| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* gistxlog.c |
| 4 |
* WAL replay logic for GiST. |
| 5 |
* |
| 6 |
* |
| 7 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 8 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/access/gist/gistxlog.c |
| 12 |
*------------------------------------------------------------------------- |
| 13 |
*/ |
| 14 |
#include "postgres.h" |
| 15 |
|
| 16 |
#include "access/bufmask.h" |
| 17 |
#include "access/gist_private.h" |
| 18 |
#include "access/gistxlog.h" |
| 19 |
#include "access/transam.h" |
| 20 |
#include "access/xloginsert.h" |
| 21 |
#include "access/xlogutils.h" |
| 22 |
#include "storage/standby.h" |
| 23 |
#include "utils/memutils.h" |
| 24 |
#include "utils/rel.h" |
| 25 |
|
| 26 |
static MemoryContext opCtx; /* working memory for operations */ |
| 27 |
|
| 28 |
/* |
| 29 |
* Replay the clearing of F_FOLLOW_RIGHT flag on a child page. |
| 30 |
* |
| 31 |
* Even if the WAL record includes a full-page image, we have to update the |
| 32 |
* follow-right flag, because that change is not included in the full-page |
| 33 |
* image. To be sure that the intermediate state with the wrong flag value is |
| 34 |
* not visible to concurrent Hot Standby queries, this function handles |
| 35 |
* restoring the full-page image as well as updating the flag. (Note that |
| 36 |
* we never need to do anything else to the child page in the current WAL |
| 37 |
* action.) |
| 38 |
*/ |
| 39 |
static void |
| 40 |
gistRedoClearFollowRight(XLogReaderState *record, uint8 block_id) |
| 41 |
{ |
| 42 |
XLogRecPtr lsn = record->EndRecPtr; |
| 43 |
Buffer buffer; |
| 44 |
Page page; |
| 45 |
XLogRedoAction action; |
| 46 |
|
| 47 |
/* |
| 48 |
* Note that we still update the page even if it was restored from a full |
| 49 |
* page image, because the updated NSN is not included in the image. |
| 50 |
*/ |
| 51 |
action = XLogReadBufferForRedo(record, block_id, &buffer); |
| 52 |
if (action == BLK_NEEDS_REDO || action == BLK_RESTORED) |
| 53 |
{ |
| 54 |
page = BufferGetPage(buffer); |
| 55 |
|
| 56 |
GistPageSetNSN(page, lsn); |
| 57 |
GistClearFollowRight(page); |
| 58 |
|
| 59 |
PageSetLSN(page, lsn); |
| 60 |
MarkBufferDirty(buffer); |
| 61 |
} |
| 62 |
if (BufferIsValid(buffer)) |
| 63 |
UnlockReleaseBuffer(buffer); |
| 64 |
} |
| 65 |
|
| 66 |
/* |
| 67 |
* redo any page update (except page split) |
| 68 |
*/ |
| 69 |
static void |
| 70 |
gistRedoPageUpdateRecord(XLogReaderState *record) |
| 71 |
{ |
| 72 |
XLogRecPtr lsn = record->EndRecPtr; |
| 73 |
gistxlogPageUpdate *xldata = (gistxlogPageUpdate *) XLogRecGetData(record); |
| 74 |
Buffer buffer; |
| 75 |
Page page; |
| 76 |
|
| 77 |
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO) |
| 78 |
{ |
| 79 |
char *begin; |
| 80 |
char *data; |
| 81 |
Size datalen; |
| 82 |
int ninserted PG_USED_FOR_ASSERTS_ONLY = 0; |
| 83 |
|
| 84 |
data = begin = XLogRecGetBlockData(record, 0, &datalen); |
| 85 |
|
| 86 |
page = (Page) BufferGetPage(buffer); |
| 87 |
|
| 88 |
if (xldata->ntodelete == 1 && xldata->ntoinsert == 1) |
| 89 |
{ |
| 90 |
/* |
| 91 |
* When replacing one tuple with one other tuple, we must use |
| 92 |
* PageIndexTupleOverwrite for consistency with gistplacetopage. |
| 93 |
*/ |
| 94 |
OffsetNumber offnum = *((OffsetNumber *) data); |
| 95 |
IndexTuple itup; |
| 96 |
Size itupsize; |
| 97 |
|
| 98 |
data += sizeof(OffsetNumber); |
| 99 |
itup = (IndexTuple) data; |
| 100 |
itupsize = IndexTupleSize(itup); |
| 101 |
if (!PageIndexTupleOverwrite(page, offnum, (Item) itup, itupsize)) |
| 102 |
elog(ERROR, "failed to add item to GiST index page, size %d bytes", |
| 103 |
(int) itupsize); |
| 104 |
data += itupsize; |
| 105 |
/* should be nothing left after consuming 1 tuple */ |
| 106 |
Assert(data - begin == datalen); |
| 107 |
/* update insertion count for assert check below */ |
| 108 |
ninserted++; |
| 109 |
} |
| 110 |
else if (xldata->ntodelete > 0) |
| 111 |
{ |
| 112 |
/* Otherwise, delete old tuples if any */ |
| 113 |
OffsetNumber *todelete = (OffsetNumber *) data; |
| 114 |
|
| 115 |
data += sizeof(OffsetNumber) * xldata->ntodelete; |
| 116 |
|
| 117 |
PageIndexMultiDelete(page, todelete, xldata->ntodelete); |
| 118 |
if (GistPageIsLeaf(page)) |
| 119 |
GistMarkTuplesDeleted(page); |
| 120 |
} |
| 121 |
|
| 122 |
/* Add new tuples if any */ |
| 123 |
if (data - begin < datalen) |
| 124 |
{ |
| 125 |
OffsetNumber off = (PageIsEmpty(page)) ? FirstOffsetNumber : |
| 126 |
OffsetNumberNext(PageGetMaxOffsetNumber(page)); |
| 127 |
|
| 128 |
while (data - begin < datalen) |
| 129 |
{ |
| 130 |
IndexTuple itup = (IndexTuple) data; |
| 131 |
Size sz = IndexTupleSize(itup); |
| 132 |
OffsetNumber l; |
| 133 |
|
| 134 |
data += sz; |
| 135 |
|
| 136 |
l = PageAddItem(page, (Item) itup, sz, off, false, false); |
| 137 |
if (l == InvalidOffsetNumber) |
| 138 |
elog(ERROR, "failed to add item to GiST index page, size %d bytes", |
| 139 |
(int) sz); |
| 140 |
off++; |
| 141 |
ninserted++; |
| 142 |
} |
| 143 |
} |
| 144 |
|
| 145 |
/* Check that XLOG record contained expected number of tuples */ |
| 146 |
Assert(ninserted == xldata->ntoinsert); |
| 147 |
|
| 148 |
PageSetLSN(page, lsn); |
| 149 |
MarkBufferDirty(buffer); |
| 150 |
} |
| 151 |
|
| 152 |
/* |
| 153 |
* Fix follow-right data on left child page |
| 154 |
* |
| 155 |
* This must be done while still holding the lock on the target page. Note |
| 156 |
* that even if the target page no longer exists, we still attempt to |
| 157 |
* replay the change on the child page. |
| 158 |
*/ |
| 159 |
if (XLogRecHasBlockRef(record, 1)) |
| 160 |
gistRedoClearFollowRight(record, 1); |
| 161 |
|
| 162 |
if (BufferIsValid(buffer)) |
| 163 |
UnlockReleaseBuffer(buffer); |
| 164 |
} |
| 165 |
|
| 166 |
|
| 167 |
/* |
| 168 |
* redo delete on gist index page to remove tuples marked as DEAD during index |
| 169 |
* tuple insertion |
| 170 |
*/ |
| 171 |
static void |
| 172 |
gistRedoDeleteRecord(XLogReaderState *record) |
| 173 |
{ |
| 174 |
XLogRecPtr lsn = record->EndRecPtr; |
| 175 |
gistxlogDelete *xldata = (gistxlogDelete *) XLogRecGetData(record); |
| 176 |
Buffer buffer; |
| 177 |
Page page; |
| 178 |
OffsetNumber *toDelete = xldata->offsets; |
| 179 |
|
| 180 |
/* |
| 181 |
* If we have any conflict processing to do, it must happen before we |
| 182 |
* update the page. |
| 183 |
* |
| 184 |
* GiST delete records can conflict with standby queries. You might think |
| 185 |
* that vacuum records would conflict as well, but we've handled that |
| 186 |
* already. XLOG_HEAP2_PRUNE_VACUUM_SCAN records provide the highest xid |
| 187 |
* cleaned by the vacuum of the heap and so we can resolve any conflicts |
| 188 |
* just once when that arrives. After that we know that no conflicts |
| 189 |
* exist from individual gist vacuum records on that index. |
| 190 |
*/ |
| 191 |
if (InHotStandby) |
| 192 |
{ |
| 193 |
RelFileLocator rlocator; |
| 194 |
|
| 195 |
XLogRecGetBlockTag(record, 0, &rlocator, NULL, NULL); |
| 196 |
|
| 197 |
ResolveRecoveryConflictWithSnapshot(xldata->snapshotConflictHorizon, |
| 198 |
xldata->isCatalogRel, |
| 199 |
rlocator); |
| 200 |
} |
| 201 |
|
| 202 |
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO) |
| 203 |
{ |
| 204 |
page = (Page) BufferGetPage(buffer); |
| 205 |
|
| 206 |
PageIndexMultiDelete(page, toDelete, xldata->ntodelete); |
| 207 |
|
| 208 |
GistClearPageHasGarbage(page); |
| 209 |
GistMarkTuplesDeleted(page); |
| 210 |
|
| 211 |
PageSetLSN(page, lsn); |
| 212 |
MarkBufferDirty(buffer); |
| 213 |
} |
| 214 |
|
| 215 |
if (BufferIsValid(buffer)) |
| 216 |
UnlockReleaseBuffer(buffer); |
| 217 |
} |
| 218 |
|
| 219 |
/* |
| 220 |
* Returns an array of index pointers. |
| 221 |
*/ |
| 222 |
static IndexTuple * |
| 223 |
decodePageSplitRecord(char *begin, int len, int *n) |
| 224 |
{ |
| 225 |
char *ptr; |
| 226 |
int i = 0; |
| 227 |
IndexTuple *tuples; |
| 228 |
|
| 229 |
/* extract the number of tuples */ |
| 230 |
memcpy(n, begin, sizeof(int)); |
| 231 |
ptr = begin + sizeof(int); |
| 232 |
|
| 233 |
tuples = palloc(*n * sizeof(IndexTuple)); |
| 234 |
|
| 235 |
for (i = 0; i < *n; i++) |
| 236 |
{ |
| 237 |
Assert(ptr - begin < len); |
| 238 |
tuples[i] = (IndexTuple) ptr; |
| 239 |
ptr += IndexTupleSize((IndexTuple) ptr); |
| 240 |
} |
| 241 |
Assert(ptr - begin == len); |
| 242 |
|
| 243 |
return tuples; |
| 244 |
} |
| 245 |
|
| 246 |
static void |
| 247 |
gistRedoPageSplitRecord(XLogReaderState *record) |
| 248 |
{ |
| 249 |
XLogRecPtr lsn = record->EndRecPtr; |
| 250 |
gistxlogPageSplit *xldata = (gistxlogPageSplit *) XLogRecGetData(record); |
| 251 |
Buffer firstbuffer = InvalidBuffer; |
| 252 |
Buffer buffer; |
| 253 |
Page page; |
| 254 |
int i; |
| 255 |
bool isrootsplit = false; |
| 256 |
|
| 257 |
/* |
| 258 |
* We must hold lock on the first-listed page throughout the action, |
| 259 |
* including while updating the left child page (if any). We can unlock |
| 260 |
* remaining pages in the list as soon as they've been written, because |
| 261 |
* there is no path for concurrent queries to reach those pages without |
| 262 |
* first visiting the first-listed page. |
| 263 |
*/ |
| 264 |
|
| 265 |
/* loop around all pages */ |
| 266 |
for (i = 0; i < xldata->npage; i++) |
| 267 |
{ |
| 268 |
int flags; |
| 269 |
char *data; |
| 270 |
Size datalen; |
| 271 |
int num; |
| 272 |
BlockNumber blkno; |
| 273 |
IndexTuple *tuples; |
| 274 |
|
| 275 |
XLogRecGetBlockTag(record, i + 1, NULL, NULL, &blkno); |
| 276 |
if (blkno == GIST_ROOT_BLKNO) |
| 277 |
{ |
| 278 |
Assert(i == 0); |
| 279 |
isrootsplit = true; |
| 280 |
} |
| 281 |
|
| 282 |
buffer = XLogInitBufferForRedo(record, i + 1); |
| 283 |
page = (Page) BufferGetPage(buffer); |
| 284 |
data = XLogRecGetBlockData(record, i + 1, &datalen); |
| 285 |
|
| 286 |
tuples = decodePageSplitRecord(data, datalen, &num); |
| 287 |
|
| 288 |
/* ok, clear buffer */ |
| 289 |
if (xldata->origleaf && blkno != GIST_ROOT_BLKNO) |
| 290 |
flags = F_LEAF; |
| 291 |
else |
| 292 |
flags = 0; |
| 293 |
GISTInitBuffer(buffer, flags); |
| 294 |
|
| 295 |
/* and fill it */ |
| 296 |
gistfillbuffer(page, tuples, num, FirstOffsetNumber); |
| 297 |
|
| 298 |
if (blkno == GIST_ROOT_BLKNO) |
| 299 |
{ |
| 300 |
GistPageGetOpaque(page)->rightlink = InvalidBlockNumber; |
| 301 |
GistPageSetNSN(page, xldata->orignsn); |
| 302 |
GistClearFollowRight(page); |
| 303 |
} |
| 304 |
else |
| 305 |
{ |
| 306 |
if (i < xldata->npage - 1) |
| 307 |
{ |
| 308 |
BlockNumber nextblkno; |
| 309 |
|
| 310 |
XLogRecGetBlockTag(record, i + 2, NULL, NULL, &nextblkno); |
| 311 |
GistPageGetOpaque(page)->rightlink = nextblkno; |
| 312 |
} |
| 313 |
else |
| 314 |
GistPageGetOpaque(page)->rightlink = xldata->origrlink; |
| 315 |
GistPageSetNSN(page, xldata->orignsn); |
| 316 |
if (i < xldata->npage - 1 && !isrootsplit && |
| 317 |
xldata->markfollowright) |
| 318 |
GistMarkFollowRight(page); |
| 319 |
else |
| 320 |
GistClearFollowRight(page); |
| 321 |
} |
| 322 |
|
| 323 |
PageSetLSN(page, lsn); |
| 324 |
MarkBufferDirty(buffer); |
| 325 |
|
| 326 |
if (i == 0) |
| 327 |
firstbuffer = buffer; |
| 328 |
else |
| 329 |
UnlockReleaseBuffer(buffer); |
| 330 |
} |
| 331 |
|
| 332 |
/* Fix follow-right data on left child page, if any */ |
| 333 |
if (XLogRecHasBlockRef(record, 0)) |
| 334 |
gistRedoClearFollowRight(record, 0); |
| 335 |
|
| 336 |
/* Finally, release lock on the first page */ |
| 337 |
UnlockReleaseBuffer(firstbuffer); |
| 338 |
} |
| 339 |
|
| 340 |
/* redo page deletion */ |
| 341 |
static void |
| 342 |
gistRedoPageDelete(XLogReaderState *record) |
| 343 |
{ |
| 344 |
XLogRecPtr lsn = record->EndRecPtr; |
| 345 |
gistxlogPageDelete *xldata = (gistxlogPageDelete *) XLogRecGetData(record); |
| 346 |
Buffer parentBuffer; |
| 347 |
Buffer leafBuffer; |
| 348 |
|
| 349 |
if (XLogReadBufferForRedo(record, 0, &leafBuffer) == BLK_NEEDS_REDO) |
| 350 |
{ |
| 351 |
Page page = (Page) BufferGetPage(leafBuffer); |
| 352 |
|
| 353 |
GistPageSetDeleted(page, xldata->deleteXid); |
| 354 |
|
| 355 |
PageSetLSN(page, lsn); |
| 356 |
MarkBufferDirty(leafBuffer); |
| 357 |
} |
| 358 |
|
| 359 |
if (XLogReadBufferForRedo(record, 1, &parentBuffer) == BLK_NEEDS_REDO) |
| 360 |
{ |
| 361 |
Page page = (Page) BufferGetPage(parentBuffer); |
| 362 |
|
| 363 |
PageIndexTupleDelete(page, xldata->downlinkOffset); |
| 364 |
|
| 365 |
PageSetLSN(page, lsn); |
| 366 |
MarkBufferDirty(parentBuffer); |
| 367 |
} |
| 368 |
|
| 369 |
if (BufferIsValid(parentBuffer)) |
| 370 |
UnlockReleaseBuffer(parentBuffer); |
| 371 |
if (BufferIsValid(leafBuffer)) |
| 372 |
UnlockReleaseBuffer(leafBuffer); |
| 373 |
} |
| 374 |
|
| 375 |
static void |
| 376 |
gistRedoPageReuse(XLogReaderState *record) |
| 377 |
{ |
| 378 |
gistxlogPageReuse *xlrec = (gistxlogPageReuse *) XLogRecGetData(record); |
| 379 |
|
| 380 |
/* |
| 381 |
* PAGE_REUSE records exist to provide a conflict point when we reuse |
| 382 |
* pages in the index via the FSM. That's all they do though. |
| 383 |
* |
| 384 |
* snapshotConflictHorizon was the page's deleteXid. The |
| 385 |
* GlobalVisCheckRemovableFullXid(deleteXid) test in gistPageRecyclable() |
| 386 |
* conceptually mirrors the PGPROC->xmin > limitXmin test in |
| 387 |
* GetConflictingVirtualXIDs(). Consequently, one XID value achieves the |
| 388 |
* same exclusion effect on primary and standby. |
| 389 |
*/ |
| 390 |
if (InHotStandby) |
| 391 |
ResolveRecoveryConflictWithSnapshotFullXid(xlrec->snapshotConflictHorizon, |
| 392 |
xlrec->isCatalogRel, |
| 393 |
xlrec->locator); |
| 394 |
} |
| 395 |
|
| 396 |
void |
| 397 |
gist_redo(XLogReaderState *record) |
| 398 |
{ |
| 399 |
uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK; |
| 400 |
MemoryContext oldCxt; |
| 401 |
|
| 402 |
/* |
| 403 |
* GiST indexes do not require any conflict processing. NB: If we ever |
| 404 |
* implement a similar optimization we have in b-tree, and remove killed |
| 405 |
* tuples outside VACUUM, we'll need to handle that here. |
| 406 |
*/ |
| 407 |
|
| 408 |
oldCxt = MemoryContextSwitchTo(opCtx); |
| 409 |
switch (info) |
| 410 |
{ |
| 411 |
case XLOG_GIST_PAGE_UPDATE: |
| 412 |
gistRedoPageUpdateRecord(record); |
| 413 |
break; |
| 414 |
case XLOG_GIST_DELETE: |
| 415 |
gistRedoDeleteRecord(record); |
| 416 |
break; |
| 417 |
case XLOG_GIST_PAGE_REUSE: |
| 418 |
gistRedoPageReuse(record); |
| 419 |
break; |
| 420 |
case XLOG_GIST_PAGE_SPLIT: |
| 421 |
gistRedoPageSplitRecord(record); |
| 422 |
break; |
| 423 |
case XLOG_GIST_PAGE_DELETE: |
| 424 |
gistRedoPageDelete(record); |
| 425 |
break; |
| 426 |
case XLOG_GIST_ASSIGN_LSN: |
| 427 |
/* nop. See gistGetFakeLSN(). */ |
| 428 |
break; |
| 429 |
default: |
| 430 |
elog(PANIC, "gist_redo: unknown op code %u", info); |
| 431 |
} |
| 432 |
|
| 433 |
MemoryContextSwitchTo(oldCxt); |
| 434 |
MemoryContextReset(opCtx); |
| 435 |
} |
| 436 |
|
| 437 |
void |
| 438 |
gist_xlog_startup(void) |
| 439 |
{ |
| 440 |
opCtx = createTempGistContext(); |
| 441 |
} |
| 442 |
|
| 443 |
void |
| 444 |
gist_xlog_cleanup(void) |
| 445 |
{ |
| 446 |
MemoryContextDelete(opCtx); |
| 447 |
} |
| 448 |
|
| 449 |
/* |
| 450 |
* Mask a Gist page before running consistency checks on it. |
| 451 |
*/ |
| 452 |
void |
| 453 |
gist_mask(char *pagedata, BlockNumber blkno) |
| 454 |
{ |
| 455 |
Page page = (Page) pagedata; |
| 456 |
|
| 457 |
mask_page_lsn_and_checksum(page); |
| 458 |
|
| 459 |
mask_page_hint_bits(page); |
| 460 |
mask_unused_space(page); |
| 461 |
|
| 462 |
/* |
| 463 |
* NSN is nothing but a special purpose LSN. Hence, mask it for the same |
| 464 |
* reason as mask_page_lsn_and_checksum. |
| 465 |
*/ |
| 466 |
GistPageSetNSN(page, (uint64) MASK_MARKER); |
| 467 |
|
| 468 |
/* |
| 469 |
* We update F_FOLLOW_RIGHT flag on the left child after writing WAL |
| 470 |
* record. Hence, mask this flag. See gistplacetopage() for details. |
| 471 |
*/ |
| 472 |
GistMarkFollowRight(page); |
| 473 |
|
| 474 |
if (GistPageIsLeaf(page)) |
| 475 |
{ |
| 476 |
/* |
| 477 |
* In gist leaf pages, it is possible to modify the LP_FLAGS without |
| 478 |
* emitting any WAL record. Hence, mask the line pointer flags. See |
| 479 |
* gistkillitems() for details. |
| 480 |
*/ |
| 481 |
mask_lp_flags(page); |
| 482 |
} |
| 483 |
|
| 484 |
/* |
| 485 |
* During gist redo, we never mark a page as garbage. Hence, mask it to |
| 486 |
* ignore any differences. |
| 487 |
*/ |
| 488 |
GistClearPageHasGarbage(page); |
| 489 |
} |
| 490 |
|
| 491 |
/* |
| 492 |
* Write WAL record of a page split. |
| 493 |
*/ |
| 494 |
XLogRecPtr |
| 495 |
gistXLogSplit(bool page_is_leaf, |
| 496 |
SplitPageLayout *dist, |
| 497 |
BlockNumber origrlink, GistNSN orignsn, |
| 498 |
Buffer leftchildbuf, bool markfollowright) |
| 499 |
{ |
| 500 |
gistxlogPageSplit xlrec; |
| 501 |
SplitPageLayout *ptr; |
| 502 |
int npage = 0; |
| 503 |
XLogRecPtr recptr; |
| 504 |
int i; |
| 505 |
|
| 506 |
for (ptr = dist; ptr; ptr = ptr->next) |
| 507 |
npage++; |
| 508 |
|
| 509 |
xlrec.origrlink = origrlink; |
| 510 |
xlrec.orignsn = orignsn; |
| 511 |
xlrec.origleaf = page_is_leaf; |
| 512 |
xlrec.npage = (uint16) npage; |
| 513 |
xlrec.markfollowright = markfollowright; |
| 514 |
|
| 515 |
XLogBeginInsert(); |
| 516 |
|
| 517 |
/* |
| 518 |
* Include a full page image of the child buf. (only necessary if a |
| 519 |
* checkpoint happened since the child page was split) |
| 520 |
*/ |
| 521 |
if (BufferIsValid(leftchildbuf)) |
| 522 |
XLogRegisterBuffer(0, leftchildbuf, REGBUF_STANDARD); |
| 523 |
|
| 524 |
/* |
| 525 |
* NOTE: We register a lot of data. The caller must've called |
| 526 |
* XLogEnsureRecordSpace() to prepare for that. We cannot do it here, |
| 527 |
* because we're already in a critical section. If you change the number |
| 528 |
* of buffer or data registrations here, make sure you modify the |
| 529 |
* XLogEnsureRecordSpace() calls accordingly! |
| 530 |
*/ |
| 531 |
XLogRegisterData(&xlrec, sizeof(gistxlogPageSplit)); |
| 532 |
|
| 533 |
i = 1; |
| 534 |
for (ptr = dist; ptr; ptr = ptr->next) |
| 535 |
{ |
| 536 |
XLogRegisterBuffer(i, ptr->buffer, REGBUF_WILL_INIT); |
| 537 |
XLogRegisterBufData(i, &(ptr->block.num), sizeof(int)); |
| 538 |
XLogRegisterBufData(i, ptr->list, ptr->lenlist); |
| 539 |
i++; |
| 540 |
} |
| 541 |
|
| 542 |
recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_PAGE_SPLIT); |
| 543 |
|
| 544 |
return recptr; |
| 545 |
} |
| 546 |
|
| 547 |
/* |
| 548 |
* Write XLOG record describing a page deletion. This also includes removal of |
| 549 |
* downlink from the parent page. |
| 550 |
*/ |
| 551 |
XLogRecPtr |
| 552 |
gistXLogPageDelete(Buffer buffer, FullTransactionId xid, |
| 553 |
Buffer parentBuffer, OffsetNumber downlinkOffset) |
| 554 |
{ |
| 555 |
gistxlogPageDelete xlrec; |
| 556 |
XLogRecPtr recptr; |
| 557 |
|
| 558 |
xlrec.deleteXid = xid; |
| 559 |
xlrec.downlinkOffset = downlinkOffset; |
| 560 |
|
| 561 |
XLogBeginInsert(); |
| 562 |
XLogRegisterData(&xlrec, SizeOfGistxlogPageDelete); |
| 563 |
|
| 564 |
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD); |
| 565 |
XLogRegisterBuffer(1, parentBuffer, REGBUF_STANDARD); |
| 566 |
|
| 567 |
recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_PAGE_DELETE); |
| 568 |
|
| 569 |
return recptr; |
| 570 |
} |
| 571 |
|
| 572 |
/* |
| 573 |
* Write an empty XLOG record to assign a distinct LSN. |
| 574 |
*/ |
| 575 |
XLogRecPtr |
| 576 |
gistXLogAssignLSN(void) |
| 577 |
{ |
| 578 |
int dummy = 0; |
| 579 |
|
| 580 |
/* |
| 581 |
* Records other than XLOG_SWITCH must have content. We use an integer 0 |
| 582 |
* to follow the restriction. |
| 583 |
*/ |
| 584 |
XLogBeginInsert(); |
| 585 |
XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT); |
| 586 |
XLogRegisterData(&dummy, sizeof(dummy)); |
| 587 |
return XLogInsert(RM_GIST_ID, XLOG_GIST_ASSIGN_LSN); |
| 588 |
} |
| 589 |
|
| 590 |
/* |
| 591 |
* Write XLOG record about reuse of a deleted page. |
| 592 |
*/ |
| 593 |
void |
| 594 |
gistXLogPageReuse(Relation rel, Relation heaprel, |
| 595 |
BlockNumber blkno, FullTransactionId deleteXid) |
| 596 |
{ |
| 597 |
gistxlogPageReuse xlrec_reuse; |
| 598 |
|
| 599 |
/* |
| 600 |
* Note that we don't register the buffer with the record, because this |
| 601 |
* operation doesn't modify the page. This record only exists to provide a |
| 602 |
* conflict point for Hot Standby. |
| 603 |
*/ |
| 604 |
|
| 605 |
/* XLOG stuff */ |
| 606 |
xlrec_reuse.isCatalogRel = RelationIsAccessibleInLogicalDecoding(heaprel); |
| 607 |
xlrec_reuse.locator = rel->rd_locator; |
| 608 |
xlrec_reuse.block = blkno; |
| 609 |
xlrec_reuse.snapshotConflictHorizon = deleteXid; |
| 610 |
|
| 611 |
XLogBeginInsert(); |
| 612 |
XLogRegisterData(&xlrec_reuse, SizeOfGistxlogPageReuse); |
| 613 |
|
| 614 |
XLogInsert(RM_GIST_ID, XLOG_GIST_PAGE_REUSE); |
| 615 |
} |
| 616 |
|
| 617 |
/* |
| 618 |
* Write XLOG record describing a page update. The update can include any |
| 619 |
* number of deletions and/or insertions of tuples on a single index page. |
| 620 |
* |
| 621 |
* If this update inserts a downlink for a split page, also record that |
| 622 |
* the F_FOLLOW_RIGHT flag on the child page is cleared and NSN set. |
| 623 |
* |
| 624 |
* Note that both the todelete array and the tuples are marked as belonging |
| 625 |
* to the target buffer; they need not be stored in XLOG if XLogInsert decides |
| 626 |
* to log the whole buffer contents instead. |
| 627 |
*/ |
| 628 |
XLogRecPtr |
| 629 |
gistXLogUpdate(Buffer buffer, |
| 630 |
OffsetNumber *todelete, int ntodelete, |
| 631 |
IndexTuple *itup, int ituplen, |
| 632 |
Buffer leftchildbuf) |
| 633 |
{ |
| 634 |
gistxlogPageUpdate xlrec; |
| 635 |
int i; |
| 636 |
XLogRecPtr recptr; |
| 637 |
|
| 638 |
xlrec.ntodelete = ntodelete; |
| 639 |
xlrec.ntoinsert = ituplen; |
| 640 |
|
| 641 |
XLogBeginInsert(); |
| 642 |
XLogRegisterData(&xlrec, sizeof(gistxlogPageUpdate)); |
| 643 |
|
| 644 |
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD); |
| 645 |
XLogRegisterBufData(0, todelete, sizeof(OffsetNumber) * ntodelete); |
| 646 |
|
| 647 |
/* new tuples */ |
| 648 |
for (i = 0; i < ituplen; i++) |
| 649 |
XLogRegisterBufData(0, itup[i], IndexTupleSize(itup[i])); |
| 650 |
|
| 651 |
/* |
| 652 |
* Include a full page image of the child buf. (only necessary if a |
| 653 |
* checkpoint happened since the child page was split) |
| 654 |
*/ |
| 655 |
if (BufferIsValid(leftchildbuf)) |
| 656 |
XLogRegisterBuffer(1, leftchildbuf, REGBUF_STANDARD); |
| 657 |
|
| 658 |
recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_PAGE_UPDATE); |
| 659 |
|
| 660 |
return recptr; |
| 661 |
} |
| 662 |
|
| 663 |
/* |
| 664 |
* Write XLOG record describing a delete of leaf index tuples marked as DEAD |
| 665 |
* during new tuple insertion. One may think that this case is already covered |
| 666 |
* by gistXLogUpdate(). But deletion of index tuples might conflict with |
| 667 |
* standby queries and needs special handling. |
| 668 |
*/ |
| 669 |
XLogRecPtr |
| 670 |
gistXLogDelete(Buffer buffer, OffsetNumber *todelete, int ntodelete, |
| 671 |
TransactionId snapshotConflictHorizon, Relation heaprel) |
| 672 |
{ |
| 673 |
gistxlogDelete xlrec; |
| 674 |
XLogRecPtr recptr; |
| 675 |
|
| 676 |
xlrec.isCatalogRel = RelationIsAccessibleInLogicalDecoding(heaprel); |
| 677 |
xlrec.snapshotConflictHorizon = snapshotConflictHorizon; |
| 678 |
xlrec.ntodelete = ntodelete; |
| 679 |
|
| 680 |
XLogBeginInsert(); |
| 681 |
XLogRegisterData(&xlrec, SizeOfGistxlogDelete); |
| 682 |
|
| 683 |
/* |
| 684 |
* We need the target-offsets array whether or not we store the whole |
| 685 |
* buffer, to allow us to find the snapshotConflictHorizon on a standby |
| 686 |
* server. |
| 687 |
*/ |
| 688 |
XLogRegisterData(todelete, ntodelete * sizeof(OffsetNumber)); |
| 689 |
|
| 690 |
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD); |
| 691 |
|
| 692 |
recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_DELETE); |
| 693 |
|
| 694 |
return recptr; |
| 695 |
} |