64 :
65 490813 : restart_insert:
66 :
67 : /*
68 : * Read the metapage. We don't lock it yet; HashMaxItemSize() will
69 : * examine pd_pagesize_version, but that can't change so we can examine it
70 : * without a lock.
71 : */
72 490813 : metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_NOLOCK, LH_META_PAGE);
73 490813 : metapage = BufferGetPage(metabuf);
74 :
75 : /*
76 : * Check whether the item can fit on a hash page at all. (Eventually, we
77 : * ought to try to apply TOAST methods if not.) Note that at this point,
78 : * itemsz doesn't include the ItemId.
79 : *
80 : * XXX this is useless code if we are only storing hash keys.
81 : */
82 490813 : if (itemsz > HashMaxItemSize(metapage))
83 0 : ereport(ERROR,
84 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
85 : errmsg("index row size %zu exceeds hash maximum %zu",
86 : itemsz, HashMaxItemSize(metapage)),
87 : errhint("Values larger than a buffer page cannot be indexed.")));
88 :
89 : /* Lock the primary bucket page for the target bucket. */
90 490813 : buf = _hash_getbucketbuf_from_hashkey(rel, hashkey, HASH_WRITE,
91 : &usedmetap);
92 : Assert(usedmetap != NULL);
93 :
94 490813 : CheckForSerializableConflictIn(rel, NULL, BufferGetBlockNumber(buf));
95 :
96 : /* remember the primary bucket buffer to release the pin on it at end. */
97 490807 : bucket_buf = buf;
98 :
99 490807 : page = BufferGetPage(buf);
100 490807 : pageopaque = HashPageGetOpaque(page);
101 490807 : bucket = pageopaque->hasho_bucket;
102 :
103 : /*
104 : * If this bucket is in the process of being split, try to finish the
105 : * split before inserting, because that might create room for the
106 : * insertion to proceed without allocating an additional overflow page.
107 : * It's only interesting to finish the split if we're trying to insert
108 : * into the bucket from which we're removing tuples (the "old" bucket),
109 : * not if we're trying to insert into the bucket into which tuples are
110 : * being moved (the "new" bucket).
111 : */
112 490807 : if (H_BUCKET_BEING_SPLIT(pageopaque) && IsBufferCleanupOK(buf))
113 : {
114 : /* release the lock on bucket buffer, before completing the split. */
115 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
116 :
117 0 : _hash_finish_split(rel, metabuf, buf, bucket,
118 0 : usedmetap->hashm_maxbucket,
119 0 : usedmetap->hashm_highmask,
120 0 : usedmetap->hashm_lowmask);
121 :
122 : /* release the pin on old and meta buffer. retry for insert. */
123 0 : _hash_dropbuf(rel, buf);
124 0 : _hash_dropbuf(rel, metabuf);
125 0 : goto restart_insert;
126 : }
127 :
128 : /* Do the insertion */
129 846949 : while (PageGetFreeSpace(page) < itemsz)
130 : {
131 : BlockNumber nextblkno;
132 :
133 : /*
134 : * Check if current page has any DEAD tuples. If yes, delete these
135 : * tuples and see if we can get a space for the new item to be
136 : * inserted before moving to the next page in the bucket chain.
137 : */
138 356146 : if (H_HAS_DEAD_TUPLES(pageopaque))
139 : {
140 :
141 4 : if (IsBufferCleanupOK(buf))
142 : {
143 4 : _hash_vacuum_one_page(rel, heapRel, metabuf, buf);
144 :
145 4 : if (PageGetFreeSpace(page) >= itemsz)
146 4 : break; /* OK, now we have enough space */
147 : }
148 : }
149 :
150 : /*
151 : * no space on this page; check for an overflow page
152 : */
153 356142 : nextblkno = pageopaque->hasho_nextblkno;
154 :
155 356142 : if (BlockNumberIsValid(nextblkno))
156 : {
157 : /*
158 : * ovfl page exists; go get it. if it doesn't have room, we'll
159 : * find out next pass through the loop test above. we always
160 : * release both the lock and pin if this is an overflow page, but
161 : * only the lock if this is the primary bucket page, since the pin
162 : * on the primary bucket must be retained throughout the scan.
163 : */
164 355916 : if (buf != bucket_buf)
165 289024 : _hash_relbuf(rel, buf);
166 : else
167 66892 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
168 355916 : buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
169 355916 : page = BufferGetPage(buf);
170 : }
171 : else
172 : {
173 : /*
174 : * we're at the end of the bucket chain and we haven't found a
175 : * page with enough room. allocate a new overflow page.
176 : */
177 :
178 : /* release our write lock without modifying buffer */
179 226 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
180 :
181 : /* chain to a new overflow page */
182 226 : buf = _hash_addovflpage(rel, metabuf, buf, (buf == bucket_buf));
183 226 : page = BufferGetPage(buf);
184 :
185 : /* should fit now, given test above */
186 : Assert(PageGetFreeSpace(page) >= itemsz);
187 : }
188 356142 : pageopaque = HashPageGetOpaque(page);
189 : Assert((pageopaque->hasho_flag & LH_PAGE_TYPE) == LH_OVERFLOW_PAGE);
190 : Assert(pageopaque->hasho_bucket == bucket);
191 : }
192 :
193 : /*
194 : * Write-lock the metapage so we can increment the tuple count. After
195 : * incrementing it, check to see if it's time for a split.
196 : */
197 490807 : LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
198 :
199 : /* Do the update. No ereport(ERROR) until changes are logged */
200 490807 : START_CRIT_SECTION();
201 :
202 : /* found page with enough space, so add the item here */
203 490807 : itup_off = _hash_pgaddtup(rel, buf, itemsz, itup, sorted);
204 490807 : MarkBufferDirty(buf);
205 :
206 : /* metapage operations */
207 490807 : metap = HashPageGetMeta(metapage);
208 490807 : metap->hashm_ntuples += 1;
209 :
210 : /* Make sure this stays in sync with _hash_expandtable() */
211 490807 : do_expand = metap->hashm_ntuples >
212 490807 : (double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1);
213 :
214 490807 : MarkBufferDirty(metabuf);
215 :
216 : /* XLOG stuff */
217 490807 : if (RelationNeedsWAL(rel))
218 407717 : {
219 : xl_hash_insert xlrec;
220 :
221 407717 : xlrec.offnum = itup_off;
222 :
223 407717 : XLogBeginInsert();
224 407717 : XLogRegisterData(&xlrec, SizeOfHashInsert);
225 :
226 407717 : XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
227 :
228 407717 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
229 407717 : XLogRegisterBufData(0, itup, IndexTupleSize(itup));
230 :
231 407717 : recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INSERT);
232 : }
233 : else
234 83090 : recptr = XLogGetFakeLSN(rel);
235 :
236 490807 : PageSetLSN(BufferGetPage(buf), recptr);
237 490807 : PageSetLSN(BufferGetPage(metabuf), recptr);
238 :
239 490807 : END_CRIT_SECTION();
240 :
241 : /* drop lock on metapage, but keep pin */
242 490807 : LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
243 :
244 : /*
245 : * Release the modified page and ensure to release the pin on primary
246 : * page.
247 : */
248 490807 : _hash_relbuf(rel, buf);
249 490807 : if (buf != bucket_buf)
250 66966 : _hash_dropbuf(rel, bucket_buf);
251 :
252 : /* Attempt to split if a split is needed */
253 490807 : if (do_expand)
254 893 : _hash_expandtable(rel, metabuf);
255 :
256 : /* Finally drop our pin on the metapage */
257 490807 : _hash_dropbuf(rel, metabuf);
258 490807 : }
259 :
260 : /*
261 : * _hash_pgaddtup() -- add a tuple to a particular page in the index.
262 : *
263 : * This routine adds the tuple to the page as requested; it does not write out
264 : * the page. It is an error to call this function without pin and write lock
265 : * on the target buffer.
266 : *
267 : * Returns the offset number at which the tuple was inserted. This function
268 : * is responsible for preserving the condition that tuples in a hash index
269 : * page are sorted by hashkey value, however, if the caller is certain that
270 : * the hashkey for the tuple being added is >= the hashkeys of all existing
271 : * tuples on the page, then the 'appendtup' flag may be passed as true. This
272 : * saves from having to binary search for the correct location to insert the
273 : * tuple.
274 : */
275 : OffsetNumber
276 490807 : _hash_pgaddtup(Relation rel, Buffer buf, Size itemsize, IndexTuple itup,
277 : bool appendtup)
278 : {
279 : OffsetNumber itup_off;
280 : Page page;
281 :
282 490807 : _hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
283 490807 : page = BufferGetPage(buf);
284 :
285 : /*
286 : * Find where to insert the tuple (preserving page's hashkey ordering). If
287 : * 'appendtup' is true then we just insert it at the end.
288 : */
289 490807 : if (appendtup)
290 : {
291 70500 : itup_off = PageGetMaxOffsetNumber(page) + 1;
292 :
293 : #ifdef USE_ASSERT_CHECKING
294 : /* ensure this tuple's hashkey is >= the final existing tuple */
295 : if (PageGetMaxOffsetNumber(page) > 0)
296 : {
297 : IndexTuple lasttup;
298 : ItemId itemid;
299 :
300 : itemid = PageGetItemId(page, PageGetMaxOffsetNumber(page));
301 : lasttup = (IndexTuple) PageGetItem(page, itemid);
302 :
303 : Assert(_hash_get_indextuple_hashkey(lasttup) <=
304 : _hash_get_indextuple_hashkey(itup));
305 : }
306 : #endif
307 : }
308 : else
309 : {
310 420307 : uint32 hashkey = _hash_get_indextuple_hashkey(itup);
311 :
312 420307 : itup_off = _hash_binsearch(page, hashkey);
313 : }
314 :
315 490807 : if (PageAddItem(page, itup, itemsize, itup_off, false, false) == InvalidOffsetNumber)
316 0 : elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(rel));
317 :
318 490807 : return itup_off;
319 : }
320 :
321 : /*
322 : * _hash_pgaddmultitup() -- add a tuple vector to a particular page in the
323 : * index.
324 : *
325 : * This routine has same requirements for locking and tuple ordering as
326 : * _hash_pgaddtup().
327 : *
328 : * Returns the offset number array at which the tuples were inserted.
329 : */
330 : void
331 1005 : _hash_pgaddmultitup(Relation rel, Buffer buf, IndexTuple *itups,
332 : OffsetNumber *itup_offsets, uint16 nitups)
333 : {
334 : OffsetNumber itup_off;
335 : Page page;
336 : uint32 hashkey;
337 : int i;
338 :
339 1005 : _hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
340 1005 : page = BufferGetPage(buf);
341 :
342 88496 : for (i = 0; i < nitups; i++)
343 : {
344 : Size itemsize;
345 :
346 87491 : itemsize = IndexTupleSize(itups[i]);
347 87491 : itemsize = MAXALIGN(itemsize);
348 :
349 : /* Find where to insert the tuple (preserving page's hashkey ordering) */
350 87491 : hashkey = _hash_get_indextuple_hashkey(itups[i]);
351 87491 : itup_off = _hash_binsearch(page, hashkey);
352 :
353 87491 : itup_offsets[i] = itup_off;
354 :
355 87491 : if (PageAddItem(page, itups[i], itemsize, itup_off, false, false) == InvalidOffsetNumber)
356 0 : elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(rel));
357 : }
358 1005 : }
359 :
360 : /*
361 : * _hash_vacuum_one_page - vacuum just one index page.
362 : *
363 : * Try to remove LP_DEAD items from the given page. We must acquire cleanup
364 : * lock on the page being modified before calling this function.
365 : */
366 :
367 : static void
368 4 : _hash_vacuum_one_page(Relation rel, Relation hrel, Buffer metabuf, Buffer buf)
369 : {
370 : OffsetNumber deletable[MaxOffsetNumber];
371 4 : int ndeletable = 0;
372 : OffsetNumber offnum,
373 : maxoff;
374 4 : Page page = BufferGetPage(buf);
375 : HashPageOpaque pageopaque;
376 : HashMetaPage metap;
377 : XLogRecPtr recptr;
378 :
379 : /* Scan each tuple in page to see if it is marked as LP_DEAD */
380 4 : maxoff = PageGetMaxOffsetNumber(page);
381 4 : for (offnum = FirstOffsetNumber;
382 1632 : offnum <= maxoff;
383 1628 : offnum = OffsetNumberNext(offnum))
384 : {
385 1628 : ItemId itemId = PageGetItemId(page, offnum);
386 :
387 1628 : if (ItemIdIsDead(itemId))
388 884 : deletable[ndeletable++] = offnum;
389 : }
390 :
391 4 : if (ndeletable > 0)
392 : {
393 : TransactionId snapshotConflictHorizon;
394 :
395 : snapshotConflictHorizon =
396 4 : index_compute_xid_horizon_for_tuples(rel, hrel, buf,
397 : deletable, ndeletable);
398 :
399 : /*
400 : * Write-lock the meta page so that we can decrement tuple count.
401 : */
402 4 : LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
403 :
404 : /* No ereport(ERROR) until changes are logged */
405 4 : START_CRIT_SECTION();
406 :
407 4 : PageIndexMultiDelete(page, deletable, ndeletable);
408 :
409 : /*
410 : * Mark the page as not containing any LP_DEAD items. This is not
411 : * certainly true (there might be some that have recently been marked,
412 : * but weren't included in our target-item list), but it will almost
413 : * always be true and it doesn't seem worth an additional page scan to
414 : * check it. Remember that LH_PAGE_HAS_DEAD_TUPLES is only a hint
415 : * anyway.
416 : */
417 4 : pageopaque = HashPageGetOpaque(page);
418 4 : pageopaque->hasho_flag &= ~LH_PAGE_HAS_DEAD_TUPLES;
419 :
420 4 : metap = HashPageGetMeta(BufferGetPage(metabuf));
421 4 : metap->hashm_ntuples -= ndeletable;
422 :
423 4 : MarkBufferDirty(buf);
424 4 : MarkBufferDirty(metabuf);
425 :
426 : /* XLOG stuff */
427 4 : if (RelationNeedsWAL(rel))
428 4 : {
429 : xl_hash_vacuum_one_page xlrec;
430 :
431 4 : xlrec.isCatalogRel = RelationIsAccessibleInLogicalDecoding(hrel);
432 4 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
433 4 : xlrec.ntuples = ndeletable;
434 :
435 4 : XLogBeginInsert();
436 4 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
437 4 : XLogRegisterData(&xlrec, SizeOfHashVacuumOnePage);
438 :
439 : /*
440 : * We need the target-offsets array whether or not we store the
441 : * whole buffer, to allow us to find the snapshotConflictHorizon
442 : * on a standby server.
443 : */
444 4 : XLogRegisterData(deletable,
445 : ndeletable * sizeof(OffsetNumber));
446 :
447 4 : XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
448 :
449 4 : recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_VACUUM_ONE_PAGE);
450 : }
451 : else
452 0 : recptr = XLogGetFakeLSN(rel);
453 :
454 4 : PageSetLSN(BufferGetPage(buf), recptr);
455 4 : PageSetLSN(BufferGetPage(metabuf), recptr);
456 :
457 4 : END_CRIT_SECTION();
458 :
459 : /*
460 : * Releasing write lock on meta page as we have updated the tuple
461 : * count.
462 : */
463 4 : LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
464 : }
465 4 : }
|
