Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * hashpage.c
4 : * Hash table page management code for the Postgres hash access method
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/hash/hashpage.c
12 : *
13 : * NOTES
14 : * Postgres hash pages look like ordinary relation pages. The opaque
15 : * data at high addresses includes information about the page including
16 : * whether a page is an overflow page or a true bucket, the bucket
17 : * number, and the block numbers of the preceding and following pages
18 : * in the same bucket.
19 : *
20 : * The first page in a hash relation, page zero, is special -- it stores
21 : * information describing the hash table; it is referred to as the
22 : * "meta page." Pages one and higher store the actual data.
23 : *
24 : * There are also bitmap pages, which are not manipulated here;
25 : * see hashovfl.c.
26 : *
27 : *-------------------------------------------------------------------------
28 : */
29 : #include "postgres.h"
30 :
31 : #include "access/hash.h"
32 : #include "access/hash_xlog.h"
33 : #include "access/xloginsert.h"
34 : #include "miscadmin.h"
35 : #include "port/pg_bitutils.h"
36 : #include "storage/predicate.h"
37 : #include "storage/smgr.h"
38 : #include "utils/rel.h"
39 :
40 : static bool _hash_alloc_buckets(Relation rel, BlockNumber firstblock,
41 : uint32 nblocks);
42 : static void _hash_splitbucket(Relation rel, Buffer metabuf,
43 : Bucket obucket, Bucket nbucket,
44 : Buffer obuf,
45 : Buffer nbuf,
46 : HTAB *htab,
47 : uint32 maxbucket,
48 : uint32 highmask, uint32 lowmask);
49 : static void log_split_page(Relation rel, Buffer buf);
50 :
51 :
52 : /*
53 : * _hash_getbuf() -- Get a buffer by block number for read or write.
54 : *
55 : * 'access' must be HASH_READ, HASH_WRITE, or HASH_NOLOCK.
56 : * 'flags' is a bitwise OR of the allowed page types.
57 : *
58 : * This must be used only to fetch pages that are expected to be valid
59 : * already. _hash_checkpage() is applied using the given flags.
60 : *
61 : * When this routine returns, the appropriate lock is set on the
62 : * requested buffer and its reference count has been incremented
63 : * (ie, the buffer is "locked and pinned").
64 : *
65 : * P_NEW is disallowed because this routine can only be used
66 : * to access pages that are known to be before the filesystem EOF.
67 : * Extending the index should be done with _hash_getnewbuf.
68 : */
69 : Buffer
70 1880586 : _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
71 : {
72 : Buffer buf;
73 :
74 1880586 : if (blkno == P_NEW)
75 0 : elog(ERROR, "hash AM does not use P_NEW");
76 :
77 1880586 : buf = ReadBuffer(rel, blkno);
78 :
79 1880586 : if (access != HASH_NOLOCK)
80 1166680 : LockBuffer(buf, access);
81 :
82 : /* ref count and lock type are correct */
83 :
84 1880586 : _hash_checkpage(rel, buf, flags);
85 :
86 1880586 : return buf;
87 : }
88 :
89 : /*
90 : * _hash_getbuf_with_condlock_cleanup() -- Try to get a buffer for cleanup.
91 : *
92 : * We read the page and try to acquire a cleanup lock. If we get it,
93 : * we return the buffer; otherwise, we return InvalidBuffer.
94 : */
95 : Buffer
96 1098 : _hash_getbuf_with_condlock_cleanup(Relation rel, BlockNumber blkno, int flags)
97 : {
98 : Buffer buf;
99 :
100 1098 : if (blkno == P_NEW)
101 0 : elog(ERROR, "hash AM does not use P_NEW");
102 :
103 1098 : buf = ReadBuffer(rel, blkno);
104 :
105 1098 : if (!ConditionalLockBufferForCleanup(buf))
106 : {
107 0 : ReleaseBuffer(buf);
108 0 : return InvalidBuffer;
109 : }
110 :
111 : /* ref count and lock type are correct */
112 :
113 1098 : _hash_checkpage(rel, buf, flags);
114 :
115 1098 : return buf;
116 : }
117 :
118 : /*
119 : * _hash_getinitbuf() -- Get and initialize a buffer by block number.
120 : *
121 : * This must be used only to fetch pages that are known to be before
122 : * the index's filesystem EOF, but are to be filled from scratch.
123 : * _hash_pageinit() is applied automatically. Otherwise it has
124 : * effects similar to _hash_getbuf() with access = HASH_WRITE.
125 : *
126 : * When this routine returns, a write lock is set on the
127 : * requested buffer and its reference count has been incremented
128 : * (ie, the buffer is "locked and pinned").
129 : *
130 : * P_NEW is disallowed because this routine can only be used
131 : * to access pages that are known to be before the filesystem EOF.
132 : * Extending the index should be done with _hash_getnewbuf.
133 : */
134 : Buffer
135 66 : _hash_getinitbuf(Relation rel, BlockNumber blkno)
136 : {
137 : Buffer buf;
138 :
139 66 : if (blkno == P_NEW)
140 0 : elog(ERROR, "hash AM does not use P_NEW");
141 :
142 66 : buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_ZERO_AND_LOCK,
143 : NULL);
144 :
145 : /* ref count and lock type are correct */
146 :
147 : /* initialize the page */
148 66 : _hash_pageinit(BufferGetPage(buf), BufferGetPageSize(buf));
149 :
150 66 : return buf;
151 : }
152 :
153 : /*
154 : * _hash_initbuf() -- Get and initialize a buffer by bucket number.
155 : */
156 : void
157 8160 : _hash_initbuf(Buffer buf, uint32 max_bucket, uint32 num_bucket, uint32 flag,
158 : bool initpage)
159 : {
160 : HashPageOpaque pageopaque;
161 : Page page;
162 :
163 8160 : page = BufferGetPage(buf);
164 :
165 : /* initialize the page */
166 8160 : if (initpage)
167 308 : _hash_pageinit(page, BufferGetPageSize(buf));
168 :
169 8160 : pageopaque = HashPageGetOpaque(page);
170 :
171 : /*
172 : * Set hasho_prevblkno with current hashm_maxbucket. This value will be
173 : * used to validate cached HashMetaPageData. See
174 : * _hash_getbucketbuf_from_hashkey().
175 : */
176 8160 : pageopaque->hasho_prevblkno = max_bucket;
177 8160 : pageopaque->hasho_nextblkno = InvalidBlockNumber;
178 8160 : pageopaque->hasho_bucket = num_bucket;
179 8160 : pageopaque->hasho_flag = flag;
180 8160 : pageopaque->hasho_page_id = HASHO_PAGE_ID;
181 8160 : }
182 :
183 : /*
184 : * _hash_getnewbuf() -- Get a new page at the end of the index.
185 : *
186 : * This has the same API as _hash_getinitbuf, except that we are adding
187 : * a page to the index, and hence expect the page to be past the
188 : * logical EOF. (However, we have to support the case where it isn't,
189 : * since a prior try might have crashed after extending the filesystem
190 : * EOF but before updating the metapage to reflect the added page.)
191 : *
192 : * It is caller's responsibility to ensure that only one process can
193 : * extend the index at a time. In practice, this function is called
194 : * only while holding write lock on the metapage, because adding a page
195 : * is always associated with an update of metapage data.
196 : */
197 : Buffer
198 9902 : _hash_getnewbuf(Relation rel, BlockNumber blkno, ForkNumber forkNum)
199 : {
200 9902 : BlockNumber nblocks = RelationGetNumberOfBlocksInFork(rel, forkNum);
201 : Buffer buf;
202 :
203 9902 : if (blkno == P_NEW)
204 0 : elog(ERROR, "hash AM does not use P_NEW");
205 9902 : if (blkno > nblocks)
206 0 : elog(ERROR, "access to noncontiguous page in hash index \"%s\"",
207 : RelationGetRelationName(rel));
208 :
209 : /* smgr insists we explicitly extend the relation */
210 9902 : if (blkno == nblocks)
211 : {
212 8804 : buf = ExtendBufferedRel(BMR_REL(rel), forkNum, NULL,
213 : EB_LOCK_FIRST | EB_SKIP_EXTENSION_LOCK);
214 8804 : if (BufferGetBlockNumber(buf) != blkno)
215 0 : elog(ERROR, "unexpected hash relation size: %u, should be %u",
216 : BufferGetBlockNumber(buf), blkno);
217 : }
218 : else
219 : {
220 1098 : buf = ReadBufferExtended(rel, forkNum, blkno, RBM_ZERO_AND_LOCK,
221 : NULL);
222 : }
223 :
224 : /* ref count and lock type are correct */
225 :
226 : /* initialize the page */
227 9902 : _hash_pageinit(BufferGetPage(buf), BufferGetPageSize(buf));
228 :
229 9902 : return buf;
230 : }
231 :
232 : /*
233 : * _hash_getbuf_with_strategy() -- Get a buffer with nondefault strategy.
234 : *
235 : * This is identical to _hash_getbuf() but also allows a buffer access
236 : * strategy to be specified. We use this for VACUUM operations.
237 : */
238 : Buffer
239 1306 : _hash_getbuf_with_strategy(Relation rel, BlockNumber blkno,
240 : int access, int flags,
241 : BufferAccessStrategy bstrategy)
242 : {
243 : Buffer buf;
244 :
245 1306 : if (blkno == P_NEW)
246 0 : elog(ERROR, "hash AM does not use P_NEW");
247 :
248 1306 : buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
249 :
250 1306 : if (access != HASH_NOLOCK)
251 1306 : LockBuffer(buf, access);
252 :
253 : /* ref count and lock type are correct */
254 :
255 1306 : _hash_checkpage(rel, buf, flags);
256 :
257 1306 : return buf;
258 : }
259 :
260 : /*
261 : * _hash_relbuf() -- release a locked buffer.
262 : *
263 : * Lock and pin (refcount) are both dropped.
264 : */
265 : void
266 1092816 : _hash_relbuf(Relation rel, Buffer buf)
267 : {
268 1092816 : UnlockReleaseBuffer(buf);
269 1092816 : }
270 :
271 : /*
272 : * _hash_dropbuf() -- release an unlocked buffer.
273 : *
274 : * This is used to unpin a buffer on which we hold no lock.
275 : */
276 : void
277 800534 : _hash_dropbuf(Relation rel, Buffer buf)
278 : {
279 800534 : ReleaseBuffer(buf);
280 800534 : }
281 :
282 : /*
283 : * _hash_dropscanbuf() -- release buffers used in scan.
284 : *
285 : * This routine unpins the buffers used during scan on which we
286 : * hold no lock.
287 : */
288 : void
289 1418 : _hash_dropscanbuf(Relation rel, HashScanOpaque so)
290 : {
291 : /* release pin we hold on primary bucket page */
292 1418 : if (BufferIsValid(so->hashso_bucket_buf) &&
293 598 : so->hashso_bucket_buf != so->currPos.buf)
294 156 : _hash_dropbuf(rel, so->hashso_bucket_buf);
295 1418 : so->hashso_bucket_buf = InvalidBuffer;
296 :
297 : /* release pin we hold on primary bucket page of bucket being split */
298 1418 : if (BufferIsValid(so->hashso_split_bucket_buf) &&
299 0 : so->hashso_split_bucket_buf != so->currPos.buf)
300 0 : _hash_dropbuf(rel, so->hashso_split_bucket_buf);
301 1418 : so->hashso_split_bucket_buf = InvalidBuffer;
302 :
303 : /* release any pin we still hold */
304 1418 : if (BufferIsValid(so->currPos.buf))
305 442 : _hash_dropbuf(rel, so->currPos.buf);
306 1418 : so->currPos.buf = InvalidBuffer;
307 :
308 : /* reset split scan */
309 1418 : so->hashso_buc_populated = false;
310 1418 : so->hashso_buc_split = false;
311 1418 : }
312 :
313 :
314 : /*
315 : * _hash_init() -- Initialize the metadata page of a hash index,
316 : * the initial buckets, and the initial bitmap page.
317 : *
318 : * The initial number of buckets is dependent on num_tuples, an estimate
319 : * of the number of tuples to be loaded into the index initially. The
320 : * chosen number of buckets is returned.
321 : *
322 : * We are fairly cavalier about locking here, since we know that no one else
323 : * could be accessing this index. In particular the rule about not holding
324 : * multiple buffer locks is ignored.
325 : */
326 : uint32
327 324 : _hash_init(Relation rel, double num_tuples, ForkNumber forkNum)
328 : {
329 : Buffer metabuf;
330 : Buffer buf;
331 : Buffer bitmapbuf;
332 : Page pg;
333 : HashMetaPage metap;
334 : RegProcedure procid;
335 : int32 data_width;
336 : int32 item_width;
337 : int32 ffactor;
338 : uint32 num_buckets;
339 : uint32 i;
340 : bool use_wal;
341 :
342 : /* safety check */
343 324 : if (RelationGetNumberOfBlocksInFork(rel, forkNum) != 0)
344 0 : elog(ERROR, "cannot initialize non-empty hash index \"%s\"",
345 : RelationGetRelationName(rel));
346 :
347 : /*
348 : * WAL log creation of pages if the relation is persistent, or this is the
349 : * init fork. Init forks for unlogged relations always need to be WAL
350 : * logged.
351 : */
352 324 : use_wal = RelationNeedsWAL(rel) || forkNum == INIT_FORKNUM;
353 :
354 : /*
355 : * Determine the target fill factor (in tuples per bucket) for this index.
356 : * The idea is to make the fill factor correspond to pages about as full
357 : * as the user-settable fillfactor parameter says. We can compute it
358 : * exactly since the index datatype (i.e. uint32 hash key) is fixed-width.
359 : */
360 324 : data_width = sizeof(uint32);
361 324 : item_width = MAXALIGN(sizeof(IndexTupleData)) + MAXALIGN(data_width) +
362 : sizeof(ItemIdData); /* include the line pointer */
363 324 : ffactor = HashGetTargetPageUsage(rel) / item_width;
364 : /* keep to a sane range */
365 324 : if (ffactor < 10)
366 0 : ffactor = 10;
367 :
368 324 : procid = index_getprocid(rel, 1, HASHSTANDARD_PROC);
369 :
370 : /*
371 : * We initialize the metapage, the first N bucket pages, and the first
372 : * bitmap page in sequence, using _hash_getnewbuf to cause smgrextend()
373 : * calls to occur. This ensures that the smgr level has the right idea of
374 : * the physical index length.
375 : *
376 : * Critical section not required, because on error the creation of the
377 : * whole relation will be rolled back.
378 : */
379 324 : metabuf = _hash_getnewbuf(rel, HASH_METAPAGE, forkNum);
380 324 : _hash_init_metabuffer(metabuf, num_tuples, procid, ffactor, false);
381 324 : MarkBufferDirty(metabuf);
382 :
383 324 : pg = BufferGetPage(metabuf);
384 324 : metap = HashPageGetMeta(pg);
385 :
386 : /* XLOG stuff */
387 324 : if (use_wal)
388 : {
389 : xl_hash_init_meta_page xlrec;
390 : XLogRecPtr recptr;
391 :
392 196 : xlrec.num_tuples = num_tuples;
393 196 : xlrec.procid = metap->hashm_procid;
394 196 : xlrec.ffactor = metap->hashm_ffactor;
395 :
396 196 : XLogBeginInsert();
397 196 : XLogRegisterData((char *) &xlrec, SizeOfHashInitMetaPage);
398 196 : XLogRegisterBuffer(0, metabuf, REGBUF_WILL_INIT | REGBUF_STANDARD);
399 :
400 196 : recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_META_PAGE);
401 :
402 196 : PageSetLSN(BufferGetPage(metabuf), recptr);
403 : }
404 :
405 324 : num_buckets = metap->hashm_maxbucket + 1;
406 :
407 : /*
408 : * Release buffer lock on the metapage while we initialize buckets.
409 : * Otherwise, we'll be in interrupt holdoff and the CHECK_FOR_INTERRUPTS
410 : * won't accomplish anything. It's a bad idea to hold buffer locks for
411 : * long intervals in any case, since that can block the bgwriter.
412 : */
413 324 : LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
414 :
415 : /*
416 : * Initialize and WAL Log the first N buckets
417 : */
418 8176 : for (i = 0; i < num_buckets; i++)
419 : {
420 : BlockNumber blkno;
421 :
422 : /* Allow interrupts, in case N is huge */
423 7852 : CHECK_FOR_INTERRUPTS();
424 :
425 7852 : blkno = BUCKET_TO_BLKNO(metap, i);
426 7852 : buf = _hash_getnewbuf(rel, blkno, forkNum);
427 7852 : _hash_initbuf(buf, metap->hashm_maxbucket, i, LH_BUCKET_PAGE, false);
428 7852 : MarkBufferDirty(buf);
429 :
430 7852 : if (use_wal)
431 5304 : log_newpage(&rel->rd_locator,
432 : forkNum,
433 : blkno,
434 : BufferGetPage(buf),
435 : true);
436 7852 : _hash_relbuf(rel, buf);
437 : }
438 :
439 : /* Now reacquire buffer lock on metapage */
440 324 : LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
441 :
442 : /*
443 : * Initialize bitmap page
444 : */
445 324 : bitmapbuf = _hash_getnewbuf(rel, num_buckets + 1, forkNum);
446 324 : _hash_initbitmapbuffer(bitmapbuf, metap->hashm_bmsize, false);
447 324 : MarkBufferDirty(bitmapbuf);
448 :
449 : /* add the new bitmap page to the metapage's list of bitmaps */
450 : /* metapage already has a write lock */
451 324 : if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
452 0 : ereport(ERROR,
453 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
454 : errmsg("out of overflow pages in hash index \"%s\"",
455 : RelationGetRelationName(rel))));
456 :
457 324 : metap->hashm_mapp[metap->hashm_nmaps] = num_buckets + 1;
458 :
459 324 : metap->hashm_nmaps++;
460 324 : MarkBufferDirty(metabuf);
461 :
462 : /* XLOG stuff */
463 324 : if (use_wal)
464 : {
465 : xl_hash_init_bitmap_page xlrec;
466 : XLogRecPtr recptr;
467 :
468 196 : xlrec.bmsize = metap->hashm_bmsize;
469 :
470 196 : XLogBeginInsert();
471 196 : XLogRegisterData((char *) &xlrec, SizeOfHashInitBitmapPage);
472 196 : XLogRegisterBuffer(0, bitmapbuf, REGBUF_WILL_INIT);
473 :
474 : /*
475 : * This is safe only because nobody else can be modifying the index at
476 : * this stage; it's only visible to the transaction that is creating
477 : * it.
478 : */
479 196 : XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
480 :
481 196 : recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_BITMAP_PAGE);
482 :
483 196 : PageSetLSN(BufferGetPage(bitmapbuf), recptr);
484 196 : PageSetLSN(BufferGetPage(metabuf), recptr);
485 : }
486 :
487 : /* all done */
488 324 : _hash_relbuf(rel, bitmapbuf);
489 324 : _hash_relbuf(rel, metabuf);
490 :
491 324 : return num_buckets;
492 : }
493 :
494 : /*
495 : * _hash_init_metabuffer() -- Initialize the metadata page of a hash index.
496 : */
497 : void
498 372 : _hash_init_metabuffer(Buffer buf, double num_tuples, RegProcedure procid,
499 : uint16 ffactor, bool initpage)
500 : {
501 : HashMetaPage metap;
502 : HashPageOpaque pageopaque;
503 : Page page;
504 : double dnumbuckets;
505 : uint32 num_buckets;
506 : uint32 spare_index;
507 : uint32 lshift;
508 :
509 : /*
510 : * Choose the number of initial bucket pages to match the fill factor
511 : * given the estimated number of tuples. We round up the result to the
512 : * total number of buckets which has to be allocated before using its
513 : * hashm_spares element. However always force at least 2 bucket pages. The
514 : * upper limit is determined by considerations explained in
515 : * _hash_expandtable().
516 : */
517 372 : dnumbuckets = num_tuples / ffactor;
518 372 : if (dnumbuckets <= 2.0)
519 110 : num_buckets = 2;
520 262 : else if (dnumbuckets >= (double) 0x40000000)
521 0 : num_buckets = 0x40000000;
522 : else
523 262 : num_buckets = _hash_get_totalbuckets(_hash_spareindex(dnumbuckets));
524 :
525 372 : spare_index = _hash_spareindex(num_buckets);
526 : Assert(spare_index < HASH_MAX_SPLITPOINTS);
527 :
528 372 : page = BufferGetPage(buf);
529 372 : if (initpage)
530 48 : _hash_pageinit(page, BufferGetPageSize(buf));
531 :
532 372 : pageopaque = HashPageGetOpaque(page);
533 372 : pageopaque->hasho_prevblkno = InvalidBlockNumber;
534 372 : pageopaque->hasho_nextblkno = InvalidBlockNumber;
535 372 : pageopaque->hasho_bucket = InvalidBucket;
536 372 : pageopaque->hasho_flag = LH_META_PAGE;
537 372 : pageopaque->hasho_page_id = HASHO_PAGE_ID;
538 :
539 372 : metap = HashPageGetMeta(page);
540 :
541 372 : metap->hashm_magic = HASH_MAGIC;
542 372 : metap->hashm_version = HASH_VERSION;
543 372 : metap->hashm_ntuples = 0;
544 372 : metap->hashm_nmaps = 0;
545 372 : metap->hashm_ffactor = ffactor;
546 372 : metap->hashm_bsize = HashGetMaxBitmapSize(page);
547 :
548 : /* find largest bitmap array size that will fit in page size */
549 372 : lshift = pg_leftmost_one_pos32(metap->hashm_bsize);
550 : Assert(lshift > 0);
551 372 : metap->hashm_bmsize = 1 << lshift;
552 372 : metap->hashm_bmshift = lshift + BYTE_TO_BIT;
553 : Assert((1 << BMPG_SHIFT(metap)) == (BMPG_MASK(metap) + 1));
554 :
555 : /*
556 : * Label the index with its primary hash support function's OID. This is
557 : * pretty useless for normal operation (in fact, hashm_procid is not used
558 : * anywhere), but it might be handy for forensic purposes so we keep it.
559 : */
560 372 : metap->hashm_procid = procid;
561 :
562 : /*
563 : * We initialize the index with N buckets, 0 .. N-1, occupying physical
564 : * blocks 1 to N. The first freespace bitmap page is in block N+1.
565 : */
566 372 : metap->hashm_maxbucket = num_buckets - 1;
567 :
568 : /*
569 : * Set highmask as next immediate ((2 ^ x) - 1), which should be
570 : * sufficient to cover num_buckets.
571 : */
572 372 : metap->hashm_highmask = pg_nextpower2_32(num_buckets + 1) - 1;
573 372 : metap->hashm_lowmask = (metap->hashm_highmask >> 1);
574 :
575 372 : MemSet(metap->hashm_spares, 0, sizeof(metap->hashm_spares));
576 372 : MemSet(metap->hashm_mapp, 0, sizeof(metap->hashm_mapp));
577 :
578 : /* Set up mapping for one spare page after the initial splitpoints */
579 372 : metap->hashm_spares[spare_index] = 1;
580 372 : metap->hashm_ovflpoint = spare_index;
581 372 : metap->hashm_firstfree = 0;
582 :
583 : /*
584 : * Set pd_lower just past the end of the metadata. This is essential,
585 : * because without doing so, metadata will be lost if xlog.c compresses
586 : * the page.
587 : */
588 372 : ((PageHeader) page)->pd_lower =
589 372 : ((char *) metap + sizeof(HashMetaPageData)) - (char *) page;
590 372 : }
591 :
592 : /*
593 : * _hash_pageinit() -- Initialize a new hash index page.
594 : */
595 : void
596 10642 : _hash_pageinit(Page page, Size size)
597 : {
598 10642 : PageInit(page, size, sizeof(HashPageOpaqueData));
599 10642 : }
600 :
601 : /*
602 : * Attempt to expand the hash table by creating one new bucket.
603 : *
604 : * This will silently do nothing if we don't get cleanup lock on old or
605 : * new bucket.
606 : *
607 : * Complete the pending splits and remove the tuples from old bucket,
608 : * if there are any left over from the previous split.
609 : *
610 : * The caller must hold a pin, but no lock, on the metapage buffer.
611 : * The buffer is returned in the same state.
612 : */
613 : void
614 1098 : _hash_expandtable(Relation rel, Buffer metabuf)
615 : {
616 : HashMetaPage metap;
617 : Bucket old_bucket;
618 : Bucket new_bucket;
619 : uint32 spare_ndx;
620 : BlockNumber start_oblkno;
621 : BlockNumber start_nblkno;
622 : Buffer buf_nblkno;
623 : Buffer buf_oblkno;
624 : Page opage;
625 : Page npage;
626 : HashPageOpaque oopaque;
627 : HashPageOpaque nopaque;
628 : uint32 maxbucket;
629 : uint32 highmask;
630 : uint32 lowmask;
631 1098 : bool metap_update_masks = false;
632 1098 : bool metap_update_splitpoint = false;
633 :
634 1098 : restart_expand:
635 :
636 : /*
637 : * Write-lock the meta page. It used to be necessary to acquire a
638 : * heavyweight lock to begin a split, but that is no longer required.
639 : */
640 1098 : LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
641 :
642 1098 : _hash_checkpage(rel, metabuf, LH_META_PAGE);
643 1098 : metap = HashPageGetMeta(BufferGetPage(metabuf));
644 :
645 : /*
646 : * Check to see if split is still needed; someone else might have already
647 : * done one while we waited for the lock.
648 : *
649 : * Make sure this stays in sync with _hash_doinsert()
650 : */
651 1098 : if (metap->hashm_ntuples <=
652 1098 : (double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1))
653 0 : goto fail;
654 :
655 : /*
656 : * Can't split anymore if maxbucket has reached its maximum possible
657 : * value.
658 : *
659 : * Ideally we'd allow bucket numbers up to UINT_MAX-1 (no higher because
660 : * the calculation maxbucket+1 mustn't overflow). Currently we restrict
661 : * to half that to prevent failure of pg_ceil_log2_32() and insufficient
662 : * space in hashm_spares[]. It's moot anyway because an index with 2^32
663 : * buckets would certainly overflow BlockNumber and hence
664 : * _hash_alloc_buckets() would fail, but if we supported buckets smaller
665 : * than a disk block then this would be an independent constraint.
666 : *
667 : * If you change this, see also the maximum initial number of buckets in
668 : * _hash_init().
669 : */
670 1098 : if (metap->hashm_maxbucket >= (uint32) 0x7FFFFFFE)
671 0 : goto fail;
672 :
673 : /*
674 : * Determine which bucket is to be split, and attempt to take cleanup lock
675 : * on the old bucket. If we can't get the lock, give up.
676 : *
677 : * The cleanup lock protects us not only against other backends, but
678 : * against our own backend as well.
679 : *
680 : * The cleanup lock is mainly to protect the split from concurrent
681 : * inserts. See src/backend/access/hash/README, Lock Definitions for
682 : * further details. Due to this locking restriction, if there is any
683 : * pending scan, the split will give up which is not good, but harmless.
684 : */
685 1098 : new_bucket = metap->hashm_maxbucket + 1;
686 :
687 1098 : old_bucket = (new_bucket & metap->hashm_lowmask);
688 :
689 1098 : start_oblkno = BUCKET_TO_BLKNO(metap, old_bucket);
690 :
691 1098 : buf_oblkno = _hash_getbuf_with_condlock_cleanup(rel, start_oblkno, LH_BUCKET_PAGE);
692 1098 : if (!buf_oblkno)
693 0 : goto fail;
694 :
695 1098 : opage = BufferGetPage(buf_oblkno);
696 1098 : oopaque = HashPageGetOpaque(opage);
697 :
698 : /*
699 : * We want to finish the split from a bucket as there is no apparent
700 : * benefit by not doing so and it will make the code complicated to finish
701 : * the split that involves multiple buckets considering the case where new
702 : * split also fails. We don't need to consider the new bucket for
703 : * completing the split here as it is not possible that a re-split of new
704 : * bucket starts when there is still a pending split from old bucket.
705 : */
706 1098 : if (H_BUCKET_BEING_SPLIT(oopaque))
707 : {
708 : /*
709 : * Copy bucket mapping info now; refer the comment in code below where
710 : * we copy this information before calling _hash_splitbucket to see
711 : * why this is okay.
712 : */
713 0 : maxbucket = metap->hashm_maxbucket;
714 0 : highmask = metap->hashm_highmask;
715 0 : lowmask = metap->hashm_lowmask;
716 :
717 : /*
718 : * Release the lock on metapage and old_bucket, before completing the
719 : * split.
720 : */
721 0 : LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
722 0 : LockBuffer(buf_oblkno, BUFFER_LOCK_UNLOCK);
723 :
724 0 : _hash_finish_split(rel, metabuf, buf_oblkno, old_bucket, maxbucket,
725 : highmask, lowmask);
726 :
727 : /* release the pin on old buffer and retry for expand. */
728 0 : _hash_dropbuf(rel, buf_oblkno);
729 :
730 0 : goto restart_expand;
731 : }
732 :
733 : /*
734 : * Clean the tuples remained from the previous split. This operation
735 : * requires cleanup lock and we already have one on the old bucket, so
736 : * let's do it. We also don't want to allow further splits from the bucket
737 : * till the garbage of previous split is cleaned. This has two
738 : * advantages; first, it helps in avoiding the bloat due to garbage and
739 : * second is, during cleanup of bucket, we are always sure that the
740 : * garbage tuples belong to most recently split bucket. On the contrary,
741 : * if we allow cleanup of bucket after meta page is updated to indicate
742 : * the new split and before the actual split, the cleanup operation won't
743 : * be able to decide whether the tuple has been moved to the newly created
744 : * bucket and ended up deleting such tuples.
745 : */
746 1098 : if (H_NEEDS_SPLIT_CLEANUP(oopaque))
747 : {
748 : /*
749 : * Copy bucket mapping info now; refer to the comment in code below
750 : * where we copy this information before calling _hash_splitbucket to
751 : * see why this is okay.
752 : */
753 0 : maxbucket = metap->hashm_maxbucket;
754 0 : highmask = metap->hashm_highmask;
755 0 : lowmask = metap->hashm_lowmask;
756 :
757 : /* Release the metapage lock. */
758 0 : LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
759 :
760 0 : hashbucketcleanup(rel, old_bucket, buf_oblkno, start_oblkno, NULL,
761 : maxbucket, highmask, lowmask, NULL, NULL, true,
762 : NULL, NULL);
763 :
764 0 : _hash_dropbuf(rel, buf_oblkno);
765 :
766 0 : goto restart_expand;
767 : }
768 :
769 : /*
770 : * There shouldn't be any active scan on new bucket.
771 : *
772 : * Note: it is safe to compute the new bucket's blkno here, even though we
773 : * may still need to update the BUCKET_TO_BLKNO mapping. This is because
774 : * the current value of hashm_spares[hashm_ovflpoint] correctly shows
775 : * where we are going to put a new splitpoint's worth of buckets.
776 : */
777 1098 : start_nblkno = BUCKET_TO_BLKNO(metap, new_bucket);
778 :
779 : /*
780 : * If the split point is increasing we need to allocate a new batch of
781 : * bucket pages.
782 : */
783 1098 : spare_ndx = _hash_spareindex(new_bucket + 1);
784 1098 : if (spare_ndx > metap->hashm_ovflpoint)
785 : {
786 : uint32 buckets_to_add;
787 :
788 : Assert(spare_ndx == metap->hashm_ovflpoint + 1);
789 :
790 : /*
791 : * We treat allocation of buckets as a separate WAL-logged action.
792 : * Even if we fail after this operation, won't leak bucket pages;
793 : * rather, the next split will consume this space. In any case, even
794 : * without failure we don't use all the space in one split operation.
795 : */
796 58 : buckets_to_add = _hash_get_totalbuckets(spare_ndx) - new_bucket;
797 58 : if (!_hash_alloc_buckets(rel, start_nblkno, buckets_to_add))
798 : {
799 : /* can't split due to BlockNumber overflow */
800 0 : _hash_relbuf(rel, buf_oblkno);
801 0 : goto fail;
802 : }
803 : }
804 :
805 : /*
806 : * Physically allocate the new bucket's primary page. We want to do this
807 : * before changing the metapage's mapping info, in case we can't get the
808 : * disk space.
809 : *
810 : * XXX It doesn't make sense to call _hash_getnewbuf first, zeroing the
811 : * buffer, and then only afterwards check whether we have a cleanup lock.
812 : * However, since no scan can be accessing the buffer yet, any concurrent
813 : * accesses will just be from processes like the bgwriter or checkpointer
814 : * which don't care about its contents, so it doesn't really matter.
815 : */
816 1098 : buf_nblkno = _hash_getnewbuf(rel, start_nblkno, MAIN_FORKNUM);
817 1098 : if (!IsBufferCleanupOK(buf_nblkno))
818 : {
819 0 : _hash_relbuf(rel, buf_oblkno);
820 0 : _hash_relbuf(rel, buf_nblkno);
821 0 : goto fail;
822 : }
823 :
824 : /*
825 : * Since we are scribbling on the pages in the shared buffers, establish a
826 : * critical section. Any failure in this next code leaves us with a big
827 : * problem: the metapage is effectively corrupt but could get written back
828 : * to disk.
829 : */
830 1098 : START_CRIT_SECTION();
831 :
832 : /*
833 : * Okay to proceed with split. Update the metapage bucket mapping info.
834 : */
835 1098 : metap->hashm_maxbucket = new_bucket;
836 :
837 1098 : if (new_bucket > metap->hashm_highmask)
838 : {
839 : /* Starting a new doubling */
840 24 : metap->hashm_lowmask = metap->hashm_highmask;
841 24 : metap->hashm_highmask = new_bucket | metap->hashm_lowmask;
842 24 : metap_update_masks = true;
843 : }
844 :
845 : /*
846 : * If the split point is increasing we need to adjust the hashm_spares[]
847 : * array and hashm_ovflpoint so that future overflow pages will be created
848 : * beyond this new batch of bucket pages.
849 : */
850 1098 : if (spare_ndx > metap->hashm_ovflpoint)
851 : {
852 58 : metap->hashm_spares[spare_ndx] = metap->hashm_spares[metap->hashm_ovflpoint];
853 58 : metap->hashm_ovflpoint = spare_ndx;
854 58 : metap_update_splitpoint = true;
855 : }
856 :
857 1098 : MarkBufferDirty(metabuf);
858 :
859 : /*
860 : * Copy bucket mapping info now; this saves re-accessing the meta page
861 : * inside _hash_splitbucket's inner loop. Note that once we drop the
862 : * split lock, other splits could begin, so these values might be out of
863 : * date before _hash_splitbucket finishes. That's okay, since all it
864 : * needs is to tell which of these two buckets to map hashkeys into.
865 : */
866 1098 : maxbucket = metap->hashm_maxbucket;
867 1098 : highmask = metap->hashm_highmask;
868 1098 : lowmask = metap->hashm_lowmask;
869 :
870 1098 : opage = BufferGetPage(buf_oblkno);
871 1098 : oopaque = HashPageGetOpaque(opage);
872 :
873 : /*
874 : * Mark the old bucket to indicate that split is in progress. (At
875 : * operation end, we will clear the split-in-progress flag.) Also, for a
876 : * primary bucket page, hasho_prevblkno stores the number of buckets that
877 : * existed as of the last split, so we must update that value here.
878 : */
879 1098 : oopaque->hasho_flag |= LH_BUCKET_BEING_SPLIT;
880 1098 : oopaque->hasho_prevblkno = maxbucket;
881 :
882 1098 : MarkBufferDirty(buf_oblkno);
883 :
884 1098 : npage = BufferGetPage(buf_nblkno);
885 :
886 : /*
887 : * initialize the new bucket's primary page and mark it to indicate that
888 : * split is in progress.
889 : */
890 1098 : nopaque = HashPageGetOpaque(npage);
891 1098 : nopaque->hasho_prevblkno = maxbucket;
892 1098 : nopaque->hasho_nextblkno = InvalidBlockNumber;
893 1098 : nopaque->hasho_bucket = new_bucket;
894 1098 : nopaque->hasho_flag = LH_BUCKET_PAGE | LH_BUCKET_BEING_POPULATED;
895 1098 : nopaque->hasho_page_id = HASHO_PAGE_ID;
896 :
897 1098 : MarkBufferDirty(buf_nblkno);
898 :
899 : /* XLOG stuff */
900 1098 : if (RelationNeedsWAL(rel))
901 : {
902 : xl_hash_split_allocate_page xlrec;
903 : XLogRecPtr recptr;
904 :
905 846 : xlrec.new_bucket = maxbucket;
906 846 : xlrec.old_bucket_flag = oopaque->hasho_flag;
907 846 : xlrec.new_bucket_flag = nopaque->hasho_flag;
908 846 : xlrec.flags = 0;
909 :
910 846 : XLogBeginInsert();
911 :
912 846 : XLogRegisterBuffer(0, buf_oblkno, REGBUF_STANDARD);
913 846 : XLogRegisterBuffer(1, buf_nblkno, REGBUF_WILL_INIT);
914 846 : XLogRegisterBuffer(2, metabuf, REGBUF_STANDARD);
915 :
916 846 : if (metap_update_masks)
917 : {
918 24 : xlrec.flags |= XLH_SPLIT_META_UPDATE_MASKS;
919 24 : XLogRegisterBufData(2, (char *) &metap->hashm_lowmask, sizeof(uint32));
920 24 : XLogRegisterBufData(2, (char *) &metap->hashm_highmask, sizeof(uint32));
921 : }
922 :
923 846 : if (metap_update_splitpoint)
924 : {
925 50 : xlrec.flags |= XLH_SPLIT_META_UPDATE_SPLITPOINT;
926 50 : XLogRegisterBufData(2, (char *) &metap->hashm_ovflpoint,
927 : sizeof(uint32));
928 50 : XLogRegisterBufData(2,
929 50 : (char *) &metap->hashm_spares[metap->hashm_ovflpoint],
930 : sizeof(uint32));
931 : }
932 :
933 846 : XLogRegisterData((char *) &xlrec, SizeOfHashSplitAllocPage);
934 :
935 846 : recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_ALLOCATE_PAGE);
936 :
937 846 : PageSetLSN(BufferGetPage(buf_oblkno), recptr);
938 846 : PageSetLSN(BufferGetPage(buf_nblkno), recptr);
939 846 : PageSetLSN(BufferGetPage(metabuf), recptr);
940 : }
941 :
942 1098 : END_CRIT_SECTION();
943 :
944 : /* drop lock, but keep pin */
945 1098 : LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
946 :
947 : /* Relocate records to the new bucket */
948 1098 : _hash_splitbucket(rel, metabuf,
949 : old_bucket, new_bucket,
950 : buf_oblkno, buf_nblkno, NULL,
951 : maxbucket, highmask, lowmask);
952 :
953 : /* all done, now release the pins on primary buckets. */
954 1098 : _hash_dropbuf(rel, buf_oblkno);
955 1098 : _hash_dropbuf(rel, buf_nblkno);
956 :
957 1098 : return;
958 :
959 : /* Here if decide not to split or fail to acquire old bucket lock */
960 0 : fail:
961 :
962 : /* We didn't write the metapage, so just drop lock */
963 0 : LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
964 : }
965 :
966 :
967 : /*
968 : * _hash_alloc_buckets -- allocate a new splitpoint's worth of bucket pages
969 : *
970 : * This does not need to initialize the new bucket pages; we'll do that as
971 : * each one is used by _hash_expandtable(). But we have to extend the logical
972 : * EOF to the end of the splitpoint; this keeps smgr's idea of the EOF in
973 : * sync with ours, so that we don't get complaints from smgr.
974 : *
975 : * We do this by writing a page of zeroes at the end of the splitpoint range.
976 : * We expect that the filesystem will ensure that the intervening pages read
977 : * as zeroes too. On many filesystems this "hole" will not be allocated
978 : * immediately, which means that the index file may end up more fragmented
979 : * than if we forced it all to be allocated now; but since we don't scan
980 : * hash indexes sequentially anyway, that probably doesn't matter.
981 : *
982 : * XXX It's annoying that this code is executed with the metapage lock held.
983 : * We need to interlock against _hash_addovflpage() adding a new overflow page
984 : * concurrently, but it'd likely be better to use LockRelationForExtension
985 : * for the purpose. OTOH, adding a splitpoint is a very infrequent operation,
986 : * so it may not be worth worrying about.
987 : *
988 : * Returns true if successful, or false if allocation failed due to
989 : * BlockNumber overflow.
990 : */
991 : static bool
992 58 : _hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
993 : {
994 : BlockNumber lastblock;
995 : PGIOAlignedBlock zerobuf;
996 : Page page;
997 : HashPageOpaque ovflopaque;
998 :
999 58 : lastblock = firstblock + nblocks - 1;
1000 :
1001 : /*
1002 : * Check for overflow in block number calculation; if so, we cannot extend
1003 : * the index anymore.
1004 : */
1005 58 : if (lastblock < firstblock || lastblock == InvalidBlockNumber)
1006 0 : return false;
1007 :
1008 58 : page = (Page) zerobuf.data;
1009 :
1010 : /*
1011 : * Initialize the page. Just zeroing the page won't work; see
1012 : * _hash_freeovflpage for similar usage. We take care to make the special
1013 : * space valid for the benefit of tools such as pageinspect.
1014 : */
1015 58 : _hash_pageinit(page, BLCKSZ);
1016 :
1017 58 : ovflopaque = HashPageGetOpaque(page);
1018 :
1019 58 : ovflopaque->hasho_prevblkno = InvalidBlockNumber;
1020 58 : ovflopaque->hasho_nextblkno = InvalidBlockNumber;
1021 58 : ovflopaque->hasho_bucket = InvalidBucket;
1022 58 : ovflopaque->hasho_flag = LH_UNUSED_PAGE;
1023 58 : ovflopaque->hasho_page_id = HASHO_PAGE_ID;
1024 :
1025 58 : if (RelationNeedsWAL(rel))
1026 50 : log_newpage(&rel->rd_locator,
1027 : MAIN_FORKNUM,
1028 : lastblock,
1029 : zerobuf.data,
1030 : true);
1031 :
1032 58 : PageSetChecksumInplace(page, lastblock);
1033 58 : smgrextend(RelationGetSmgr(rel), MAIN_FORKNUM, lastblock, zerobuf.data,
1034 : false);
1035 :
1036 58 : return true;
1037 : }
1038 :
1039 :
1040 : /*
1041 : * _hash_splitbucket -- split 'obucket' into 'obucket' and 'nbucket'
1042 : *
1043 : * This routine is used to partition the tuples between old and new bucket and
1044 : * is used to finish the incomplete split operations. To finish the previously
1045 : * interrupted split operation, the caller needs to fill htab. If htab is set,
1046 : * then we skip the movement of tuples that exists in htab, otherwise NULL
1047 : * value of htab indicates movement of all the tuples that belong to the new
1048 : * bucket.
1049 : *
1050 : * We are splitting a bucket that consists of a base bucket page and zero
1051 : * or more overflow (bucket chain) pages. We must relocate tuples that
1052 : * belong in the new bucket.
1053 : *
1054 : * The caller must hold cleanup locks on both buckets to ensure that
1055 : * no one else is trying to access them (see README).
1056 : *
1057 : * The caller must hold a pin, but no lock, on the metapage buffer.
1058 : * The buffer is returned in the same state. (The metapage is only
1059 : * touched if it becomes necessary to add or remove overflow pages.)
1060 : *
1061 : * Split needs to retain pin on primary bucket pages of both old and new
1062 : * buckets till end of operation. This is to prevent vacuum from starting
1063 : * while a split is in progress.
1064 : *
1065 : * In addition, the caller must have created the new bucket's base page,
1066 : * which is passed in buffer nbuf, pinned and write-locked. The lock will be
1067 : * released here and pin must be released by the caller. (The API is set up
1068 : * this way because we must do _hash_getnewbuf() before releasing the metapage
1069 : * write lock. So instead of passing the new bucket's start block number, we
1070 : * pass an actual buffer.)
1071 : */
1072 : static void
1073 1098 : _hash_splitbucket(Relation rel,
1074 : Buffer metabuf,
1075 : Bucket obucket,
1076 : Bucket nbucket,
1077 : Buffer obuf,
1078 : Buffer nbuf,
1079 : HTAB *htab,
1080 : uint32 maxbucket,
1081 : uint32 highmask,
1082 : uint32 lowmask)
1083 : {
1084 : Buffer bucket_obuf;
1085 : Buffer bucket_nbuf;
1086 : Page opage;
1087 : Page npage;
1088 : HashPageOpaque oopaque;
1089 : HashPageOpaque nopaque;
1090 : OffsetNumber itup_offsets[MaxIndexTuplesPerPage];
1091 : IndexTuple itups[MaxIndexTuplesPerPage];
1092 1098 : Size all_tups_size = 0;
1093 : int i;
1094 1098 : uint16 nitups = 0;
1095 :
1096 1098 : bucket_obuf = obuf;
1097 1098 : opage = BufferGetPage(obuf);
1098 1098 : oopaque = HashPageGetOpaque(opage);
1099 :
1100 1098 : bucket_nbuf = nbuf;
1101 1098 : npage = BufferGetPage(nbuf);
1102 1098 : nopaque = HashPageGetOpaque(npage);
1103 :
1104 : /* Copy the predicate locks from old bucket to new bucket. */
1105 1098 : PredicateLockPageSplit(rel,
1106 : BufferGetBlockNumber(bucket_obuf),
1107 : BufferGetBlockNumber(bucket_nbuf));
1108 :
1109 : /*
1110 : * Partition the tuples in the old bucket between the old bucket and the
1111 : * new bucket, advancing along the old bucket's overflow bucket chain and
1112 : * adding overflow pages to the new bucket as needed. Outer loop iterates
1113 : * once per page in old bucket.
1114 : */
1115 : for (;;)
1116 342 : {
1117 : BlockNumber oblkno;
1118 : OffsetNumber ooffnum;
1119 : OffsetNumber omaxoffnum;
1120 :
1121 : /* Scan each tuple in old page */
1122 1440 : omaxoffnum = PageGetMaxOffsetNumber(opage);
1123 299208 : for (ooffnum = FirstOffsetNumber;
1124 : ooffnum <= omaxoffnum;
1125 297768 : ooffnum = OffsetNumberNext(ooffnum))
1126 : {
1127 : IndexTuple itup;
1128 : Size itemsz;
1129 : Bucket bucket;
1130 297768 : bool found = false;
1131 :
1132 : /* skip dead tuples */
1133 297768 : if (ItemIdIsDead(PageGetItemId(opage, ooffnum)))
1134 0 : continue;
1135 :
1136 : /*
1137 : * Before inserting a tuple, probe the hash table containing TIDs
1138 : * of tuples belonging to new bucket, if we find a match, then
1139 : * skip that tuple, else fetch the item's hash key (conveniently
1140 : * stored in the item) and determine which bucket it now belongs
1141 : * in.
1142 : */
1143 297768 : itup = (IndexTuple) PageGetItem(opage,
1144 : PageGetItemId(opage, ooffnum));
1145 :
1146 297768 : if (htab)
1147 0 : (void) hash_search(htab, &itup->t_tid, HASH_FIND, &found);
1148 :
1149 297768 : if (found)
1150 0 : continue;
1151 :
1152 297768 : bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
1153 : maxbucket, highmask, lowmask);
1154 :
1155 297768 : if (bucket == nbucket)
1156 : {
1157 : IndexTuple new_itup;
1158 :
1159 : /*
1160 : * make a copy of index tuple as we have to scribble on it.
1161 : */
1162 121100 : new_itup = CopyIndexTuple(itup);
1163 :
1164 : /*
1165 : * mark the index tuple as moved by split, such tuples are
1166 : * skipped by scan if there is split in progress for a bucket.
1167 : */
1168 121100 : new_itup->t_info |= INDEX_MOVED_BY_SPLIT_MASK;
1169 :
1170 : /*
1171 : * insert the tuple into the new bucket. if it doesn't fit on
1172 : * the current page in the new bucket, we must allocate a new
1173 : * overflow page and place the tuple on that page instead.
1174 : */
1175 121100 : itemsz = IndexTupleSize(new_itup);
1176 121100 : itemsz = MAXALIGN(itemsz);
1177 :
1178 121100 : if (PageGetFreeSpaceForMultipleTuples(npage, nitups + 1) < (all_tups_size + itemsz))
1179 : {
1180 : /*
1181 : * Change the shared buffer state in critical section,
1182 : * otherwise any error could make it unrecoverable.
1183 : */
1184 78 : START_CRIT_SECTION();
1185 :
1186 78 : _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1187 78 : MarkBufferDirty(nbuf);
1188 : /* log the split operation before releasing the lock */
1189 78 : log_split_page(rel, nbuf);
1190 :
1191 78 : END_CRIT_SECTION();
1192 :
1193 : /* drop lock, but keep pin */
1194 78 : LockBuffer(nbuf, BUFFER_LOCK_UNLOCK);
1195 :
1196 : /* be tidy */
1197 31824 : for (i = 0; i < nitups; i++)
1198 31746 : pfree(itups[i]);
1199 78 : nitups = 0;
1200 78 : all_tups_size = 0;
1201 :
1202 : /* chain to a new overflow page */
1203 78 : nbuf = _hash_addovflpage(rel, metabuf, nbuf, (nbuf == bucket_nbuf));
1204 78 : npage = BufferGetPage(nbuf);
1205 78 : nopaque = HashPageGetOpaque(npage);
1206 : }
1207 :
1208 121100 : itups[nitups++] = new_itup;
1209 121100 : all_tups_size += itemsz;
1210 : }
1211 : else
1212 : {
1213 : /*
1214 : * the tuple stays on this page, so nothing to do.
1215 : */
1216 : Assert(bucket == obucket);
1217 : }
1218 : }
1219 :
1220 1440 : oblkno = oopaque->hasho_nextblkno;
1221 :
1222 : /* retain the pin on the old primary bucket */
1223 1440 : if (obuf == bucket_obuf)
1224 1098 : LockBuffer(obuf, BUFFER_LOCK_UNLOCK);
1225 : else
1226 342 : _hash_relbuf(rel, obuf);
1227 :
1228 : /* Exit loop if no more overflow pages in old bucket */
1229 1440 : if (!BlockNumberIsValid(oblkno))
1230 : {
1231 : /*
1232 : * Change the shared buffer state in critical section, otherwise
1233 : * any error could make it unrecoverable.
1234 : */
1235 1098 : START_CRIT_SECTION();
1236 :
1237 1098 : _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1238 1098 : MarkBufferDirty(nbuf);
1239 : /* log the split operation before releasing the lock */
1240 1098 : log_split_page(rel, nbuf);
1241 :
1242 1098 : END_CRIT_SECTION();
1243 :
1244 1098 : if (nbuf == bucket_nbuf)
1245 1092 : LockBuffer(nbuf, BUFFER_LOCK_UNLOCK);
1246 : else
1247 6 : _hash_relbuf(rel, nbuf);
1248 :
1249 : /* be tidy */
1250 90452 : for (i = 0; i < nitups; i++)
1251 89354 : pfree(itups[i]);
1252 1098 : break;
1253 : }
1254 :
1255 : /* Else, advance to next old page */
1256 342 : obuf = _hash_getbuf(rel, oblkno, HASH_READ, LH_OVERFLOW_PAGE);
1257 342 : opage = BufferGetPage(obuf);
1258 342 : oopaque = HashPageGetOpaque(opage);
1259 : }
1260 :
1261 : /*
1262 : * We're at the end of the old bucket chain, so we're done partitioning
1263 : * the tuples. Mark the old and new buckets to indicate split is
1264 : * finished.
1265 : *
1266 : * To avoid deadlocks due to locking order of buckets, first lock the old
1267 : * bucket and then the new bucket.
1268 : */
1269 1098 : LockBuffer(bucket_obuf, BUFFER_LOCK_EXCLUSIVE);
1270 1098 : opage = BufferGetPage(bucket_obuf);
1271 1098 : oopaque = HashPageGetOpaque(opage);
1272 :
1273 1098 : LockBuffer(bucket_nbuf, BUFFER_LOCK_EXCLUSIVE);
1274 1098 : npage = BufferGetPage(bucket_nbuf);
1275 1098 : nopaque = HashPageGetOpaque(npage);
1276 :
1277 1098 : START_CRIT_SECTION();
1278 :
1279 1098 : oopaque->hasho_flag &= ~LH_BUCKET_BEING_SPLIT;
1280 1098 : nopaque->hasho_flag &= ~LH_BUCKET_BEING_POPULATED;
1281 :
1282 : /*
1283 : * After the split is finished, mark the old bucket to indicate that it
1284 : * contains deletable tuples. We will clear split-cleanup flag after
1285 : * deleting such tuples either at the end of split or at the next split
1286 : * from old bucket or at the time of vacuum.
1287 : */
1288 1098 : oopaque->hasho_flag |= LH_BUCKET_NEEDS_SPLIT_CLEANUP;
1289 :
1290 : /*
1291 : * now write the buffers, here we don't release the locks as caller is
1292 : * responsible to release locks.
1293 : */
1294 1098 : MarkBufferDirty(bucket_obuf);
1295 1098 : MarkBufferDirty(bucket_nbuf);
1296 :
1297 1098 : if (RelationNeedsWAL(rel))
1298 : {
1299 : XLogRecPtr recptr;
1300 : xl_hash_split_complete xlrec;
1301 :
1302 846 : xlrec.old_bucket_flag = oopaque->hasho_flag;
1303 846 : xlrec.new_bucket_flag = nopaque->hasho_flag;
1304 :
1305 846 : XLogBeginInsert();
1306 :
1307 846 : XLogRegisterData((char *) &xlrec, SizeOfHashSplitComplete);
1308 :
1309 846 : XLogRegisterBuffer(0, bucket_obuf, REGBUF_STANDARD);
1310 846 : XLogRegisterBuffer(1, bucket_nbuf, REGBUF_STANDARD);
1311 :
1312 846 : recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_COMPLETE);
1313 :
1314 846 : PageSetLSN(BufferGetPage(bucket_obuf), recptr);
1315 846 : PageSetLSN(BufferGetPage(bucket_nbuf), recptr);
1316 : }
1317 :
1318 1098 : END_CRIT_SECTION();
1319 :
1320 : /*
1321 : * If possible, clean up the old bucket. We might not be able to do this
1322 : * if someone else has a pin on it, but if not then we can go ahead. This
1323 : * isn't absolutely necessary, but it reduces bloat; if we don't do it
1324 : * now, VACUUM will do it eventually, but maybe not until new overflow
1325 : * pages have been allocated. Note that there's no need to clean up the
1326 : * new bucket.
1327 : */
1328 1098 : if (IsBufferCleanupOK(bucket_obuf))
1329 : {
1330 1098 : LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1331 1098 : hashbucketcleanup(rel, obucket, bucket_obuf,
1332 : BufferGetBlockNumber(bucket_obuf), NULL,
1333 : maxbucket, highmask, lowmask, NULL, NULL, true,
1334 : NULL, NULL);
1335 : }
1336 : else
1337 : {
1338 0 : LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1339 0 : LockBuffer(bucket_obuf, BUFFER_LOCK_UNLOCK);
1340 : }
1341 1098 : }
1342 :
1343 : /*
1344 : * _hash_finish_split() -- Finish the previously interrupted split operation
1345 : *
1346 : * To complete the split operation, we form the hash table of TIDs in new
1347 : * bucket which is then used by split operation to skip tuples that are
1348 : * already moved before the split operation was previously interrupted.
1349 : *
1350 : * The caller must hold a pin, but no lock, on the metapage and old bucket's
1351 : * primary page buffer. The buffers are returned in the same state. (The
1352 : * metapage is only touched if it becomes necessary to add or remove overflow
1353 : * pages.)
1354 : */
1355 : void
1356 0 : _hash_finish_split(Relation rel, Buffer metabuf, Buffer obuf, Bucket obucket,
1357 : uint32 maxbucket, uint32 highmask, uint32 lowmask)
1358 : {
1359 : HASHCTL hash_ctl;
1360 : HTAB *tidhtab;
1361 0 : Buffer bucket_nbuf = InvalidBuffer;
1362 : Buffer nbuf;
1363 : Page npage;
1364 : BlockNumber nblkno;
1365 : BlockNumber bucket_nblkno;
1366 : HashPageOpaque npageopaque;
1367 : Bucket nbucket;
1368 : bool found;
1369 :
1370 : /* Initialize hash tables used to track TIDs */
1371 0 : hash_ctl.keysize = sizeof(ItemPointerData);
1372 0 : hash_ctl.entrysize = sizeof(ItemPointerData);
1373 0 : hash_ctl.hcxt = CurrentMemoryContext;
1374 :
1375 : tidhtab =
1376 0 : hash_create("bucket ctids",
1377 : 256, /* arbitrary initial size */
1378 : &hash_ctl,
1379 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1380 :
1381 0 : bucket_nblkno = nblkno = _hash_get_newblock_from_oldbucket(rel, obucket);
1382 :
1383 : /*
1384 : * Scan the new bucket and build hash table of TIDs
1385 : */
1386 : for (;;)
1387 0 : {
1388 : OffsetNumber noffnum;
1389 : OffsetNumber nmaxoffnum;
1390 :
1391 0 : nbuf = _hash_getbuf(rel, nblkno, HASH_READ,
1392 : LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
1393 :
1394 : /* remember the primary bucket buffer to acquire cleanup lock on it. */
1395 0 : if (nblkno == bucket_nblkno)
1396 0 : bucket_nbuf = nbuf;
1397 :
1398 0 : npage = BufferGetPage(nbuf);
1399 0 : npageopaque = HashPageGetOpaque(npage);
1400 :
1401 : /* Scan each tuple in new page */
1402 0 : nmaxoffnum = PageGetMaxOffsetNumber(npage);
1403 0 : for (noffnum = FirstOffsetNumber;
1404 : noffnum <= nmaxoffnum;
1405 0 : noffnum = OffsetNumberNext(noffnum))
1406 : {
1407 : IndexTuple itup;
1408 :
1409 : /* Fetch the item's TID and insert it in hash table. */
1410 0 : itup = (IndexTuple) PageGetItem(npage,
1411 : PageGetItemId(npage, noffnum));
1412 :
1413 0 : (void) hash_search(tidhtab, &itup->t_tid, HASH_ENTER, &found);
1414 :
1415 : Assert(!found);
1416 : }
1417 :
1418 0 : nblkno = npageopaque->hasho_nextblkno;
1419 :
1420 : /*
1421 : * release our write lock without modifying buffer and ensure to
1422 : * retain the pin on primary bucket.
1423 : */
1424 0 : if (nbuf == bucket_nbuf)
1425 0 : LockBuffer(nbuf, BUFFER_LOCK_UNLOCK);
1426 : else
1427 0 : _hash_relbuf(rel, nbuf);
1428 :
1429 : /* Exit loop if no more overflow pages in new bucket */
1430 0 : if (!BlockNumberIsValid(nblkno))
1431 0 : break;
1432 : }
1433 :
1434 : /*
1435 : * Conditionally get the cleanup lock on old and new buckets to perform
1436 : * the split operation. If we don't get the cleanup locks, silently give
1437 : * up and next insertion on old bucket will try again to complete the
1438 : * split.
1439 : */
1440 0 : if (!ConditionalLockBufferForCleanup(obuf))
1441 : {
1442 0 : hash_destroy(tidhtab);
1443 0 : return;
1444 : }
1445 0 : if (!ConditionalLockBufferForCleanup(bucket_nbuf))
1446 : {
1447 0 : LockBuffer(obuf, BUFFER_LOCK_UNLOCK);
1448 0 : hash_destroy(tidhtab);
1449 0 : return;
1450 : }
1451 :
1452 0 : npage = BufferGetPage(bucket_nbuf);
1453 0 : npageopaque = HashPageGetOpaque(npage);
1454 0 : nbucket = npageopaque->hasho_bucket;
1455 :
1456 0 : _hash_splitbucket(rel, metabuf, obucket,
1457 : nbucket, obuf, bucket_nbuf, tidhtab,
1458 : maxbucket, highmask, lowmask);
1459 :
1460 0 : _hash_dropbuf(rel, bucket_nbuf);
1461 0 : hash_destroy(tidhtab);
1462 : }
1463 :
1464 : /*
1465 : * log_split_page() -- Log the split operation
1466 : *
1467 : * We log the split operation when the new page in new bucket gets full,
1468 : * so we log the entire page.
1469 : *
1470 : * 'buf' must be locked by the caller which is also responsible for unlocking
1471 : * it.
1472 : */
1473 : static void
1474 1176 : log_split_page(Relation rel, Buffer buf)
1475 : {
1476 1176 : if (RelationNeedsWAL(rel))
1477 : {
1478 : XLogRecPtr recptr;
1479 :
1480 924 : XLogBeginInsert();
1481 :
1482 924 : XLogRegisterBuffer(0, buf, REGBUF_FORCE_IMAGE | REGBUF_STANDARD);
1483 :
1484 924 : recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_PAGE);
1485 :
1486 924 : PageSetLSN(BufferGetPage(buf), recptr);
1487 : }
1488 1176 : }
1489 :
1490 : /*
1491 : * _hash_getcachedmetap() -- Returns cached metapage data.
1492 : *
1493 : * If metabuf is not InvalidBuffer, caller must hold a pin, but no lock, on
1494 : * the metapage. If not set, we'll set it before returning if we have to
1495 : * refresh the cache, and return with a pin but no lock on it; caller is
1496 : * responsible for releasing the pin.
1497 : *
1498 : * We refresh the cache if it's not initialized yet or force_refresh is true.
1499 : */
1500 : HashMetaPage
1501 715120 : _hash_getcachedmetap(Relation rel, Buffer *metabuf, bool force_refresh)
1502 : {
1503 : Page page;
1504 :
1505 : Assert(metabuf);
1506 715120 : if (force_refresh || rel->rd_amcache == NULL)
1507 : {
1508 1118 : char *cache = NULL;
1509 :
1510 : /*
1511 : * It's important that we don't set rd_amcache to an invalid value.
1512 : * Either MemoryContextAlloc or _hash_getbuf could fail, so don't
1513 : * install a pointer to the newly-allocated storage in the actual
1514 : * relcache entry until both have succeeded.
1515 : */
1516 1118 : if (rel->rd_amcache == NULL)
1517 502 : cache = MemoryContextAlloc(rel->rd_indexcxt,
1518 : sizeof(HashMetaPageData));
1519 :
1520 : /* Read the metapage. */
1521 1118 : if (BufferIsValid(*metabuf))
1522 2 : LockBuffer(*metabuf, BUFFER_LOCK_SHARE);
1523 : else
1524 1116 : *metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ,
1525 : LH_META_PAGE);
1526 1118 : page = BufferGetPage(*metabuf);
1527 :
1528 : /* Populate the cache. */
1529 1118 : if (rel->rd_amcache == NULL)
1530 502 : rel->rd_amcache = cache;
1531 1118 : memcpy(rel->rd_amcache, HashPageGetMeta(page),
1532 : sizeof(HashMetaPageData));
1533 :
1534 : /* Release metapage lock, but keep the pin. */
1535 1118 : LockBuffer(*metabuf, BUFFER_LOCK_UNLOCK);
1536 : }
1537 :
1538 715120 : return (HashMetaPage) rel->rd_amcache;
1539 : }
1540 :
1541 : /*
1542 : * _hash_getbucketbuf_from_hashkey() -- Get the bucket's buffer for the given
1543 : * hashkey.
1544 : *
1545 : * Bucket pages do not move or get removed once they are allocated. This give
1546 : * us an opportunity to use the previously saved metapage contents to reach
1547 : * the target bucket buffer, instead of reading from the metapage every time.
1548 : * This saves one buffer access every time we want to reach the target bucket
1549 : * buffer, which is very helpful savings in bufmgr traffic and contention.
1550 : *
1551 : * The access type parameter (HASH_READ or HASH_WRITE) indicates whether the
1552 : * bucket buffer has to be locked for reading or writing.
1553 : *
1554 : * The out parameter cachedmetap is set with metapage contents used for
1555 : * hashkey to bucket buffer mapping. Some callers need this info to reach the
1556 : * old bucket in case of bucket split, see _hash_doinsert().
1557 : */
1558 : Buffer
1559 714474 : _hash_getbucketbuf_from_hashkey(Relation rel, uint32 hashkey, int access,
1560 : HashMetaPage *cachedmetap)
1561 : {
1562 : HashMetaPage metap;
1563 : Buffer buf;
1564 714474 : Buffer metabuf = InvalidBuffer;
1565 : Page page;
1566 : Bucket bucket;
1567 : BlockNumber blkno;
1568 : HashPageOpaque opaque;
1569 :
1570 : /* We read from target bucket buffer, hence locking is must. */
1571 : Assert(access == HASH_READ || access == HASH_WRITE);
1572 :
1573 714474 : metap = _hash_getcachedmetap(rel, &metabuf, false);
1574 : Assert(metap != NULL);
1575 :
1576 : /*
1577 : * Loop until we get a lock on the correct target bucket.
1578 : */
1579 : for (;;)
1580 : {
1581 : /*
1582 : * Compute the target bucket number, and convert to block number.
1583 : */
1584 715088 : bucket = _hash_hashkey2bucket(hashkey,
1585 : metap->hashm_maxbucket,
1586 : metap->hashm_highmask,
1587 : metap->hashm_lowmask);
1588 :
1589 715088 : blkno = BUCKET_TO_BLKNO(metap, bucket);
1590 :
1591 : /* Fetch the primary bucket page for the bucket */
1592 715088 : buf = _hash_getbuf(rel, blkno, access, LH_BUCKET_PAGE);
1593 715088 : page = BufferGetPage(buf);
1594 715088 : opaque = HashPageGetOpaque(page);
1595 : Assert(opaque->hasho_bucket == bucket);
1596 : Assert(opaque->hasho_prevblkno != InvalidBlockNumber);
1597 :
1598 : /*
1599 : * If this bucket hasn't been split, we're done.
1600 : */
1601 715088 : if (opaque->hasho_prevblkno <= metap->hashm_maxbucket)
1602 714474 : break;
1603 :
1604 : /* Drop lock on this buffer, update cached metapage, and retry. */
1605 614 : _hash_relbuf(rel, buf);
1606 614 : metap = _hash_getcachedmetap(rel, &metabuf, true);
1607 : Assert(metap != NULL);
1608 : }
1609 :
1610 714474 : if (BufferIsValid(metabuf))
1611 1116 : _hash_dropbuf(rel, metabuf);
1612 :
1613 714474 : if (cachedmetap)
1614 713876 : *cachedmetap = metap;
1615 :
1616 714474 : return buf;
1617 : }
|
