| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* spgdoinsert.c |
| 4 |
* implementation of insert algorithm |
| 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/spgist/spgdoinsert.c |
| 12 |
* |
| 13 |
*------------------------------------------------------------------------- |
| 14 |
*/ |
| 15 |
|
| 16 |
#include "postgres.h" |
| 17 |
|
| 18 |
#include "access/genam.h" |
| 19 |
#include "access/spgist_private.h" |
| 20 |
#include "access/spgxlog.h" |
| 21 |
#include "access/xloginsert.h" |
| 22 |
#include "common/int.h" |
| 23 |
#include "common/pg_prng.h" |
| 24 |
#include "miscadmin.h" |
| 25 |
#include "storage/bufmgr.h" |
| 26 |
#include "utils/rel.h" |
| 27 |
|
| 28 |
|
| 29 |
/* |
| 30 |
* SPPageDesc tracks all info about a page we are inserting into. In some |
| 31 |
* situations it actually identifies a tuple, or even a specific node within |
| 32 |
* an inner tuple. But any of the fields can be invalid. If the buffer |
| 33 |
* field is valid, it implies we hold pin and exclusive lock on that buffer. |
| 34 |
* page pointer should be valid exactly when buffer is. |
| 35 |
*/ |
| 36 |
typedef struct SPPageDesc |
| 37 |
{ |
| 38 |
BlockNumber blkno; /* block number, or InvalidBlockNumber */ |
| 39 |
Buffer buffer; /* page's buffer number, or InvalidBuffer */ |
| 40 |
Page page; /* pointer to page buffer, or NULL */ |
| 41 |
OffsetNumber offnum; /* offset of tuple, or InvalidOffsetNumber */ |
| 42 |
int node; /* node number within inner tuple, or -1 */ |
| 43 |
} SPPageDesc; |
| 44 |
|
| 45 |
|
| 46 |
/* |
| 47 |
* Set the item pointer in the nodeN'th entry in inner tuple tup. This |
| 48 |
* is used to update the parent inner tuple's downlink after a move or |
| 49 |
* split operation. |
| 50 |
*/ |
| 51 |
void |
| 52 |
spgUpdateNodeLink(SpGistInnerTuple tup, int nodeN, |
| 53 |
BlockNumber blkno, OffsetNumber offset) |
| 54 |
{ |
| 55 |
int i; |
| 56 |
SpGistNodeTuple node; |
| 57 |
|
| 58 |
SGITITERATE(tup, i, node) |
| 59 |
{ |
| 60 |
if (i == nodeN) |
| 61 |
{ |
| 62 |
ItemPointerSet(&node->t_tid, blkno, offset); |
| 63 |
return; |
| 64 |
} |
| 65 |
} |
| 66 |
|
| 67 |
elog(ERROR, "failed to find requested node %d in SPGiST inner tuple", |
| 68 |
nodeN); |
| 69 |
} |
| 70 |
|
| 71 |
/* |
| 72 |
* Form a new inner tuple containing one more node than the given one, with |
| 73 |
* the specified label datum, inserted at offset "offset" in the node array. |
| 74 |
* The new tuple's prefix is the same as the old one's. |
| 75 |
* |
| 76 |
* Note that the new node initially has an invalid downlink. We'll find a |
| 77 |
* page to point it to later. |
| 78 |
*/ |
| 79 |
static SpGistInnerTuple |
| 80 |
addNode(SpGistState *state, SpGistInnerTuple tuple, Datum label, int offset) |
| 81 |
{ |
| 82 |
SpGistNodeTuple node, |
| 83 |
*nodes; |
| 84 |
int i; |
| 85 |
|
| 86 |
/* if offset is negative, insert at end */ |
| 87 |
if (offset < 0) |
| 88 |
offset = tuple->nNodes; |
| 89 |
else if (offset > tuple->nNodes) |
| 90 |
elog(ERROR, "invalid offset for adding node to SPGiST inner tuple"); |
| 91 |
|
| 92 |
nodes = palloc(sizeof(SpGistNodeTuple) * (tuple->nNodes + 1)); |
| 93 |
SGITITERATE(tuple, i, node) |
| 94 |
{ |
| 95 |
if (i < offset) |
| 96 |
nodes[i] = node; |
| 97 |
else |
| 98 |
nodes[i + 1] = node; |
| 99 |
} |
| 100 |
|
| 101 |
nodes[offset] = spgFormNodeTuple(state, label, false); |
| 102 |
|
| 103 |
return spgFormInnerTuple(state, |
| 104 |
(tuple->prefixSize > 0), |
| 105 |
SGITDATUM(tuple, state), |
| 106 |
tuple->nNodes + 1, |
| 107 |
nodes); |
| 108 |
} |
| 109 |
|
| 110 |
/* qsort comparator for sorting OffsetNumbers */ |
| 111 |
static int |
| 112 |
cmpOffsetNumbers(const void *a, const void *b) |
| 113 |
{ |
| 114 |
return pg_cmp_u16(*(const OffsetNumber *) a, *(const OffsetNumber *) b); |
| 115 |
} |
| 116 |
|
| 117 |
/* |
| 118 |
* Delete multiple tuples from an index page, preserving tuple offset numbers. |
| 119 |
* |
| 120 |
* The first tuple in the given list is replaced with a dead tuple of type |
| 121 |
* "firststate" (REDIRECT/DEAD/PLACEHOLDER); the remaining tuples are replaced |
| 122 |
* with dead tuples of type "reststate". If either firststate or reststate |
| 123 |
* is REDIRECT, blkno/offnum specify where to link to. |
| 124 |
* |
| 125 |
* NB: this is used during WAL replay, so beware of trying to make it too |
| 126 |
* smart. In particular, it shouldn't use "state" except for calling |
| 127 |
* spgFormDeadTuple(). This is also used in a critical section, so no |
| 128 |
* pallocs either! |
| 129 |
*/ |
| 130 |
void |
| 131 |
spgPageIndexMultiDelete(SpGistState *state, Page page, |
| 132 |
OffsetNumber *itemnos, int nitems, |
| 133 |
int firststate, int reststate, |
| 134 |
BlockNumber blkno, OffsetNumber offnum) |
| 135 |
{ |
| 136 |
OffsetNumber firstItem; |
| 137 |
OffsetNumber sortednos[MaxIndexTuplesPerPage]; |
| 138 |
SpGistDeadTuple tuple = NULL; |
| 139 |
int i; |
| 140 |
|
| 141 |
if (nitems == 0) |
| 142 |
return; /* nothing to do */ |
| 143 |
|
| 144 |
/* |
| 145 |
* For efficiency we want to use PageIndexMultiDelete, which requires the |
| 146 |
* targets to be listed in sorted order, so we have to sort the itemnos |
| 147 |
* array. (This also greatly simplifies the math for reinserting the |
| 148 |
* replacement tuples.) However, we must not scribble on the caller's |
| 149 |
* array, so we have to make a copy. |
| 150 |
*/ |
| 151 |
memcpy(sortednos, itemnos, sizeof(OffsetNumber) * nitems); |
| 152 |
if (nitems > 1) |
| 153 |
qsort(sortednos, nitems, sizeof(OffsetNumber), cmpOffsetNumbers); |
| 154 |
|
| 155 |
PageIndexMultiDelete(page, sortednos, nitems); |
| 156 |
|
| 157 |
firstItem = itemnos[0]; |
| 158 |
|
| 159 |
for (i = 0; i < nitems; i++) |
| 160 |
{ |
| 161 |
OffsetNumber itemno = sortednos[i]; |
| 162 |
int tupstate; |
| 163 |
|
| 164 |
tupstate = (itemno == firstItem) ? firststate : reststate; |
| 165 |
if (tuple == NULL || tuple->tupstate != tupstate) |
| 166 |
tuple = spgFormDeadTuple(state, tupstate, blkno, offnum); |
| 167 |
|
| 168 |
if (PageAddItem(page, (Item) tuple, tuple->size, |
| 169 |
itemno, false, false) != itemno) |
| 170 |
elog(ERROR, "failed to add item of size %u to SPGiST index page", |
| 171 |
tuple->size); |
| 172 |
|
| 173 |
if (tupstate == SPGIST_REDIRECT) |
| 174 |
SpGistPageGetOpaque(page)->nRedirection++; |
| 175 |
else if (tupstate == SPGIST_PLACEHOLDER) |
| 176 |
SpGistPageGetOpaque(page)->nPlaceholder++; |
| 177 |
} |
| 178 |
} |
| 179 |
|
| 180 |
/* |
| 181 |
* Update the parent inner tuple's downlink, and mark the parent buffer |
| 182 |
* dirty (this must be the last change to the parent page in the current |
| 183 |
* WAL action). |
| 184 |
*/ |
| 185 |
static void |
| 186 |
saveNodeLink(Relation index, SPPageDesc *parent, |
| 187 |
BlockNumber blkno, OffsetNumber offnum) |
| 188 |
{ |
| 189 |
SpGistInnerTuple innerTuple; |
| 190 |
|
| 191 |
innerTuple = (SpGistInnerTuple) PageGetItem(parent->page, |
| 192 |
PageGetItemId(parent->page, parent->offnum)); |
| 193 |
|
| 194 |
spgUpdateNodeLink(innerTuple, parent->node, blkno, offnum); |
| 195 |
|
| 196 |
MarkBufferDirty(parent->buffer); |
| 197 |
} |
| 198 |
|
| 199 |
/* |
| 200 |
* Add a leaf tuple to a leaf page where there is known to be room for it |
| 201 |
*/ |
| 202 |
static void |
| 203 |
addLeafTuple(Relation index, SpGistState *state, SpGistLeafTuple leafTuple, |
| 204 |
SPPageDesc *current, SPPageDesc *parent, bool isNulls, bool isNew) |
| 205 |
{ |
| 206 |
spgxlogAddLeaf xlrec; |
| 207 |
|
| 208 |
xlrec.newPage = isNew; |
| 209 |
xlrec.storesNulls = isNulls; |
| 210 |
|
| 211 |
/* these will be filled below as needed */ |
| 212 |
xlrec.offnumLeaf = InvalidOffsetNumber; |
| 213 |
xlrec.offnumHeadLeaf = InvalidOffsetNumber; |
| 214 |
xlrec.offnumParent = InvalidOffsetNumber; |
| 215 |
xlrec.nodeI = 0; |
| 216 |
|
| 217 |
START_CRIT_SECTION(); |
| 218 |
|
| 219 |
if (current->offnum == InvalidOffsetNumber || |
| 220 |
SpGistBlockIsRoot(current->blkno)) |
| 221 |
{ |
| 222 |
/* Tuple is not part of a chain */ |
| 223 |
SGLT_SET_NEXTOFFSET(leafTuple, InvalidOffsetNumber); |
| 224 |
current->offnum = SpGistPageAddNewItem(state, current->page, |
| 225 |
(Item) leafTuple, leafTuple->size, |
| 226 |
NULL, false); |
| 227 |
|
| 228 |
xlrec.offnumLeaf = current->offnum; |
| 229 |
|
| 230 |
/* Must update parent's downlink if any */ |
| 231 |
if (parent->buffer != InvalidBuffer) |
| 232 |
{ |
| 233 |
xlrec.offnumParent = parent->offnum; |
| 234 |
xlrec.nodeI = parent->node; |
| 235 |
|
| 236 |
saveNodeLink(index, parent, current->blkno, current->offnum); |
| 237 |
} |
| 238 |
} |
| 239 |
else |
| 240 |
{ |
| 241 |
/* |
| 242 |
* Tuple must be inserted into existing chain. We mustn't change the |
| 243 |
* chain's head address, but we don't need to chase the entire chain |
| 244 |
* to put the tuple at the end; we can insert it second. |
| 245 |
* |
| 246 |
* Also, it's possible that the "chain" consists only of a DEAD tuple, |
| 247 |
* in which case we should replace the DEAD tuple in-place. |
| 248 |
*/ |
| 249 |
SpGistLeafTuple head; |
| 250 |
OffsetNumber offnum; |
| 251 |
|
| 252 |
head = (SpGistLeafTuple) PageGetItem(current->page, |
| 253 |
PageGetItemId(current->page, current->offnum)); |
| 254 |
if (head->tupstate == SPGIST_LIVE) |
| 255 |
{ |
| 256 |
SGLT_SET_NEXTOFFSET(leafTuple, SGLT_GET_NEXTOFFSET(head)); |
| 257 |
offnum = SpGistPageAddNewItem(state, current->page, |
| 258 |
(Item) leafTuple, leafTuple->size, |
| 259 |
NULL, false); |
| 260 |
|
| 261 |
/* |
| 262 |
* re-get head of list because it could have been moved on page, |
| 263 |
* and set new second element |
| 264 |
*/ |
| 265 |
head = (SpGistLeafTuple) PageGetItem(current->page, |
| 266 |
PageGetItemId(current->page, current->offnum)); |
| 267 |
SGLT_SET_NEXTOFFSET(head, offnum); |
| 268 |
|
| 269 |
xlrec.offnumLeaf = offnum; |
| 270 |
xlrec.offnumHeadLeaf = current->offnum; |
| 271 |
} |
| 272 |
else if (head->tupstate == SPGIST_DEAD) |
| 273 |
{ |
| 274 |
SGLT_SET_NEXTOFFSET(leafTuple, InvalidOffsetNumber); |
| 275 |
PageIndexTupleDelete(current->page, current->offnum); |
| 276 |
if (PageAddItem(current->page, |
| 277 |
(Item) leafTuple, leafTuple->size, |
| 278 |
current->offnum, false, false) != current->offnum) |
| 279 |
elog(ERROR, "failed to add item of size %u to SPGiST index page", |
| 280 |
leafTuple->size); |
| 281 |
|
| 282 |
/* WAL replay distinguishes this case by equal offnums */ |
| 283 |
xlrec.offnumLeaf = current->offnum; |
| 284 |
xlrec.offnumHeadLeaf = current->offnum; |
| 285 |
} |
| 286 |
else |
| 287 |
elog(ERROR, "unexpected SPGiST tuple state: %d", head->tupstate); |
| 288 |
} |
| 289 |
|
| 290 |
MarkBufferDirty(current->buffer); |
| 291 |
|
| 292 |
if (RelationNeedsWAL(index) && !state->isBuild) |
| 293 |
{ |
| 294 |
XLogRecPtr recptr; |
| 295 |
int flags; |
| 296 |
|
| 297 |
XLogBeginInsert(); |
| 298 |
XLogRegisterData(&xlrec, sizeof(xlrec)); |
| 299 |
XLogRegisterData(leafTuple, leafTuple->size); |
| 300 |
|
| 301 |
flags = REGBUF_STANDARD; |
| 302 |
if (xlrec.newPage) |
| 303 |
flags |= REGBUF_WILL_INIT; |
| 304 |
XLogRegisterBuffer(0, current->buffer, flags); |
| 305 |
if (xlrec.offnumParent != InvalidOffsetNumber) |
| 306 |
XLogRegisterBuffer(1, parent->buffer, REGBUF_STANDARD); |
| 307 |
|
| 308 |
recptr = XLogInsert(RM_SPGIST_ID, XLOG_SPGIST_ADD_LEAF); |
| 309 |
|
| 310 |
PageSetLSN(current->page, recptr); |
| 311 |
|
| 312 |
/* update parent only if we actually changed it */ |
| 313 |
if (xlrec.offnumParent != InvalidOffsetNumber) |
| 314 |
{ |
| 315 |
PageSetLSN(parent->page, recptr); |
| 316 |
} |
| 317 |
} |
| 318 |
|
| 319 |
END_CRIT_SECTION(); |
| 320 |
} |
| 321 |
|
| 322 |
/* |
| 323 |
* Count the number and total size of leaf tuples in the chain starting at |
| 324 |
* current->offnum. Return number into *nToSplit and total size as function |
| 325 |
* result. |
| 326 |
* |
| 327 |
* Klugy special case when considering the root page (i.e., root is a leaf |
| 328 |
* page, but we're about to split for the first time): return fake large |
| 329 |
* values to force spgdoinsert() to take the doPickSplit rather than |
| 330 |
* moveLeafs code path. moveLeafs is not prepared to deal with root page. |
| 331 |
*/ |
| 332 |
static int |
| 333 |
checkSplitConditions(Relation index, SpGistState *state, |
| 334 |
SPPageDesc *current, int *nToSplit) |
| 335 |
{ |
| 336 |
int i, |
| 337 |
n = 0, |
| 338 |
totalSize = 0; |
| 339 |
|
| 340 |
if (SpGistBlockIsRoot(current->blkno)) |
| 341 |
{ |
| 342 |
/* return impossible values to force split */ |
| 343 |
*nToSplit = BLCKSZ; |
| 344 |
return BLCKSZ; |
| 345 |
} |
| 346 |
|
| 347 |
i = current->offnum; |
| 348 |
while (i != InvalidOffsetNumber) |
| 349 |
{ |
| 350 |
SpGistLeafTuple it; |
| 351 |
|
| 352 |
Assert(i >= FirstOffsetNumber && |
| 353 |
i <= PageGetMaxOffsetNumber(current->page)); |
| 354 |
it = (SpGistLeafTuple) PageGetItem(current->page, |
| 355 |
PageGetItemId(current->page, i)); |
| 356 |
if (it->tupstate == SPGIST_LIVE) |
| 357 |
{ |
| 358 |
n++; |
| 359 |
totalSize += it->size + sizeof(ItemIdData); |
| 360 |
} |
| 361 |
else if (it->tupstate == SPGIST_DEAD) |
| 362 |
{ |
| 363 |
/* We could see a DEAD tuple as first/only chain item */ |
| 364 |
Assert(i == current->offnum); |
| 365 |
Assert(SGLT_GET_NEXTOFFSET(it) == InvalidOffsetNumber); |
| 366 |
/* Don't count it in result, because it won't go to other page */ |
| 367 |
} |
| 368 |
else |
| 369 |
elog(ERROR, "unexpected SPGiST tuple state: %d", it->tupstate); |
| 370 |
|
| 371 |
i = SGLT_GET_NEXTOFFSET(it); |
| 372 |
} |
| 373 |
|
| 374 |
*nToSplit = n; |
| 375 |
|
| 376 |
return totalSize; |
| 377 |
} |
| 378 |
|
| 379 |
/* |
| 380 |
* current points to a leaf-tuple chain that we wanted to add newLeafTuple to, |
| 381 |
* but the chain has to be moved because there's not enough room to add |
| 382 |
* newLeafTuple to its page. We use this method when the chain contains |
| 383 |
* very little data so a split would be inefficient. We are sure we can |
| 384 |
* fit the chain plus newLeafTuple on one other page. |
| 385 |
*/ |
| 386 |
static void |
| 387 |
moveLeafs(Relation index, SpGistState *state, |
| 388 |
SPPageDesc *current, SPPageDesc *parent, |
| 389 |
SpGistLeafTuple newLeafTuple, bool isNulls) |
| 390 |
{ |
| 391 |
int i, |
| 392 |
nDelete, |
| 393 |
nInsert, |
| 394 |
size; |
| 395 |
Buffer nbuf; |
| 396 |
Page npage; |
| 397 |
OffsetNumber r = InvalidOffsetNumber, |
| 398 |
startOffset = InvalidOffsetNumber; |
| 399 |
bool replaceDead = false; |
| 400 |
OffsetNumber *toDelete; |
| 401 |
OffsetNumber *toInsert; |
| 402 |
BlockNumber nblkno; |
| 403 |
spgxlogMoveLeafs xlrec; |
| 404 |
char *leafdata, |
| 405 |
*leafptr; |
| 406 |
|
| 407 |
/* This doesn't work on root page */ |
| 408 |
Assert(parent->buffer != InvalidBuffer); |
| 409 |
Assert(parent->buffer != current->buffer); |
| 410 |
|
| 411 |
/* Locate the tuples to be moved, and count up the space needed */ |
| 412 |
i = PageGetMaxOffsetNumber(current->page); |
| 413 |
toDelete = (OffsetNumber *) palloc(sizeof(OffsetNumber) * i); |
| 414 |
toInsert = (OffsetNumber *) palloc(sizeof(OffsetNumber) * (i + 1)); |
| 415 |
|
| 416 |
size = newLeafTuple->size + sizeof(ItemIdData); |
| 417 |
|
| 418 |
nDelete = 0; |
| 419 |
i = current->offnum; |
| 420 |
while (i != InvalidOffsetNumber) |
| 421 |
{ |
| 422 |
SpGistLeafTuple it; |
| 423 |
|
| 424 |
Assert(i >= FirstOffsetNumber && |
| 425 |
i <= PageGetMaxOffsetNumber(current->page)); |
| 426 |
it = (SpGistLeafTuple) PageGetItem(current->page, |
| 427 |
PageGetItemId(current->page, i)); |
| 428 |
|
| 429 |
if (it->tupstate == SPGIST_LIVE) |
| 430 |
{ |
| 431 |
toDelete[nDelete] = i; |
| 432 |
size += it->size + sizeof(ItemIdData); |
| 433 |
nDelete++; |
| 434 |
} |
| 435 |
else if (it->tupstate == SPGIST_DEAD) |
| 436 |
{ |
| 437 |
/* We could see a DEAD tuple as first/only chain item */ |
| 438 |
Assert(i == current->offnum); |
| 439 |
Assert(SGLT_GET_NEXTOFFSET(it) == InvalidOffsetNumber); |
| 440 |
/* We don't want to move it, so don't count it in size */ |
| 441 |
toDelete[nDelete] = i; |
| 442 |
nDelete++; |
| 443 |
replaceDead = true; |
| 444 |
} |
| 445 |
else |
| 446 |
elog(ERROR, "unexpected SPGiST tuple state: %d", it->tupstate); |
| 447 |
|
| 448 |
i = SGLT_GET_NEXTOFFSET(it); |
| 449 |
} |
| 450 |
|
| 451 |
/* Find a leaf page that will hold them */ |
| 452 |
nbuf = SpGistGetBuffer(index, GBUF_LEAF | (isNulls ? GBUF_NULLS : 0), |
| 453 |
size, &xlrec.newPage); |
| 454 |
npage = BufferGetPage(nbuf); |
| 455 |
nblkno = BufferGetBlockNumber(nbuf); |
| 456 |
Assert(nblkno != current->blkno); |
| 457 |
|
| 458 |
leafdata = leafptr = palloc(size); |
| 459 |
|
| 460 |
START_CRIT_SECTION(); |
| 461 |
|
| 462 |
/* copy all the old tuples to new page, unless they're dead */ |
| 463 |
nInsert = 0; |
| 464 |
if (!replaceDead) |
| 465 |
{ |
| 466 |
for (i = 0; i < nDelete; i++) |
| 467 |
{ |
| 468 |
SpGistLeafTuple it; |
| 469 |
|
| 470 |
it = (SpGistLeafTuple) PageGetItem(current->page, |
| 471 |
PageGetItemId(current->page, toDelete[i])); |
| 472 |
Assert(it->tupstate == SPGIST_LIVE); |
| 473 |
|
| 474 |
/* |
| 475 |
* Update chain link (notice the chain order gets reversed, but we |
| 476 |
* don't care). We're modifying the tuple on the source page |
| 477 |
* here, but it's okay since we're about to delete it. |
| 478 |
*/ |
| 479 |
SGLT_SET_NEXTOFFSET(it, r); |
| 480 |
|
| 481 |
r = SpGistPageAddNewItem(state, npage, (Item) it, it->size, |
| 482 |
&startOffset, false); |
| 483 |
|
| 484 |
toInsert[nInsert] = r; |
| 485 |
nInsert++; |
| 486 |
|
| 487 |
/* save modified tuple into leafdata as well */ |
| 488 |
memcpy(leafptr, it, it->size); |
| 489 |
leafptr += it->size; |
| 490 |
} |
| 491 |
} |
| 492 |
|
| 493 |
/* add the new tuple as well */ |
| 494 |
SGLT_SET_NEXTOFFSET(newLeafTuple, r); |
| 495 |
r = SpGistPageAddNewItem(state, npage, |
| 496 |
(Item) newLeafTuple, newLeafTuple->size, |
| 497 |
&startOffset, false); |
| 498 |
toInsert[nInsert] = r; |
| 499 |
nInsert++; |
| 500 |
memcpy(leafptr, newLeafTuple, newLeafTuple->size); |
| 501 |
leafptr += newLeafTuple->size; |
| 502 |
|
| 503 |
/* |
| 504 |
* Now delete the old tuples, leaving a redirection pointer behind for the |
| 505 |
* first one, unless we're doing an index build; in which case there can't |
| 506 |
* be any concurrent scan so we need not provide a redirect. |
| 507 |
*/ |
| 508 |
spgPageIndexMultiDelete(state, current->page, toDelete, nDelete, |
| 509 |
state->isBuild ? SPGIST_PLACEHOLDER : SPGIST_REDIRECT, |
| 510 |
SPGIST_PLACEHOLDER, |
| 511 |
nblkno, r); |
| 512 |
|
| 513 |
/* Update parent's downlink and mark parent page dirty */ |
| 514 |
saveNodeLink(index, parent, nblkno, r); |
| 515 |
|
| 516 |
/* Mark the leaf pages too */ |
| 517 |
MarkBufferDirty(current->buffer); |
| 518 |
MarkBufferDirty(nbuf); |
| 519 |
|
| 520 |
if (RelationNeedsWAL(index) && !state->isBuild) |
| 521 |
{ |
| 522 |
XLogRecPtr recptr; |
| 523 |
|
| 524 |
/* prepare WAL info */ |
| 525 |
STORE_STATE(state, xlrec.stateSrc); |
| 526 |
|
| 527 |
xlrec.nMoves = nDelete; |
| 528 |
xlrec.replaceDead = replaceDead; |
| 529 |
xlrec.storesNulls = isNulls; |
| 530 |
|
| 531 |
xlrec.offnumParent = parent->offnum; |
| 532 |
xlrec.nodeI = parent->node; |
| 533 |
|
| 534 |
XLogBeginInsert(); |
| 535 |
XLogRegisterData(&xlrec, SizeOfSpgxlogMoveLeafs); |
| 536 |
XLogRegisterData(toDelete, |
| 537 |
sizeof(OffsetNumber) * nDelete); |
| 538 |
XLogRegisterData(toInsert, |
| 539 |
sizeof(OffsetNumber) * nInsert); |
| 540 |
XLogRegisterData(leafdata, leafptr - leafdata); |
| 541 |
|
| 542 |
XLogRegisterBuffer(0, current->buffer, REGBUF_STANDARD); |
| 543 |
XLogRegisterBuffer(1, nbuf, REGBUF_STANDARD | (xlrec.newPage ? REGBUF_WILL_INIT : 0)); |
| 544 |
XLogRegisterBuffer(2, parent->buffer, REGBUF_STANDARD); |
| 545 |
|
| 546 |
recptr = XLogInsert(RM_SPGIST_ID, XLOG_SPGIST_MOVE_LEAFS); |
| 547 |
|
| 548 |
PageSetLSN(current->page, recptr); |
| 549 |
PageSetLSN(npage, recptr); |
| 550 |
PageSetLSN(parent->page, recptr); |
| 551 |
} |
| 552 |
|
| 553 |
END_CRIT_SECTION(); |
| 554 |
|
| 555 |
/* Update local free-space cache and release new buffer */ |
| 556 |
SpGistSetLastUsedPage(index, nbuf); |
| 557 |
UnlockReleaseBuffer(nbuf); |
| 558 |
} |
| 559 |
|
| 560 |
/* |
| 561 |
* Update previously-created redirection tuple with appropriate destination |
| 562 |
* |
| 563 |
* We use this when it's not convenient to know the destination first. |
| 564 |
* The tuple should have been made with the "impossible" destination of |
| 565 |
* the metapage. |
| 566 |
*/ |
| 567 |
static void |
| 568 |
setRedirectionTuple(SPPageDesc *current, OffsetNumber position, |
| 569 |
BlockNumber blkno, OffsetNumber offnum) |
| 570 |
{ |
| 571 |
SpGistDeadTuple dt; |
| 572 |
|
| 573 |
dt = (SpGistDeadTuple) PageGetItem(current->page, |
| 574 |
PageGetItemId(current->page, position)); |
| 575 |
Assert(dt->tupstate == SPGIST_REDIRECT); |
| 576 |
Assert(ItemPointerGetBlockNumber(&dt->pointer) == SPGIST_METAPAGE_BLKNO); |
| 577 |
ItemPointerSet(&dt->pointer, blkno, offnum); |
| 578 |
} |
| 579 |
|
| 580 |
/* |
| 581 |
* Test to see if the user-defined picksplit function failed to do its job, |
| 582 |
* ie, it put all the leaf tuples into the same node. |
| 583 |
* If so, randomly divide the tuples into several nodes (all with the same |
| 584 |
* label) and return true to select allTheSame mode for this inner tuple. |
| 585 |
* |
| 586 |
* (This code is also used to forcibly select allTheSame mode for nulls.) |
| 587 |
* |
| 588 |
* If we know that the leaf tuples wouldn't all fit on one page, then we |
| 589 |
* exclude the last tuple (which is the incoming new tuple that forced a split) |
| 590 |
* from the check to see if more than one node is used. The reason for this |
| 591 |
* is that if the existing tuples are put into only one chain, then even if |
| 592 |
* we move them all to an empty page, there would still not be room for the |
| 593 |
* new tuple, so we'd get into an infinite loop of picksplit attempts. |
| 594 |
* Forcing allTheSame mode dodges this problem by ensuring the old tuples will |
| 595 |
* be split across pages. (Exercise for the reader: figure out why this |
| 596 |
* fixes the problem even when there is only one old tuple.) |
| 597 |
*/ |
| 598 |
static bool |
| 599 |
checkAllTheSame(spgPickSplitIn *in, spgPickSplitOut *out, bool tooBig, |
| 600 |
bool *includeNew) |
| 601 |
{ |
| 602 |
int theNode; |
| 603 |
int limit; |
| 604 |
int i; |
| 605 |
|
| 606 |
/* For the moment, assume we can include the new leaf tuple */ |
| 607 |
*includeNew = true; |
| 608 |
|
| 609 |
/* If there's only the new leaf tuple, don't select allTheSame mode */ |
| 610 |
if (in->nTuples <= 1) |
| 611 |
return false; |
| 612 |
|
| 613 |
/* If tuple set doesn't fit on one page, ignore the new tuple in test */ |
| 614 |
limit = tooBig ? in->nTuples - 1 : in->nTuples; |
| 615 |
|
| 616 |
/* Check to see if more than one node is populated */ |
| 617 |
theNode = out->mapTuplesToNodes[0]; |
| 618 |
for (i = 1; i < limit; i++) |
| 619 |
{ |
| 620 |
if (out->mapTuplesToNodes[i] != theNode) |
| 621 |
return false; |
| 622 |
} |
| 623 |
|
| 624 |
/* Nope, so override the picksplit function's decisions */ |
| 625 |
|
| 626 |
/* If the new tuple is in its own node, it can't be included in split */ |
| 627 |
if (tooBig && out->mapTuplesToNodes[in->nTuples - 1] != theNode) |
| 628 |
*includeNew = false; |
| 629 |
|
| 630 |
out->nNodes = 8; /* arbitrary number of child nodes */ |
| 631 |
|
| 632 |
/* Random assignment of tuples to nodes (note we include new tuple) */ |
| 633 |
for (i = 0; i < in->nTuples; i++) |
| 634 |
out->mapTuplesToNodes[i] = i % out->nNodes; |
| 635 |
|
| 636 |
/* The opclass may not use node labels, but if it does, duplicate 'em */ |
| 637 |
if (out->nodeLabels) |
| 638 |
{ |
| 639 |
Datum theLabel = out->nodeLabels[theNode]; |
| 640 |
|
| 641 |
out->nodeLabels = (Datum *) palloc(sizeof(Datum) * out->nNodes); |
| 642 |
for (i = 0; i < out->nNodes; i++) |
| 643 |
out->nodeLabels[i] = theLabel; |
| 644 |
} |
| 645 |
|
| 646 |
/* We don't touch the prefix or the leaf tuple datum assignments */ |
| 647 |
|
| 648 |
return true; |
| 649 |
} |
| 650 |
|
| 651 |
/* |
| 652 |
* current points to a leaf-tuple chain that we wanted to add newLeafTuple to, |
| 653 |
* but the chain has to be split because there's not enough room to add |
| 654 |
* newLeafTuple to its page. |
| 655 |
* |
| 656 |
* This function splits the leaf tuple set according to picksplit's rules, |
| 657 |
* creating one or more new chains that are spread across the current page |
| 658 |
* and an additional leaf page (we assume that two leaf pages will be |
| 659 |
* sufficient). A new inner tuple is created, and the parent downlink |
| 660 |
* pointer is updated to point to that inner tuple instead of the leaf chain. |
| 661 |
* |
| 662 |
* On exit, current contains the address of the new inner tuple. |
| 663 |
* |
| 664 |
* Returns true if we successfully inserted newLeafTuple during this function, |
| 665 |
* false if caller still has to do it (meaning another picksplit operation is |
| 666 |
* probably needed). Failure could occur if the picksplit result is fairly |
| 667 |
* unbalanced, or if newLeafTuple is just plain too big to fit on a page. |
| 668 |
* Because we force the picksplit result to be at least two chains, each |
| 669 |
* cycle will get rid of at least one leaf tuple from the chain, so the loop |
| 670 |
* will eventually terminate if lack of balance is the issue. If the tuple |
| 671 |
* is too big, we assume that repeated picksplit operations will eventually |
| 672 |
* make it small enough by repeated prefix-stripping. A broken opclass could |
| 673 |
* make this an infinite loop, though, so spgdoinsert() checks that the |
| 674 |
* leaf datums get smaller each time. |
| 675 |
*/ |
| 676 |
static bool |
| 677 |
doPickSplit(Relation index, SpGistState *state, |
| 678 |
SPPageDesc *current, SPPageDesc *parent, |
| 679 |
SpGistLeafTuple newLeafTuple, |
| 680 |
int level, bool isNulls, bool isNew) |
| 681 |
{ |
| 682 |
bool insertedNew = false; |
| 683 |
spgPickSplitIn in; |
| 684 |
spgPickSplitOut out; |
| 685 |
FmgrInfo *procinfo; |
| 686 |
bool includeNew; |
| 687 |
int i, |
| 688 |
max, |
| 689 |
n; |
| 690 |
SpGistInnerTuple innerTuple; |
| 691 |
SpGistNodeTuple node, |
| 692 |
*nodes; |
| 693 |
Buffer newInnerBuffer, |
| 694 |
newLeafBuffer; |
| 695 |
uint8 *leafPageSelect; |
| 696 |
int *leafSizes; |
| 697 |
OffsetNumber *toDelete; |
| 698 |
OffsetNumber *toInsert; |
| 699 |
OffsetNumber redirectTuplePos = InvalidOffsetNumber; |
| 700 |
OffsetNumber startOffsets[2]; |
| 701 |
SpGistLeafTuple *oldLeafs; |
| 702 |
SpGistLeafTuple *newLeafs; |
| 703 |
Datum leafDatums[INDEX_MAX_KEYS]; |
| 704 |
bool leafIsnulls[INDEX_MAX_KEYS]; |
| 705 |
int spaceToDelete; |
| 706 |
int currentFreeSpace; |
| 707 |
int totalLeafSizes; |
| 708 |
bool allTheSame; |
| 709 |
spgxlogPickSplit xlrec; |
| 710 |
char *leafdata, |
| 711 |
*leafptr; |
| 712 |
SPPageDesc saveCurrent; |
| 713 |
int nToDelete, |
| 714 |
nToInsert, |
| 715 |
maxToInclude; |
| 716 |
|
| 717 |
in.level = level; |
| 718 |
|
| 719 |
/* |
| 720 |
* Allocate per-leaf-tuple work arrays with max possible size |
| 721 |
*/ |
| 722 |
max = PageGetMaxOffsetNumber(current->page); |
| 723 |
n = max + 1; |
| 724 |
in.datums = (Datum *) palloc(sizeof(Datum) * n); |
| 725 |
toDelete = (OffsetNumber *) palloc(sizeof(OffsetNumber) * n); |
| 726 |
toInsert = (OffsetNumber *) palloc(sizeof(OffsetNumber) * n); |
| 727 |
oldLeafs = (SpGistLeafTuple *) palloc(sizeof(SpGistLeafTuple) * n); |
| 728 |
newLeafs = (SpGistLeafTuple *) palloc(sizeof(SpGistLeafTuple) * n); |
| 729 |
leafPageSelect = (uint8 *) palloc(sizeof(uint8) * n); |
| 730 |
|
| 731 |
STORE_STATE(state, xlrec.stateSrc); |
| 732 |
|
| 733 |
/* |
| 734 |
* Form list of leaf tuples which will be distributed as split result; |
| 735 |
* also, count up the amount of space that will be freed from current. |
| 736 |
* (Note that in the non-root case, we won't actually delete the old |
| 737 |
* tuples, only replace them with redirects or placeholders.) |
| 738 |
*/ |
| 739 |
nToInsert = 0; |
| 740 |
nToDelete = 0; |
| 741 |
spaceToDelete = 0; |
| 742 |
if (SpGistBlockIsRoot(current->blkno)) |
| 743 |
{ |
| 744 |
/* |
| 745 |
* We are splitting the root (which up to now is also a leaf page). |
| 746 |
* Its tuples are not linked, so scan sequentially to get them all. We |
| 747 |
* ignore the original value of current->offnum. |
| 748 |
*/ |
| 749 |
for (i = FirstOffsetNumber; i <= max; i++) |
| 750 |
{ |
| 751 |
SpGistLeafTuple it; |
| 752 |
|
| 753 |
it = (SpGistLeafTuple) PageGetItem(current->page, |
| 754 |
PageGetItemId(current->page, i)); |
| 755 |
if (it->tupstate == SPGIST_LIVE) |
| 756 |
{ |
| 757 |
in.datums[nToInsert] = |
| 758 |
isNulls ? (Datum) 0 : SGLTDATUM(it, state); |
| 759 |
oldLeafs[nToInsert] = it; |
| 760 |
nToInsert++; |
| 761 |
toDelete[nToDelete] = i; |
| 762 |
nToDelete++; |
| 763 |
/* we will delete the tuple altogether, so count full space */ |
| 764 |
spaceToDelete += it->size + sizeof(ItemIdData); |
| 765 |
} |
| 766 |
else /* tuples on root should be live */ |
| 767 |
elog(ERROR, "unexpected SPGiST tuple state: %d", it->tupstate); |
| 768 |
} |
| 769 |
} |
| 770 |
else |
| 771 |
{ |
| 772 |
/* Normal case, just collect the leaf tuples in the chain */ |
| 773 |
i = current->offnum; |
| 774 |
while (i != InvalidOffsetNumber) |
| 775 |
{ |
| 776 |
SpGistLeafTuple it; |
| 777 |
|
| 778 |
Assert(i >= FirstOffsetNumber && i <= max); |
| 779 |
it = (SpGistLeafTuple) PageGetItem(current->page, |
| 780 |
PageGetItemId(current->page, i)); |
| 781 |
if (it->tupstate == SPGIST_LIVE) |
| 782 |
{ |
| 783 |
in.datums[nToInsert] = |
| 784 |
isNulls ? (Datum) 0 : SGLTDATUM(it, state); |
| 785 |
oldLeafs[nToInsert] = it; |
| 786 |
nToInsert++; |
| 787 |
toDelete[nToDelete] = i; |
| 788 |
nToDelete++; |
| 789 |
/* we will not delete the tuple, only replace with dead */ |
| 790 |
Assert(it->size >= SGDTSIZE); |
| 791 |
spaceToDelete += it->size - SGDTSIZE; |
| 792 |
} |
| 793 |
else if (it->tupstate == SPGIST_DEAD) |
| 794 |
{ |
| 795 |
/* We could see a DEAD tuple as first/only chain item */ |
| 796 |
Assert(i == current->offnum); |
| 797 |
Assert(SGLT_GET_NEXTOFFSET(it) == InvalidOffsetNumber); |
| 798 |
toDelete[nToDelete] = i; |
| 799 |
nToDelete++; |
| 800 |
/* replacing it with redirect will save no space */ |
| 801 |
} |
| 802 |
else |
| 803 |
elog(ERROR, "unexpected SPGiST tuple state: %d", it->tupstate); |
| 804 |
|
| 805 |
i = SGLT_GET_NEXTOFFSET(it); |
| 806 |
} |
| 807 |
} |
| 808 |
in.nTuples = nToInsert; |
| 809 |
|
| 810 |
/* |
| 811 |
* We may not actually insert new tuple because another picksplit may be |
| 812 |
* necessary due to too large value, but we will try to allocate enough |
| 813 |
* space to include it; and in any case it has to be included in the input |
| 814 |
* for the picksplit function. So don't increment nToInsert yet. |
| 815 |
*/ |
| 816 |
in.datums[in.nTuples] = |
| 817 |
isNulls ? (Datum) 0 : SGLTDATUM(newLeafTuple, state); |
| 818 |
oldLeafs[in.nTuples] = newLeafTuple; |
| 819 |
in.nTuples++; |
| 820 |
|
| 821 |
memset(&out, 0, sizeof(out)); |
| 822 |
|
| 823 |
if (!isNulls) |
| 824 |
{ |
| 825 |
/* |
| 826 |
* Perform split using user-defined method. |
| 827 |
*/ |
| 828 |
procinfo = index_getprocinfo(index, 1, SPGIST_PICKSPLIT_PROC); |
| 829 |
FunctionCall2Coll(procinfo, |
| 830 |
index->rd_indcollation[0], |
| 831 |
PointerGetDatum(&in), |
| 832 |
PointerGetDatum(&out)); |
| 833 |
|
| 834 |
/* |
| 835 |
* Form new leaf tuples and count up the total space needed. |
| 836 |
*/ |
| 837 |
totalLeafSizes = 0; |
| 838 |
for (i = 0; i < in.nTuples; i++) |
| 839 |
{ |
| 840 |
if (state->leafTupDesc->natts > 1) |
| 841 |
spgDeformLeafTuple(oldLeafs[i], |
| 842 |
state->leafTupDesc, |
| 843 |
leafDatums, |
| 844 |
leafIsnulls, |
| 845 |
isNulls); |
| 846 |
|
| 847 |
leafDatums[spgKeyColumn] = out.leafTupleDatums[i]; |
| 848 |
leafIsnulls[spgKeyColumn] = false; |
| 849 |
|
| 850 |
newLeafs[i] = spgFormLeafTuple(state, &oldLeafs[i]->heapPtr, |
| 851 |
leafDatums, |
| 852 |
leafIsnulls); |
| 853 |
totalLeafSizes += newLeafs[i]->size + sizeof(ItemIdData); |
| 854 |
} |
| 855 |
} |
| 856 |
else |
| 857 |
{ |
| 858 |
/* |
| 859 |
* Perform dummy split that puts all tuples into one node. |
| 860 |
* checkAllTheSame will override this and force allTheSame mode. |
| 861 |
*/ |
| 862 |
out.hasPrefix = false; |
| 863 |
out.nNodes = 1; |
| 864 |
out.nodeLabels = NULL; |
| 865 |
out.mapTuplesToNodes = palloc0(sizeof(int) * in.nTuples); |
| 866 |
|
| 867 |
/* |
| 868 |
* Form new leaf tuples and count up the total space needed. |
| 869 |
*/ |
| 870 |
totalLeafSizes = 0; |
| 871 |
for (i = 0; i < in.nTuples; i++) |
| 872 |
{ |
| 873 |
if (state->leafTupDesc->natts > 1) |
| 874 |
spgDeformLeafTuple(oldLeafs[i], |
| 875 |
state->leafTupDesc, |
| 876 |
leafDatums, |
| 877 |
leafIsnulls, |
| 878 |
isNulls); |
| 879 |
|
| 880 |
/* |
| 881 |
* Nulls tree can contain only null key values. |
| 882 |
*/ |
| 883 |
leafDatums[spgKeyColumn] = (Datum) 0; |
| 884 |
leafIsnulls[spgKeyColumn] = true; |
| 885 |
|
| 886 |
newLeafs[i] = spgFormLeafTuple(state, &oldLeafs[i]->heapPtr, |
| 887 |
leafDatums, |
| 888 |
leafIsnulls); |
| 889 |
totalLeafSizes += newLeafs[i]->size + sizeof(ItemIdData); |
| 890 |
} |
| 891 |
} |
| 892 |
|
| 893 |
/* |
| 894 |
* Check to see if the picksplit function failed to separate the values, |
| 895 |
* ie, it put them all into the same child node. If so, select allTheSame |
| 896 |
* mode and create a random split instead. See comments for |
| 897 |
* checkAllTheSame as to why we need to know if the new leaf tuples could |
| 898 |
* fit on one page. |
| 899 |
*/ |
| 900 |
allTheSame = checkAllTheSame(&in, &out, |
| 901 |
totalLeafSizes > SPGIST_PAGE_CAPACITY, |
| 902 |
&includeNew); |
| 903 |
|
| 904 |
/* |
| 905 |
* If checkAllTheSame decided we must exclude the new tuple, don't |
| 906 |
* consider it any further. |
| 907 |
*/ |
| 908 |
if (includeNew) |
| 909 |
maxToInclude = in.nTuples; |
| 910 |
else |
| 911 |
{ |
| 912 |
maxToInclude = in.nTuples - 1; |
| 913 |
totalLeafSizes -= newLeafs[in.nTuples - 1]->size + sizeof(ItemIdData); |
| 914 |
} |
| 915 |
|
| 916 |
/* |
| 917 |
* Allocate per-node work arrays. Since checkAllTheSame could replace |
| 918 |
* out.nNodes with a value larger than the number of tuples on the input |
| 919 |
* page, we can't allocate these arrays before here. |
| 920 |
*/ |
| 921 |
nodes = (SpGistNodeTuple *) palloc(sizeof(SpGistNodeTuple) * out.nNodes); |
| 922 |
leafSizes = (int *) palloc0(sizeof(int) * out.nNodes); |
| 923 |
|
| 924 |
/* |
| 925 |
* Form nodes of inner tuple and inner tuple itself |
| 926 |
*/ |
| 927 |
for (i = 0; i < out.nNodes; i++) |
| 928 |
{ |
| 929 |
Datum label = (Datum) 0; |
| 930 |
bool labelisnull = (out.nodeLabels == NULL); |
| 931 |
|
| 932 |
if (!labelisnull) |
| 933 |
label = out.nodeLabels[i]; |
| 934 |
nodes[i] = spgFormNodeTuple(state, label, labelisnull); |
| 935 |
} |
| 936 |
innerTuple = spgFormInnerTuple(state, |
| 937 |
out.hasPrefix, out.prefixDatum, |
| 938 |
out.nNodes, nodes); |
| 939 |
innerTuple->allTheSame = allTheSame; |
| 940 |
|
| 941 |
/* |
| 942 |
* Update nodes[] array to point into the newly formed innerTuple, so that |
| 943 |
* we can adjust their downlinks below. |
| 944 |
*/ |
| 945 |
SGITITERATE(innerTuple, i, node) |
| 946 |
{ |
| 947 |
nodes[i] = node; |
| 948 |
} |
| 949 |
|
| 950 |
/* |
| 951 |
* Re-scan new leaf tuples and count up the space needed under each node. |
| 952 |
*/ |
| 953 |
for (i = 0; i < maxToInclude; i++) |
| 954 |
{ |
| 955 |
n = out.mapTuplesToNodes[i]; |
| 956 |
if (n < 0 || n >= out.nNodes) |
| 957 |
elog(ERROR, "inconsistent result of SPGiST picksplit function"); |
| 958 |
leafSizes[n] += newLeafs[i]->size + sizeof(ItemIdData); |
| 959 |
} |
| 960 |
|
| 961 |
/* |
| 962 |
* To perform the split, we must insert a new inner tuple, which can't go |
| 963 |
* on a leaf page; and unless we are splitting the root page, we must then |
| 964 |
* update the parent tuple's downlink to point to the inner tuple. If |
| 965 |
* there is room, we'll put the new inner tuple on the same page as the |
| 966 |
* parent tuple, otherwise we need another non-leaf buffer. But if the |
| 967 |
* parent page is the root, we can't add the new inner tuple there, |
| 968 |
* because the root page must have only one inner tuple. |
| 969 |
*/ |
| 970 |
xlrec.initInner = false; |
| 971 |
if (parent->buffer != InvalidBuffer && |
| 972 |
!SpGistBlockIsRoot(parent->blkno) && |
| 973 |
(SpGistPageGetFreeSpace(parent->page, 1) >= |
| 974 |
innerTuple->size + sizeof(ItemIdData))) |
| 975 |
{ |
| 976 |
/* New inner tuple will fit on parent page */ |
| 977 |
newInnerBuffer = parent->buffer; |
| 978 |
} |
| 979 |
else if (parent->buffer != InvalidBuffer) |
| 980 |
{ |
| 981 |
/* Send tuple to page with next triple parity (see README) */ |
| 982 |
newInnerBuffer = SpGistGetBuffer(index, |
| 983 |
GBUF_INNER_PARITY(parent->blkno + 1) | |
| 984 |
(isNulls ? GBUF_NULLS : 0), |
| 985 |
innerTuple->size + sizeof(ItemIdData), |
| 986 |
&xlrec.initInner); |
| 987 |
} |
| 988 |
else |
| 989 |
{ |
| 990 |
/* Root page split ... inner tuple will go to root page */ |
| 991 |
newInnerBuffer = InvalidBuffer; |
| 992 |
} |
| 993 |
|
| 994 |
/* |
| 995 |
* The new leaf tuples converted from the existing ones should require the |
| 996 |
* same or less space, and therefore should all fit onto one page |
| 997 |
* (although that's not necessarily the current page, since we can't |
| 998 |
* delete the old tuples but only replace them with placeholders). |
| 999 |
* However, the incoming new tuple might not also fit, in which case we |
| 1000 |
* might need another picksplit cycle to reduce it some more. |
| 1001 |
* |
| 1002 |
* If there's not room to put everything back onto the current page, then |
| 1003 |
* we decide on a per-node basis which tuples go to the new page. (We do |
| 1004 |
* it like that because leaf tuple chains can't cross pages, so we must |
| 1005 |
* place all leaf tuples belonging to the same parent node on the same |
| 1006 |
* page.) |
| 1007 |
* |
| 1008 |
* If we are splitting the root page (turning it from a leaf page into an |
| 1009 |
* inner page), then no leaf tuples can go back to the current page; they |
| 1010 |
* must all go somewhere else. |
| 1011 |
*/ |
| 1012 |
if (!SpGistBlockIsRoot(current->blkno)) |
| 1013 |
currentFreeSpace = PageGetExactFreeSpace(current->page) + spaceToDelete; |
| 1014 |
else |
| 1015 |
currentFreeSpace = 0; /* prevent assigning any tuples to current */ |
| 1016 |
|
| 1017 |
xlrec.initDest = false; |
| 1018 |
|
| 1019 |
if (totalLeafSizes <= currentFreeSpace) |
| 1020 |
{ |
| 1021 |
/* All the leaf tuples will fit on current page */ |
| 1022 |
newLeafBuffer = InvalidBuffer; |
| 1023 |
/* mark new leaf tuple as included in insertions, if allowed */ |
| 1024 |
if (includeNew) |
| 1025 |
{ |
| 1026 |
nToInsert++; |
| 1027 |
insertedNew = true; |
| 1028 |
} |
| 1029 |
for (i = 0; i < nToInsert; i++) |
| 1030 |
leafPageSelect[i] = 0; /* signifies current page */ |
| 1031 |
} |
| 1032 |
else if (in.nTuples == 1 && totalLeafSizes > SPGIST_PAGE_CAPACITY) |
| 1033 |
{ |
| 1034 |
/* |
| 1035 |
* We're trying to split up a long value by repeated suffixing, but |
| 1036 |
* it's not going to fit yet. Don't bother allocating a second leaf |
| 1037 |
* buffer that we won't be able to use. |
| 1038 |
*/ |
| 1039 |
newLeafBuffer = InvalidBuffer; |
| 1040 |
Assert(includeNew); |
| 1041 |
Assert(nToInsert == 0); |
| 1042 |
} |
| 1043 |
else |
| 1044 |
{ |
| 1045 |
/* We will need another leaf page */ |
| 1046 |
uint8 *nodePageSelect; |
| 1047 |
int curspace; |
| 1048 |
int newspace; |
| 1049 |
|
| 1050 |
newLeafBuffer = SpGistGetBuffer(index, |
| 1051 |
GBUF_LEAF | (isNulls ? GBUF_NULLS : 0), |
| 1052 |
Min(totalLeafSizes, |
| 1053 |
SPGIST_PAGE_CAPACITY), |
| 1054 |
&xlrec.initDest); |
| 1055 |
|
| 1056 |
/* |
| 1057 |
* Attempt to assign node groups to the two pages. We might fail to |
| 1058 |
* do so, even if totalLeafSizes is less than the available space, |
| 1059 |
* because we can't split a group across pages. |
| 1060 |
*/ |
| 1061 |
nodePageSelect = (uint8 *) palloc(sizeof(uint8) * out.nNodes); |
| 1062 |
|
| 1063 |
curspace = currentFreeSpace; |
| 1064 |
newspace = PageGetExactFreeSpace(BufferGetPage(newLeafBuffer)); |
| 1065 |
for (i = 0; i < out.nNodes; i++) |
| 1066 |
{ |
| 1067 |
if (leafSizes[i] <= curspace) |
| 1068 |
{ |
| 1069 |
nodePageSelect[i] = 0; /* signifies current page */ |
| 1070 |
curspace -= leafSizes[i]; |
| 1071 |
} |
| 1072 |
else |
| 1073 |
{ |
| 1074 |
nodePageSelect[i] = 1; /* signifies new leaf page */ |
| 1075 |
newspace -= leafSizes[i]; |
| 1076 |
} |
| 1077 |
} |
| 1078 |
if (curspace >= 0 && newspace >= 0) |
| 1079 |
{ |
| 1080 |
/* Successful assignment, so we can include the new leaf tuple */ |
| 1081 |
if (includeNew) |
| 1082 |
{ |
| 1083 |
nToInsert++; |
| 1084 |
insertedNew = true; |
| 1085 |
} |
| 1086 |
} |
| 1087 |
else if (includeNew) |
| 1088 |
{ |
| 1089 |
/* We must exclude the new leaf tuple from the split */ |
| 1090 |
int nodeOfNewTuple = out.mapTuplesToNodes[in.nTuples - 1]; |
| 1091 |
|
| 1092 |
leafSizes[nodeOfNewTuple] -= |
| 1093 |
newLeafs[in.nTuples - 1]->size + sizeof(ItemIdData); |
| 1094 |
|
| 1095 |
/* Repeat the node assignment process --- should succeed now */ |
| 1096 |
curspace = currentFreeSpace; |
| 1097 |
newspace = PageGetExactFreeSpace(BufferGetPage(newLeafBuffer)); |
| 1098 |
for (i = 0; i < out.nNodes; i++) |
| 1099 |
{ |
| 1100 |
if (leafSizes[i] <= curspace) |
| 1101 |
{ |
| 1102 |
nodePageSelect[i] = 0; /* signifies current page */ |
| 1103 |
curspace -= leafSizes[i]; |
| 1104 |
} |
| 1105 |
else |
| 1106 |
{ |
| 1107 |
nodePageSelect[i] = 1; /* signifies new leaf page */ |
| 1108 |
newspace -= leafSizes[i]; |
| 1109 |
} |
| 1110 |
} |
| 1111 |
if (curspace < 0 || newspace < 0) |
| 1112 |
elog(ERROR, "failed to divide leaf tuple groups across pages"); |
| 1113 |
} |
| 1114 |
else |
| 1115 |
{ |
| 1116 |
/* oops, we already excluded new tuple ... should not get here */ |
| 1117 |
elog(ERROR, "failed to divide leaf tuple groups across pages"); |
| 1118 |
} |
| 1119 |
/* Expand the per-node assignments to be shown per leaf tuple */ |
| 1120 |
for (i = 0; i < nToInsert; i++) |
| 1121 |
{ |
| 1122 |
n = out.mapTuplesToNodes[i]; |
| 1123 |
leafPageSelect[i] = nodePageSelect[n]; |
| 1124 |
} |
| 1125 |
} |
| 1126 |
|
| 1127 |
/* Start preparing WAL record */ |
| 1128 |
xlrec.nDelete = 0; |
| 1129 |
xlrec.initSrc = isNew; |
| 1130 |
xlrec.storesNulls = isNulls; |
| 1131 |
xlrec.isRootSplit = SpGistBlockIsRoot(current->blkno); |
| 1132 |
|
| 1133 |
leafdata = leafptr = (char *) palloc(totalLeafSizes); |
| 1134 |
|
| 1135 |
/* Here we begin making the changes to the target pages */ |
| 1136 |
START_CRIT_SECTION(); |
| 1137 |
|
| 1138 |
/* |
| 1139 |
* Delete old leaf tuples from current buffer, except when we're splitting |
| 1140 |
* the root; in that case there's no need because we'll re-init the page |
| 1141 |
* below. We do this first to make room for reinserting new leaf tuples. |
| 1142 |
*/ |
| 1143 |
if (!SpGistBlockIsRoot(current->blkno)) |
| 1144 |
{ |
| 1145 |
/* |
| 1146 |
* Init buffer instead of deleting individual tuples, but only if |
| 1147 |
* there aren't any other live tuples and only during build; otherwise |
| 1148 |
* we need to set a redirection tuple for concurrent scans. |
| 1149 |
*/ |
| 1150 |
if (state->isBuild && |
| 1151 |
nToDelete + SpGistPageGetOpaque(current->page)->nPlaceholder == |
| 1152 |
PageGetMaxOffsetNumber(current->page)) |
| 1153 |
{ |
| 1154 |
SpGistInitBuffer(current->buffer, |
| 1155 |
SPGIST_LEAF | (isNulls ? SPGIST_NULLS : 0)); |
| 1156 |
xlrec.initSrc = true; |
| 1157 |
} |
| 1158 |
else if (isNew) |
| 1159 |
{ |
| 1160 |
/* don't expose the freshly init'd buffer as a backup block */ |
| 1161 |
Assert(nToDelete == 0); |
| 1162 |
} |
| 1163 |
else |
| 1164 |
{ |
| 1165 |
xlrec.nDelete = nToDelete; |
| 1166 |
|
| 1167 |
if (!state->isBuild) |
| 1168 |
{ |
| 1169 |
/* |
| 1170 |
* Need to create redirect tuple (it will point to new inner |
| 1171 |
* tuple) but right now the new tuple's location is not known |
| 1172 |
* yet. So, set the redirection pointer to "impossible" value |
| 1173 |
* and remember its position to update tuple later. |
| 1174 |
*/ |
| 1175 |
if (nToDelete > 0) |
| 1176 |
redirectTuplePos = toDelete[0]; |
| 1177 |
spgPageIndexMultiDelete(state, current->page, |
| 1178 |
toDelete, nToDelete, |
| 1179 |
SPGIST_REDIRECT, |
| 1180 |
SPGIST_PLACEHOLDER, |
| 1181 |
SPGIST_METAPAGE_BLKNO, |
| 1182 |
FirstOffsetNumber); |
| 1183 |
} |
| 1184 |
else |
| 1185 |
{ |
| 1186 |
/* |
| 1187 |
* During index build there is not concurrent searches, so we |
| 1188 |
* don't need to create redirection tuple. |
| 1189 |
*/ |
| 1190 |
spgPageIndexMultiDelete(state, current->page, |
| 1191 |
toDelete, nToDelete, |
| 1192 |
SPGIST_PLACEHOLDER, |
| 1193 |
SPGIST_PLACEHOLDER, |
| 1194 |
InvalidBlockNumber, |
| 1195 |
InvalidOffsetNumber); |
| 1196 |
} |
| 1197 |
} |
| 1198 |
} |
| 1199 |
|
| 1200 |
/* |
| 1201 |
* Put leaf tuples on proper pages, and update downlinks in innerTuple's |
| 1202 |
* nodes. |
| 1203 |
*/ |
| 1204 |
startOffsets[0] = startOffsets[1] = InvalidOffsetNumber; |
| 1205 |
for (i = 0; i < nToInsert; i++) |
| 1206 |
{ |
| 1207 |
SpGistLeafTuple it = newLeafs[i]; |
| 1208 |
Buffer leafBuffer; |
| 1209 |
BlockNumber leafBlock; |
| 1210 |
OffsetNumber newoffset; |
| 1211 |
|
| 1212 |
/* Which page is it going to? */ |
| 1213 |
leafBuffer = leafPageSelect[i] ? newLeafBuffer : current->buffer; |
| 1214 |
leafBlock = BufferGetBlockNumber(leafBuffer); |
| 1215 |
|
| 1216 |
/* Link tuple into correct chain for its node */ |
| 1217 |
n = out.mapTuplesToNodes[i]; |
| 1218 |
|
| 1219 |
if (ItemPointerIsValid(&nodes[n]->t_tid)) |
| 1220 |
{ |
| 1221 |
Assert(ItemPointerGetBlockNumber(&nodes[n]->t_tid) == leafBlock); |
| 1222 |
SGLT_SET_NEXTOFFSET(it, ItemPointerGetOffsetNumber(&nodes[n]->t_tid)); |
| 1223 |
} |
| 1224 |
else |
| 1225 |
SGLT_SET_NEXTOFFSET(it, InvalidOffsetNumber); |
| 1226 |
|
| 1227 |
/* Insert it on page */ |
| 1228 |
newoffset = SpGistPageAddNewItem(state, BufferGetPage(leafBuffer), |
| 1229 |
(Item) it, it->size, |
| 1230 |
&startOffsets[leafPageSelect[i]], |
| 1231 |
false); |
| 1232 |
toInsert[i] = newoffset; |
| 1233 |
|
| 1234 |
/* ... and complete the chain linking */ |
| 1235 |
ItemPointerSet(&nodes[n]->t_tid, leafBlock, newoffset); |
| 1236 |
|
| 1237 |
/* Also copy leaf tuple into WAL data */ |
| 1238 |
memcpy(leafptr, newLeafs[i], newLeafs[i]->size); |
| 1239 |
leafptr += newLeafs[i]->size; |
| 1240 |
} |
| 1241 |
|
| 1242 |
/* |
| 1243 |
* We're done modifying the other leaf buffer (if any), so mark it dirty. |
| 1244 |
* current->buffer will be marked below, after we're entirely done |
| 1245 |
* modifying it. |
| 1246 |
*/ |
| 1247 |
if (newLeafBuffer != InvalidBuffer) |
| 1248 |
{ |
| 1249 |
MarkBufferDirty(newLeafBuffer); |
| 1250 |
} |
| 1251 |
|
| 1252 |
/* Remember current buffer, since we're about to change "current" */ |
| 1253 |
saveCurrent = *current; |
| 1254 |
|
| 1255 |
/* |
| 1256 |
* Store the new innerTuple |
| 1257 |
*/ |
| 1258 |
if (newInnerBuffer == parent->buffer && newInnerBuffer != InvalidBuffer) |
| 1259 |
{ |
| 1260 |
/* |
| 1261 |
* new inner tuple goes to parent page |
| 1262 |
*/ |
| 1263 |
Assert(current->buffer != parent->buffer); |
| 1264 |
|
| 1265 |
/* Repoint "current" at the new inner tuple */ |
| 1266 |
current->blkno = parent->blkno; |
| 1267 |
current->buffer = parent->buffer; |
| 1268 |
current->page = parent->page; |
| 1269 |
xlrec.offnumInner = current->offnum = |
| 1270 |
SpGistPageAddNewItem(state, current->page, |
| 1271 |
(Item) innerTuple, innerTuple->size, |
| 1272 |
NULL, false); |
| 1273 |
|
| 1274 |
/* |
| 1275 |
* Update parent node link and mark parent page dirty |
| 1276 |
*/ |
| 1277 |
xlrec.innerIsParent = true; |
| 1278 |
xlrec.offnumParent = parent->offnum; |
| 1279 |
xlrec.nodeI = parent->node; |
| 1280 |
saveNodeLink(index, parent, current->blkno, current->offnum); |
| 1281 |
|
| 1282 |
/* |
| 1283 |
* Update redirection link (in old current buffer) |
| 1284 |
*/ |
| 1285 |
if (redirectTuplePos != InvalidOffsetNumber) |
| 1286 |
setRedirectionTuple(&saveCurrent, redirectTuplePos, |
| 1287 |
current->blkno, current->offnum); |
| 1288 |
|
| 1289 |
/* Done modifying old current buffer, mark it dirty */ |
| 1290 |
MarkBufferDirty(saveCurrent.buffer); |
| 1291 |
} |
| 1292 |
else if (parent->buffer != InvalidBuffer) |
| 1293 |
{ |
| 1294 |
/* |
| 1295 |
* new inner tuple will be stored on a new page |
| 1296 |
*/ |
| 1297 |
Assert(newInnerBuffer != InvalidBuffer); |
| 1298 |
|
| 1299 |
/* Repoint "current" at the new inner tuple */ |
| 1300 |
current->buffer = newInnerBuffer; |
| 1301 |
current->blkno = BufferGetBlockNumber(current->buffer); |
| 1302 |
current->page = BufferGetPage(current->buffer); |
| 1303 |
xlrec.offnumInner = current->offnum = |
| 1304 |
SpGistPageAddNewItem(state, current->page, |
| 1305 |
(Item) innerTuple, innerTuple->size, |
| 1306 |
NULL, false); |
| 1307 |
|
| 1308 |
/* Done modifying new current buffer, mark it dirty */ |
| 1309 |
MarkBufferDirty(current->buffer); |
| 1310 |
|
| 1311 |
/* |
| 1312 |
* Update parent node link and mark parent page dirty |
| 1313 |
*/ |
| 1314 |
xlrec.innerIsParent = (parent->buffer == current->buffer); |
| 1315 |
xlrec.offnumParent = parent->offnum; |
| 1316 |
xlrec.nodeI = parent->node; |
| 1317 |
saveNodeLink(index, parent, current->blkno, current->offnum); |
| 1318 |
|
| 1319 |
/* |
| 1320 |
* Update redirection link (in old current buffer) |
| 1321 |
*/ |
| 1322 |
if (redirectTuplePos != InvalidOffsetNumber) |
| 1323 |
setRedirectionTuple(&saveCurrent, redirectTuplePos, |
| 1324 |
current->blkno, current->offnum); |
| 1325 |
|
| 1326 |
/* Done modifying old current buffer, mark it dirty */ |
| 1327 |
MarkBufferDirty(saveCurrent.buffer); |
| 1328 |
} |
| 1329 |
else |
| 1330 |
{ |
| 1331 |
/* |
| 1332 |
* Splitting root page, which was a leaf but now becomes inner page |
| 1333 |
* (and so "current" continues to point at it) |
| 1334 |
*/ |
| 1335 |
Assert(SpGistBlockIsRoot(current->blkno)); |
| 1336 |
Assert(redirectTuplePos == InvalidOffsetNumber); |
| 1337 |
|
| 1338 |
SpGistInitBuffer(current->buffer, (isNulls ? SPGIST_NULLS : 0)); |
| 1339 |
xlrec.initInner = true; |
| 1340 |
xlrec.innerIsParent = false; |
| 1341 |
|
| 1342 |
xlrec.offnumInner = current->offnum = |
| 1343 |
PageAddItem(current->page, (Item) innerTuple, innerTuple->size, |
| 1344 |
InvalidOffsetNumber, false, false); |
| 1345 |
if (current->offnum != FirstOffsetNumber) |
| 1346 |
elog(ERROR, "failed to add item of size %u to SPGiST index page", |
| 1347 |
innerTuple->size); |
| 1348 |
|
| 1349 |
/* No parent link to update, nor redirection to do */ |
| 1350 |
xlrec.offnumParent = InvalidOffsetNumber; |
| 1351 |
xlrec.nodeI = 0; |
| 1352 |
|
| 1353 |
/* Done modifying new current buffer, mark it dirty */ |
| 1354 |
MarkBufferDirty(current->buffer); |
| 1355 |
|
| 1356 |
/* saveCurrent doesn't represent a different buffer */ |
| 1357 |
saveCurrent.buffer = InvalidBuffer; |
| 1358 |
} |
| 1359 |
|
| 1360 |
if (RelationNeedsWAL(index) && !state->isBuild) |
| 1361 |
{ |
| 1362 |
XLogRecPtr recptr; |
| 1363 |
int flags; |
| 1364 |
|
| 1365 |
XLogBeginInsert(); |
| 1366 |
|
| 1367 |
xlrec.nInsert = nToInsert; |
| 1368 |
XLogRegisterData(&xlrec, SizeOfSpgxlogPickSplit); |
| 1369 |
|
| 1370 |
XLogRegisterData(toDelete, |
| 1371 |
sizeof(OffsetNumber) * xlrec.nDelete); |
| 1372 |
XLogRegisterData(toInsert, |
| 1373 |
sizeof(OffsetNumber) * xlrec.nInsert); |
| 1374 |
XLogRegisterData(leafPageSelect, |
| 1375 |
sizeof(uint8) * xlrec.nInsert); |
| 1376 |
XLogRegisterData(innerTuple, innerTuple->size); |
| 1377 |
XLogRegisterData(leafdata, leafptr - leafdata); |
| 1378 |
|
| 1379 |
/* Old leaf page */ |
| 1380 |
if (BufferIsValid(saveCurrent.buffer)) |
| 1381 |
{ |
| 1382 |
flags = REGBUF_STANDARD; |
| 1383 |
if (xlrec.initSrc) |
| 1384 |
flags |= REGBUF_WILL_INIT; |
| 1385 |
XLogRegisterBuffer(0, saveCurrent.buffer, flags); |
| 1386 |
} |
| 1387 |
|
| 1388 |
/* New leaf page */ |
| 1389 |
if (BufferIsValid(newLeafBuffer)) |
| 1390 |
{ |
| 1391 |
flags = REGBUF_STANDARD; |
| 1392 |
if (xlrec.initDest) |
| 1393 |
flags |= REGBUF_WILL_INIT; |
| 1394 |
XLogRegisterBuffer(1, newLeafBuffer, flags); |
| 1395 |
} |
| 1396 |
|
| 1397 |
/* Inner page */ |
| 1398 |
flags = REGBUF_STANDARD; |
| 1399 |
if (xlrec.initInner) |
| 1400 |
flags |= REGBUF_WILL_INIT; |
| 1401 |
XLogRegisterBuffer(2, current->buffer, flags); |
| 1402 |
|
| 1403 |
/* Parent page, if different from inner page */ |
| 1404 |
if (parent->buffer != InvalidBuffer) |
| 1405 |
{ |
| 1406 |
if (parent->buffer != current->buffer) |
| 1407 |
XLogRegisterBuffer(3, parent->buffer, REGBUF_STANDARD); |
| 1408 |
else |
| 1409 |
Assert(xlrec.innerIsParent); |
| 1410 |
} |
| 1411 |
|
| 1412 |
/* Issue the WAL record */ |
| 1413 |
recptr = XLogInsert(RM_SPGIST_ID, XLOG_SPGIST_PICKSPLIT); |
| 1414 |
|
| 1415 |
/* Update page LSNs on all affected pages */ |
| 1416 |
if (newLeafBuffer != InvalidBuffer) |
| 1417 |
{ |
| 1418 |
Page page = BufferGetPage(newLeafBuffer); |
| 1419 |
|
| 1420 |
PageSetLSN(page, recptr); |
| 1421 |
} |
| 1422 |
|
| 1423 |
if (saveCurrent.buffer != InvalidBuffer) |
| 1424 |
{ |
| 1425 |
Page page = BufferGetPage(saveCurrent.buffer); |
| 1426 |
|
| 1427 |
PageSetLSN(page, recptr); |
| 1428 |
} |
| 1429 |
|
| 1430 |
PageSetLSN(current->page, recptr); |
| 1431 |
|
| 1432 |
if (parent->buffer != InvalidBuffer) |
| 1433 |
{ |
| 1434 |
PageSetLSN(parent->page, recptr); |
| 1435 |
} |
| 1436 |
} |
| 1437 |
|
| 1438 |
END_CRIT_SECTION(); |
| 1439 |
|
| 1440 |
/* Update local free-space cache and unlock buffers */ |
| 1441 |
if (newLeafBuffer != InvalidBuffer) |
| 1442 |
{ |
| 1443 |
SpGistSetLastUsedPage(index, newLeafBuffer); |
| 1444 |
UnlockReleaseBuffer(newLeafBuffer); |
| 1445 |
} |
| 1446 |
if (saveCurrent.buffer != InvalidBuffer) |
| 1447 |
{ |
| 1448 |
SpGistSetLastUsedPage(index, saveCurrent.buffer); |
| 1449 |
UnlockReleaseBuffer(saveCurrent.buffer); |
| 1450 |
} |
| 1451 |
|
| 1452 |
return insertedNew; |
| 1453 |
} |
| 1454 |
|
| 1455 |
/* |
| 1456 |
* spgMatchNode action: descend to N'th child node of current inner tuple |
| 1457 |
*/ |
| 1458 |
static void |
| 1459 |
spgMatchNodeAction(Relation index, SpGistState *state, |
| 1460 |
SpGistInnerTuple innerTuple, |
| 1461 |
SPPageDesc *current, SPPageDesc *parent, int nodeN) |
| 1462 |
{ |
| 1463 |
int i; |
| 1464 |
SpGistNodeTuple node; |
| 1465 |
|
| 1466 |
/* Release previous parent buffer if any */ |
| 1467 |
if (parent->buffer != InvalidBuffer && |
| 1468 |
parent->buffer != current->buffer) |
| 1469 |
{ |
| 1470 |
SpGistSetLastUsedPage(index, parent->buffer); |
| 1471 |
UnlockReleaseBuffer(parent->buffer); |
| 1472 |
} |
| 1473 |
|
| 1474 |
/* Repoint parent to specified node of current inner tuple */ |
| 1475 |
parent->blkno = current->blkno; |
| 1476 |
parent->buffer = current->buffer; |
| 1477 |
parent->page = current->page; |
| 1478 |
parent->offnum = current->offnum; |
| 1479 |
parent->node = nodeN; |
| 1480 |
|
| 1481 |
/* Locate that node */ |
| 1482 |
SGITITERATE(innerTuple, i, node) |
| 1483 |
{ |
| 1484 |
if (i == nodeN) |
| 1485 |
break; |
| 1486 |
} |
| 1487 |
|
| 1488 |
if (i != nodeN) |
| 1489 |
elog(ERROR, "failed to find requested node %d in SPGiST inner tuple", |
| 1490 |
nodeN); |
| 1491 |
|
| 1492 |
/* Point current to the downlink location, if any */ |
| 1493 |
if (ItemPointerIsValid(&node->t_tid)) |
| 1494 |
{ |
| 1495 |
current->blkno = ItemPointerGetBlockNumber(&node->t_tid); |
| 1496 |
current->offnum = ItemPointerGetOffsetNumber(&node->t_tid); |
| 1497 |
} |
| 1498 |
else |
| 1499 |
{ |
| 1500 |
/* Downlink is empty, so we'll need to find a new page */ |
| 1501 |
current->blkno = InvalidBlockNumber; |
| 1502 |
current->offnum = InvalidOffsetNumber; |
| 1503 |
} |
| 1504 |
|
| 1505 |
current->buffer = InvalidBuffer; |
| 1506 |
current->page = NULL; |
| 1507 |
} |
| 1508 |
|
| 1509 |
/* |
| 1510 |
* spgAddNode action: add a node to the inner tuple at current |
| 1511 |
*/ |
| 1512 |
static void |
| 1513 |
spgAddNodeAction(Relation index, SpGistState *state, |
| 1514 |
SpGistInnerTuple innerTuple, |
| 1515 |
SPPageDesc *current, SPPageDesc *parent, |
| 1516 |
int nodeN, Datum nodeLabel) |
| 1517 |
{ |
| 1518 |
SpGistInnerTuple newInnerTuple; |
| 1519 |
spgxlogAddNode xlrec; |
| 1520 |
|
| 1521 |
/* Should not be applied to nulls */ |
| 1522 |
Assert(!SpGistPageStoresNulls(current->page)); |
| 1523 |
|
| 1524 |
/* Construct new inner tuple with additional node */ |
| 1525 |
newInnerTuple = addNode(state, innerTuple, nodeLabel, nodeN); |
| 1526 |
|
| 1527 |
/* Prepare WAL record */ |
| 1528 |
STORE_STATE(state, xlrec.stateSrc); |
| 1529 |
xlrec.offnum = current->offnum; |
| 1530 |
|
| 1531 |
/* we don't fill these unless we need to change the parent downlink */ |
| 1532 |
xlrec.parentBlk = -1; |
| 1533 |
xlrec.offnumParent = InvalidOffsetNumber; |
| 1534 |
xlrec.nodeI = 0; |
| 1535 |
|
| 1536 |
/* we don't fill these unless tuple has to be moved */ |
| 1537 |
xlrec.offnumNew = InvalidOffsetNumber; |
| 1538 |
xlrec.newPage = false; |
| 1539 |
|
| 1540 |
if (PageGetExactFreeSpace(current->page) >= |
| 1541 |
newInnerTuple->size - innerTuple->size) |
| 1542 |
{ |
| 1543 |
/* |
| 1544 |
* We can replace the inner tuple by new version in-place |
| 1545 |
*/ |
| 1546 |
START_CRIT_SECTION(); |
| 1547 |
|
| 1548 |
PageIndexTupleDelete(current->page, current->offnum); |
| 1549 |
if (PageAddItem(current->page, |
| 1550 |
(Item) newInnerTuple, newInnerTuple->size, |
| 1551 |
current->offnum, false, false) != current->offnum) |
| 1552 |
elog(ERROR, "failed to add item of size %u to SPGiST index page", |
| 1553 |
newInnerTuple->size); |
| 1554 |
|
| 1555 |
MarkBufferDirty(current->buffer); |
| 1556 |
|
| 1557 |
if (RelationNeedsWAL(index) && !state->isBuild) |
| 1558 |
{ |
| 1559 |
XLogRecPtr recptr; |
| 1560 |
|
| 1561 |
XLogBeginInsert(); |
| 1562 |
XLogRegisterData(&xlrec, sizeof(xlrec)); |
| 1563 |
XLogRegisterData(newInnerTuple, newInnerTuple->size); |
| 1564 |
|
| 1565 |
XLogRegisterBuffer(0, current->buffer, REGBUF_STANDARD); |
| 1566 |
|
| 1567 |
recptr = XLogInsert(RM_SPGIST_ID, XLOG_SPGIST_ADD_NODE); |
| 1568 |
|
| 1569 |
PageSetLSN(current->page, recptr); |
| 1570 |
} |
| 1571 |
|
| 1572 |
END_CRIT_SECTION(); |
| 1573 |
} |
| 1574 |
else |
| 1575 |
{ |
| 1576 |
/* |
| 1577 |
* move inner tuple to another page, and update parent |
| 1578 |
*/ |
| 1579 |
SpGistDeadTuple dt; |
| 1580 |
SPPageDesc saveCurrent; |
| 1581 |
|
| 1582 |
/* |
| 1583 |
* It should not be possible to get here for the root page, since we |
| 1584 |
* allow only one inner tuple on the root page, and spgFormInnerTuple |
| 1585 |
* always checks that inner tuples don't exceed the size of a page. |
| 1586 |
*/ |
| 1587 |
if (SpGistBlockIsRoot(current->blkno)) |
| 1588 |
elog(ERROR, "cannot enlarge root tuple any more"); |
| 1589 |
Assert(parent->buffer != InvalidBuffer); |
| 1590 |
|
| 1591 |
saveCurrent = *current; |
| 1592 |
|
| 1593 |
xlrec.offnumParent = parent->offnum; |
| 1594 |
xlrec.nodeI = parent->node; |
| 1595 |
|
| 1596 |
/* |
| 1597 |
* obtain new buffer with the same parity as current, since it will be |
| 1598 |
* a child of same parent tuple |
| 1599 |
*/ |
| 1600 |
current->buffer = SpGistGetBuffer(index, |
| 1601 |
GBUF_INNER_PARITY(current->blkno), |
| 1602 |
newInnerTuple->size + sizeof(ItemIdData), |
| 1603 |
&xlrec.newPage); |
| 1604 |
current->blkno = BufferGetBlockNumber(current->buffer); |
| 1605 |
current->page = BufferGetPage(current->buffer); |
| 1606 |
|
| 1607 |
/* |
| 1608 |
* Let's just make real sure new current isn't same as old. Right now |
| 1609 |
* that's impossible, but if SpGistGetBuffer ever got smart enough to |
| 1610 |
* delete placeholder tuples before checking space, maybe it wouldn't |
| 1611 |
* be impossible. The case would appear to work except that WAL |
| 1612 |
* replay would be subtly wrong, so I think a mere assert isn't enough |
| 1613 |
* here. |
| 1614 |
*/ |
| 1615 |
if (current->blkno == saveCurrent.blkno) |
| 1616 |
elog(ERROR, "SPGiST new buffer shouldn't be same as old buffer"); |
| 1617 |
|
| 1618 |
/* |
| 1619 |
* New current and parent buffer will both be modified; but note that |
| 1620 |
* parent buffer could be same as either new or old current. |
| 1621 |
*/ |
| 1622 |
if (parent->buffer == saveCurrent.buffer) |
| 1623 |
xlrec.parentBlk = 0; |
| 1624 |
else if (parent->buffer == current->buffer) |
| 1625 |
xlrec.parentBlk = 1; |
| 1626 |
else |
| 1627 |
xlrec.parentBlk = 2; |
| 1628 |
|
| 1629 |
START_CRIT_SECTION(); |
| 1630 |
|
| 1631 |
/* insert new ... */ |
| 1632 |
xlrec.offnumNew = current->offnum = |
| 1633 |
SpGistPageAddNewItem(state, current->page, |
| 1634 |
(Item) newInnerTuple, newInnerTuple->size, |
| 1635 |
NULL, false); |
| 1636 |
|
| 1637 |
MarkBufferDirty(current->buffer); |
| 1638 |
|
| 1639 |
/* update parent's downlink and mark parent page dirty */ |
| 1640 |
saveNodeLink(index, parent, current->blkno, current->offnum); |
| 1641 |
|
| 1642 |
/* |
| 1643 |
* Replace old tuple with a placeholder or redirection tuple. Unless |
| 1644 |
* doing an index build, we have to insert a redirection tuple for |
| 1645 |
* possible concurrent scans. We can't just delete it in any case, |
| 1646 |
* because that could change the offsets of other tuples on the page, |
| 1647 |
* breaking downlinks from their parents. |
| 1648 |
*/ |
| 1649 |
if (state->isBuild) |
| 1650 |
dt = spgFormDeadTuple(state, SPGIST_PLACEHOLDER, |
| 1651 |
InvalidBlockNumber, InvalidOffsetNumber); |
| 1652 |
else |
| 1653 |
dt = spgFormDeadTuple(state, SPGIST_REDIRECT, |
| 1654 |
current->blkno, current->offnum); |
| 1655 |
|
| 1656 |
PageIndexTupleDelete(saveCurrent.page, saveCurrent.offnum); |
| 1657 |
if (PageAddItem(saveCurrent.page, (Item) dt, dt->size, |
| 1658 |
saveCurrent.offnum, |
| 1659 |
false, false) != saveCurrent.offnum) |
| 1660 |
elog(ERROR, "failed to add item of size %u to SPGiST index page", |
| 1661 |
dt->size); |
| 1662 |
|
| 1663 |
if (state->isBuild) |
| 1664 |
SpGistPageGetOpaque(saveCurrent.page)->nPlaceholder++; |
| 1665 |
else |
| 1666 |
SpGistPageGetOpaque(saveCurrent.page)->nRedirection++; |
| 1667 |
|
| 1668 |
MarkBufferDirty(saveCurrent.buffer); |
| 1669 |
|
| 1670 |
if (RelationNeedsWAL(index) && !state->isBuild) |
| 1671 |
{ |
| 1672 |
XLogRecPtr recptr; |
| 1673 |
int flags; |
| 1674 |
|
| 1675 |
XLogBeginInsert(); |
| 1676 |
|
| 1677 |
/* orig page */ |
| 1678 |
XLogRegisterBuffer(0, saveCurrent.buffer, REGBUF_STANDARD); |
| 1679 |
/* new page */ |
| 1680 |
flags = REGBUF_STANDARD; |
| 1681 |
if (xlrec.newPage) |
| 1682 |
flags |= REGBUF_WILL_INIT; |
| 1683 |
XLogRegisterBuffer(1, current->buffer, flags); |
| 1684 |
/* parent page (if different from orig and new) */ |
| 1685 |
if (xlrec.parentBlk == 2) |
| 1686 |
XLogRegisterBuffer(2, parent->buffer, REGBUF_STANDARD); |
| 1687 |
|
| 1688 |
XLogRegisterData(&xlrec, sizeof(xlrec)); |
| 1689 |
XLogRegisterData(newInnerTuple, newInnerTuple->size); |
| 1690 |
|
| 1691 |
recptr = XLogInsert(RM_SPGIST_ID, XLOG_SPGIST_ADD_NODE); |
| 1692 |
|
| 1693 |
/* we don't bother to check if any of these are redundant */ |
| 1694 |
PageSetLSN(current->page, recptr); |
| 1695 |
PageSetLSN(parent->page, recptr); |
| 1696 |
PageSetLSN(saveCurrent.page, recptr); |
| 1697 |
} |
| 1698 |
|
| 1699 |
END_CRIT_SECTION(); |
| 1700 |
|
| 1701 |
/* Release saveCurrent if it's not same as current or parent */ |
| 1702 |
if (saveCurrent.buffer != current->buffer && |
| 1703 |
saveCurrent.buffer != parent->buffer) |
| 1704 |
{ |
| 1705 |
SpGistSetLastUsedPage(index, saveCurrent.buffer); |
| 1706 |
UnlockReleaseBuffer(saveCurrent.buffer); |
| 1707 |
} |
| 1708 |
} |
| 1709 |
} |
| 1710 |
|
| 1711 |
/* |
| 1712 |
* spgSplitNode action: split inner tuple at current into prefix and postfix |
| 1713 |
*/ |
| 1714 |
static void |
| 1715 |
spgSplitNodeAction(Relation index, SpGistState *state, |
| 1716 |
SpGistInnerTuple innerTuple, |
| 1717 |
SPPageDesc *current, spgChooseOut *out) |
| 1718 |
{ |
| 1719 |
SpGistInnerTuple prefixTuple, |
| 1720 |
postfixTuple; |
| 1721 |
SpGistNodeTuple node, |
| 1722 |
*nodes; |
| 1723 |
BlockNumber postfixBlkno; |
| 1724 |
OffsetNumber postfixOffset; |
| 1725 |
int i; |
| 1726 |
spgxlogSplitTuple xlrec; |
| 1727 |
Buffer newBuffer = InvalidBuffer; |
| 1728 |
|
| 1729 |
/* Should not be applied to nulls */ |
| 1730 |
Assert(!SpGistPageStoresNulls(current->page)); |
| 1731 |
|
| 1732 |
/* Check opclass gave us sane values */ |
| 1733 |
if (out->result.splitTuple.prefixNNodes <= 0 || |
| 1734 |
out->result.splitTuple.prefixNNodes > SGITMAXNNODES) |
| 1735 |
elog(ERROR, "invalid number of prefix nodes: %d", |
| 1736 |
out->result.splitTuple.prefixNNodes); |
| 1737 |
if (out->result.splitTuple.childNodeN < 0 || |
| 1738 |
out->result.splitTuple.childNodeN >= |
| 1739 |
out->result.splitTuple.prefixNNodes) |
| 1740 |
elog(ERROR, "invalid child node number: %d", |
| 1741 |
out->result.splitTuple.childNodeN); |
| 1742 |
|
| 1743 |
/* |
| 1744 |
* Construct new prefix tuple with requested number of nodes. We'll fill |
| 1745 |
* in the childNodeN'th node's downlink below. |
| 1746 |
*/ |
| 1747 |
nodes = (SpGistNodeTuple *) palloc(sizeof(SpGistNodeTuple) * |
| 1748 |
out->result.splitTuple.prefixNNodes); |
| 1749 |
|
| 1750 |
for (i = 0; i < out->result.splitTuple.prefixNNodes; i++) |
| 1751 |
{ |
| 1752 |
Datum label = (Datum) 0; |
| 1753 |
bool labelisnull; |
| 1754 |
|
| 1755 |
labelisnull = (out->result.splitTuple.prefixNodeLabels == NULL); |
| 1756 |
if (!labelisnull) |
| 1757 |
label = out->result.splitTuple.prefixNodeLabels[i]; |
| 1758 |
nodes[i] = spgFormNodeTuple(state, label, labelisnull); |
| 1759 |
} |
| 1760 |
|
| 1761 |
prefixTuple = spgFormInnerTuple(state, |
| 1762 |
out->result.splitTuple.prefixHasPrefix, |
| 1763 |
out->result.splitTuple.prefixPrefixDatum, |
| 1764 |
out->result.splitTuple.prefixNNodes, |
| 1765 |
nodes); |
| 1766 |
|
| 1767 |
/* it must fit in the space that innerTuple now occupies */ |
| 1768 |
if (prefixTuple->size > innerTuple->size) |
| 1769 |
elog(ERROR, "SPGiST inner-tuple split must not produce longer prefix"); |
| 1770 |
|
| 1771 |
/* |
| 1772 |
* Construct new postfix tuple, containing all nodes of innerTuple with |
| 1773 |
* same node datums, but with the prefix specified by the picksplit |
| 1774 |
* function. |
| 1775 |
*/ |
| 1776 |
nodes = palloc(sizeof(SpGistNodeTuple) * innerTuple->nNodes); |
| 1777 |
SGITITERATE(innerTuple, i, node) |
| 1778 |
{ |
| 1779 |
nodes[i] = node; |
| 1780 |
} |
| 1781 |
|
| 1782 |
postfixTuple = spgFormInnerTuple(state, |
| 1783 |
out->result.splitTuple.postfixHasPrefix, |
| 1784 |
out->result.splitTuple.postfixPrefixDatum, |
| 1785 |
innerTuple->nNodes, nodes); |
| 1786 |
|
| 1787 |
/* Postfix tuple is allTheSame if original tuple was */ |
| 1788 |
postfixTuple->allTheSame = innerTuple->allTheSame; |
| 1789 |
|
| 1790 |
/* prep data for WAL record */ |
| 1791 |
xlrec.newPage = false; |
| 1792 |
|
| 1793 |
/* |
| 1794 |
* If we can't fit both tuples on the current page, get a new page for the |
| 1795 |
* postfix tuple. In particular, can't split to the root page. |
| 1796 |
* |
| 1797 |
* For the space calculation, note that prefixTuple replaces innerTuple |
| 1798 |
* but postfixTuple will be a new entry. |
| 1799 |
*/ |
| 1800 |
if (SpGistBlockIsRoot(current->blkno) || |
| 1801 |
SpGistPageGetFreeSpace(current->page, 1) + innerTuple->size < |
| 1802 |
prefixTuple->size + postfixTuple->size + sizeof(ItemIdData)) |
| 1803 |
{ |
| 1804 |
/* |
| 1805 |
* Choose page with next triple parity, because postfix tuple is a |
| 1806 |
* child of prefix one |
| 1807 |
*/ |
| 1808 |
newBuffer = SpGistGetBuffer(index, |
| 1809 |
GBUF_INNER_PARITY(current->blkno + 1), |
| 1810 |
postfixTuple->size + sizeof(ItemIdData), |
| 1811 |
&xlrec.newPage); |
| 1812 |
} |
| 1813 |
|
| 1814 |
START_CRIT_SECTION(); |
| 1815 |
|
| 1816 |
/* |
| 1817 |
* Replace old tuple by prefix tuple |
| 1818 |
*/ |
| 1819 |
PageIndexTupleDelete(current->page, current->offnum); |
| 1820 |
xlrec.offnumPrefix = PageAddItem(current->page, |
| 1821 |
(Item) prefixTuple, prefixTuple->size, |
| 1822 |
current->offnum, false, false); |
| 1823 |
if (xlrec.offnumPrefix != current->offnum) |
| 1824 |
elog(ERROR, "failed to add item of size %u to SPGiST index page", |
| 1825 |
prefixTuple->size); |
| 1826 |
|
| 1827 |
/* |
| 1828 |
* put postfix tuple into appropriate page |
| 1829 |
*/ |
| 1830 |
if (newBuffer == InvalidBuffer) |
| 1831 |
{ |
| 1832 |
postfixBlkno = current->blkno; |
| 1833 |
xlrec.offnumPostfix = postfixOffset = |
| 1834 |
SpGistPageAddNewItem(state, current->page, |
| 1835 |
(Item) postfixTuple, postfixTuple->size, |
| 1836 |
NULL, false); |
| 1837 |
xlrec.postfixBlkSame = true; |
| 1838 |
} |
| 1839 |
else |
| 1840 |
{ |
| 1841 |
postfixBlkno = BufferGetBlockNumber(newBuffer); |
| 1842 |
xlrec.offnumPostfix = postfixOffset = |
| 1843 |
SpGistPageAddNewItem(state, BufferGetPage(newBuffer), |
| 1844 |
(Item) postfixTuple, postfixTuple->size, |
| 1845 |
NULL, false); |
| 1846 |
MarkBufferDirty(newBuffer); |
| 1847 |
xlrec.postfixBlkSame = false; |
| 1848 |
} |
| 1849 |
|
| 1850 |
/* |
| 1851 |
* And set downlink pointer in the prefix tuple to point to postfix tuple. |
| 1852 |
* (We can't avoid this step by doing the above two steps in opposite |
| 1853 |
* order, because there might not be enough space on the page to insert |
| 1854 |
* the postfix tuple first.) We have to update the local copy of the |
| 1855 |
* prefixTuple too, because that's what will be written to WAL. |
| 1856 |
*/ |
| 1857 |
spgUpdateNodeLink(prefixTuple, out->result.splitTuple.childNodeN, |
| 1858 |
postfixBlkno, postfixOffset); |
| 1859 |
prefixTuple = (SpGistInnerTuple) PageGetItem(current->page, |
| 1860 |
PageGetItemId(current->page, current->offnum)); |
| 1861 |
spgUpdateNodeLink(prefixTuple, out->result.splitTuple.childNodeN, |
| 1862 |
postfixBlkno, postfixOffset); |
| 1863 |
|
| 1864 |
MarkBufferDirty(current->buffer); |
| 1865 |
|
| 1866 |
if (RelationNeedsWAL(index) && !state->isBuild) |
| 1867 |
{ |
| 1868 |
XLogRecPtr recptr; |
| 1869 |
|
| 1870 |
XLogBeginInsert(); |
| 1871 |
XLogRegisterData(&xlrec, sizeof(xlrec)); |
| 1872 |
XLogRegisterData(prefixTuple, prefixTuple->size); |
| 1873 |
XLogRegisterData(postfixTuple, postfixTuple->size); |
| 1874 |
|
| 1875 |
XLogRegisterBuffer(0, current->buffer, REGBUF_STANDARD); |
| 1876 |
if (newBuffer != InvalidBuffer) |
| 1877 |
{ |
| 1878 |
int flags; |
| 1879 |
|
| 1880 |
flags = REGBUF_STANDARD; |
| 1881 |
if (xlrec.newPage) |
| 1882 |
flags |= REGBUF_WILL_INIT; |
| 1883 |
XLogRegisterBuffer(1, newBuffer, flags); |
| 1884 |
} |
| 1885 |
|
| 1886 |
recptr = XLogInsert(RM_SPGIST_ID, XLOG_SPGIST_SPLIT_TUPLE); |
| 1887 |
|
| 1888 |
PageSetLSN(current->page, recptr); |
| 1889 |
|
| 1890 |
if (newBuffer != InvalidBuffer) |
| 1891 |
{ |
| 1892 |
PageSetLSN(BufferGetPage(newBuffer), recptr); |
| 1893 |
} |
| 1894 |
} |
| 1895 |
|
| 1896 |
END_CRIT_SECTION(); |
| 1897 |
|
| 1898 |
/* Update local free-space cache and release buffer */ |
| 1899 |
if (newBuffer != InvalidBuffer) |
| 1900 |
{ |
| 1901 |
SpGistSetLastUsedPage(index, newBuffer); |
| 1902 |
UnlockReleaseBuffer(newBuffer); |
| 1903 |
} |
| 1904 |
} |
| 1905 |
|
| 1906 |
/* |
| 1907 |
* Insert one item into the index. |
| 1908 |
* |
| 1909 |
* Returns true on success, false if we failed to complete the insertion |
| 1910 |
* (typically because of conflict with a concurrent insert). In the latter |
| 1911 |
* case, caller should re-call spgdoinsert() with the same args. |
| 1912 |
*/ |
| 1913 |
bool |
| 1914 |
spgdoinsert(Relation index, SpGistState *state, |
| 1915 |
ItemPointer heapPtr, Datum *datums, bool *isnulls) |
| 1916 |
{ |
| 1917 |
bool result = true; |
| 1918 |
TupleDesc leafDescriptor = state->leafTupDesc; |
| 1919 |
bool isnull = isnulls[spgKeyColumn]; |
| 1920 |
int level = 0; |
| 1921 |
Datum leafDatums[INDEX_MAX_KEYS]; |
| 1922 |
int leafSize; |
| 1923 |
int bestLeafSize; |
| 1924 |
int numNoProgressCycles = 0; |
| 1925 |
SPPageDesc current, |
| 1926 |
parent; |
| 1927 |
FmgrInfo *procinfo = NULL; |
| 1928 |
|
| 1929 |
/* |
| 1930 |
* Look up FmgrInfo of the user-defined choose function once, to save |
| 1931 |
* cycles in the loop below. |
| 1932 |
*/ |
| 1933 |
if (!isnull) |
| 1934 |
procinfo = index_getprocinfo(index, 1, SPGIST_CHOOSE_PROC); |
| 1935 |
|
| 1936 |
/* |
| 1937 |
* Prepare the leaf datum to insert. |
| 1938 |
* |
| 1939 |
* If an optional "compress" method is provided, then call it to form the |
| 1940 |
* leaf key datum from the input datum. Otherwise, store the input datum |
| 1941 |
* as is. Since we don't use index_form_tuple in this AM, we have to make |
| 1942 |
* sure value to be inserted is not toasted; FormIndexDatum doesn't |
| 1943 |
* guarantee that. But we assume the "compress" method to return an |
| 1944 |
* untoasted value. |
| 1945 |
*/ |
| 1946 |
if (!isnull) |
| 1947 |
{ |
| 1948 |
if (OidIsValid(index_getprocid(index, 1, SPGIST_COMPRESS_PROC))) |
| 1949 |
{ |
| 1950 |
FmgrInfo *compressProcinfo = NULL; |
| 1951 |
|
| 1952 |
compressProcinfo = index_getprocinfo(index, 1, SPGIST_COMPRESS_PROC); |
| 1953 |
leafDatums[spgKeyColumn] = |
| 1954 |
FunctionCall1Coll(compressProcinfo, |
| 1955 |
index->rd_indcollation[spgKeyColumn], |
| 1956 |
datums[spgKeyColumn]); |
| 1957 |
} |
| 1958 |
else |
| 1959 |
{ |
| 1960 |
Assert(state->attLeafType.type == state->attType.type); |
| 1961 |
|
| 1962 |
if (state->attType.attlen == -1) |
| 1963 |
leafDatums[spgKeyColumn] = |
| 1964 |
PointerGetDatum(PG_DETOAST_DATUM(datums[spgKeyColumn])); |
| 1965 |
else |
| 1966 |
leafDatums[spgKeyColumn] = datums[spgKeyColumn]; |
| 1967 |
} |
| 1968 |
} |
| 1969 |
else |
| 1970 |
leafDatums[spgKeyColumn] = (Datum) 0; |
| 1971 |
|
| 1972 |
/* Likewise, ensure that any INCLUDE values are not toasted */ |
| 1973 |
for (int i = spgFirstIncludeColumn; i < leafDescriptor->natts; i++) |
| 1974 |
{ |
| 1975 |
if (!isnulls[i]) |
| 1976 |
{ |
| 1977 |
if (TupleDescCompactAttr(leafDescriptor, i)->attlen == -1) |
| 1978 |
leafDatums[i] = PointerGetDatum(PG_DETOAST_DATUM(datums[i])); |
| 1979 |
else |
| 1980 |
leafDatums[i] = datums[i]; |
| 1981 |
} |
| 1982 |
else |
| 1983 |
leafDatums[i] = (Datum) 0; |
| 1984 |
} |
| 1985 |
|
| 1986 |
/* |
| 1987 |
* Compute space needed for a leaf tuple containing the given data. |
| 1988 |
*/ |
| 1989 |
leafSize = SpGistGetLeafTupleSize(leafDescriptor, leafDatums, isnulls); |
| 1990 |
/* Account for an item pointer, too */ |
| 1991 |
leafSize += sizeof(ItemIdData); |
| 1992 |
|
| 1993 |
/* |
| 1994 |
* If it isn't gonna fit, and the opclass can't reduce the datum size by |
| 1995 |
* suffixing, bail out now rather than doing a lot of useless work. |
| 1996 |
*/ |
| 1997 |
if (leafSize > SPGIST_PAGE_CAPACITY && |
| 1998 |
(isnull || !state->config.longValuesOK)) |
| 1999 |
ereport(ERROR, |
| 2000 |
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
| 2001 |
errmsg("index row size %zu exceeds maximum %zu for index \"%s\"", |
| 2002 |
leafSize - sizeof(ItemIdData), |
| 2003 |
SPGIST_PAGE_CAPACITY - sizeof(ItemIdData), |
| 2004 |
RelationGetRelationName(index)), |
| 2005 |
errhint("Values larger than a buffer page cannot be indexed."))); |
| 2006 |
bestLeafSize = leafSize; |
| 2007 |
|
| 2008 |
/* Initialize "current" to the appropriate root page */ |
| 2009 |
current.blkno = isnull ? SPGIST_NULL_BLKNO : SPGIST_ROOT_BLKNO; |
| 2010 |
current.buffer = InvalidBuffer; |
| 2011 |
current.page = NULL; |
| 2012 |
current.offnum = FirstOffsetNumber; |
| 2013 |
current.node = -1; |
| 2014 |
|
| 2015 |
/* "parent" is invalid for the moment */ |
| 2016 |
parent.blkno = InvalidBlockNumber; |
| 2017 |
parent.buffer = InvalidBuffer; |
| 2018 |
parent.page = NULL; |
| 2019 |
parent.offnum = InvalidOffsetNumber; |
| 2020 |
parent.node = -1; |
| 2021 |
|
| 2022 |
/* |
| 2023 |
* Before entering the loop, try to clear any pending interrupt condition. |
| 2024 |
* If a query cancel is pending, we might as well accept it now not later; |
| 2025 |
* while if a non-canceling condition is pending, servicing it here avoids |
| 2026 |
* having to restart the insertion and redo all the work so far. |
| 2027 |
*/ |
| 2028 |
CHECK_FOR_INTERRUPTS(); |
| 2029 |
|
| 2030 |
for (;;) |
| 2031 |
{ |
| 2032 |
bool isNew = false; |
| 2033 |
|
| 2034 |
/* |
| 2035 |
* Bail out if query cancel is pending. We must have this somewhere |
| 2036 |
* in the loop since a broken opclass could produce an infinite |
| 2037 |
* picksplit loop. However, because we'll be holding buffer lock(s) |
| 2038 |
* after the first iteration, ProcessInterrupts() wouldn't be able to |
| 2039 |
* throw a cancel error here. Hence, if we see that an interrupt is |
| 2040 |
* pending, break out of the loop and deal with the situation below. |
| 2041 |
* Set result = false because we must restart the insertion if the |
| 2042 |
* interrupt isn't a query-cancel-or-die case. |
| 2043 |
*/ |
| 2044 |
if (INTERRUPTS_PENDING_CONDITION()) |
| 2045 |
{ |
| 2046 |
result = false; |
| 2047 |
break; |
| 2048 |
} |
| 2049 |
|
| 2050 |
if (current.blkno == InvalidBlockNumber) |
| 2051 |
{ |
| 2052 |
/* |
| 2053 |
* Create a leaf page. If leafSize is too large to fit on a page, |
| 2054 |
* we won't actually use the page yet, but it simplifies the API |
| 2055 |
* for doPickSplit to always have a leaf page at hand; so just |
| 2056 |
* quietly limit our request to a page size. |
| 2057 |
*/ |
| 2058 |
current.buffer = |
| 2059 |
SpGistGetBuffer(index, |
| 2060 |
GBUF_LEAF | (isnull ? GBUF_NULLS : 0), |
| 2061 |
Min(leafSize, SPGIST_PAGE_CAPACITY), |
| 2062 |
&isNew); |
| 2063 |
current.blkno = BufferGetBlockNumber(current.buffer); |
| 2064 |
} |
| 2065 |
else if (parent.buffer == InvalidBuffer) |
| 2066 |
{ |
| 2067 |
/* we hold no parent-page lock, so no deadlock is possible */ |
| 2068 |
current.buffer = ReadBuffer(index, current.blkno); |
| 2069 |
LockBuffer(current.buffer, BUFFER_LOCK_EXCLUSIVE); |
| 2070 |
} |
| 2071 |
else if (current.blkno != parent.blkno) |
| 2072 |
{ |
| 2073 |
/* descend to a new child page */ |
| 2074 |
current.buffer = ReadBuffer(index, current.blkno); |
| 2075 |
|
| 2076 |
/* |
| 2077 |
* Attempt to acquire lock on child page. We must beware of |
| 2078 |
* deadlock against another insertion process descending from that |
| 2079 |
* page to our parent page (see README). If we fail to get lock, |
| 2080 |
* abandon the insertion and tell our caller to start over. |
| 2081 |
* |
| 2082 |
* XXX this could be improved, because failing to get lock on a |
| 2083 |
* buffer is not proof of a deadlock situation; the lock might be |
| 2084 |
* held by a reader, or even just background writer/checkpointer |
| 2085 |
* process. Perhaps it'd be worth retrying after sleeping a bit? |
| 2086 |
*/ |
| 2087 |
if (!ConditionalLockBuffer(current.buffer)) |
| 2088 |
{ |
| 2089 |
ReleaseBuffer(current.buffer); |
| 2090 |
UnlockReleaseBuffer(parent.buffer); |
| 2091 |
return false; |
| 2092 |
} |
| 2093 |
} |
| 2094 |
else |
| 2095 |
{ |
| 2096 |
/* inner tuple can be stored on the same page as parent one */ |
| 2097 |
current.buffer = parent.buffer; |
| 2098 |
} |
| 2099 |
current.page = BufferGetPage(current.buffer); |
| 2100 |
|
| 2101 |
/* should not arrive at a page of the wrong type */ |
| 2102 |
if (isnull ? !SpGistPageStoresNulls(current.page) : |
| 2103 |
SpGistPageStoresNulls(current.page)) |
| 2104 |
elog(ERROR, "SPGiST index page %u has wrong nulls flag", |
| 2105 |
current.blkno); |
| 2106 |
|
| 2107 |
if (SpGistPageIsLeaf(current.page)) |
| 2108 |
{ |
| 2109 |
SpGistLeafTuple leafTuple; |
| 2110 |
int nToSplit, |
| 2111 |
sizeToSplit; |
| 2112 |
|
| 2113 |
leafTuple = spgFormLeafTuple(state, heapPtr, leafDatums, isnulls); |
| 2114 |
if (leafTuple->size + sizeof(ItemIdData) <= |
| 2115 |
SpGistPageGetFreeSpace(current.page, 1)) |
| 2116 |
{ |
| 2117 |
/* it fits on page, so insert it and we're done */ |
| 2118 |
addLeafTuple(index, state, leafTuple, |
| 2119 |
¤t, &parent, isnull, isNew); |
| 2120 |
break; |
| 2121 |
} |
| 2122 |
else if ((sizeToSplit = |
| 2123 |
checkSplitConditions(index, state, ¤t, |
| 2124 |
&nToSplit)) < SPGIST_PAGE_CAPACITY / 2 && |
| 2125 |
nToSplit < 64 && |
| 2126 |
leafTuple->size + sizeof(ItemIdData) + sizeToSplit <= SPGIST_PAGE_CAPACITY) |
| 2127 |
{ |
| 2128 |
/* |
| 2129 |
* the amount of data is pretty small, so just move the whole |
| 2130 |
* chain to another leaf page rather than splitting it. |
| 2131 |
*/ |
| 2132 |
Assert(!isNew); |
| 2133 |
moveLeafs(index, state, ¤t, &parent, leafTuple, isnull); |
| 2134 |
break; /* we're done */ |
| 2135 |
} |
| 2136 |
else |
| 2137 |
{ |
| 2138 |
/* picksplit */ |
| 2139 |
if (doPickSplit(index, state, ¤t, &parent, |
| 2140 |
leafTuple, level, isnull, isNew)) |
| 2141 |
break; /* doPickSplit installed new tuples */ |
| 2142 |
|
| 2143 |
/* leaf tuple will not be inserted yet */ |
| 2144 |
pfree(leafTuple); |
| 2145 |
|
| 2146 |
/* |
| 2147 |
* current now describes new inner tuple, go insert into it |
| 2148 |
*/ |
| 2149 |
Assert(!SpGistPageIsLeaf(current.page)); |
| 2150 |
goto process_inner_tuple; |
| 2151 |
} |
| 2152 |
} |
| 2153 |
else /* non-leaf page */ |
| 2154 |
{ |
| 2155 |
/* |
| 2156 |
* Apply the opclass choose function to figure out how to insert |
| 2157 |
* the given datum into the current inner tuple. |
| 2158 |
*/ |
| 2159 |
SpGistInnerTuple innerTuple; |
| 2160 |
spgChooseIn in; |
| 2161 |
spgChooseOut out; |
| 2162 |
|
| 2163 |
/* |
| 2164 |
* spgAddNode and spgSplitTuple cases will loop back to here to |
| 2165 |
* complete the insertion operation. Just in case the choose |
| 2166 |
* function is broken and produces add or split requests |
| 2167 |
* repeatedly, check for query cancel (see comments above). |
| 2168 |
*/ |
| 2169 |
process_inner_tuple: |
| 2170 |
if (INTERRUPTS_PENDING_CONDITION()) |
| 2171 |
{ |
| 2172 |
result = false; |
| 2173 |
break; |
| 2174 |
} |
| 2175 |
|
| 2176 |
innerTuple = (SpGistInnerTuple) PageGetItem(current.page, |
| 2177 |
PageGetItemId(current.page, current.offnum)); |
| 2178 |
|
| 2179 |
in.datum = datums[spgKeyColumn]; |
| 2180 |
in.leafDatum = leafDatums[spgKeyColumn]; |
| 2181 |
in.level = level; |
| 2182 |
in.allTheSame = innerTuple->allTheSame; |
| 2183 |
in.hasPrefix = (innerTuple->prefixSize > 0); |
| 2184 |
in.prefixDatum = SGITDATUM(innerTuple, state); |
| 2185 |
in.nNodes = innerTuple->nNodes; |
| 2186 |
in.nodeLabels = spgExtractNodeLabels(state, innerTuple); |
| 2187 |
|
| 2188 |
memset(&out, 0, sizeof(out)); |
| 2189 |
|
| 2190 |
if (!isnull) |
| 2191 |
{ |
| 2192 |
/* use user-defined choose method */ |
| 2193 |
FunctionCall2Coll(procinfo, |
| 2194 |
index->rd_indcollation[0], |
| 2195 |
PointerGetDatum(&in), |
| 2196 |
PointerGetDatum(&out)); |
| 2197 |
} |
| 2198 |
else |
| 2199 |
{ |
| 2200 |
/* force "match" action (to insert to random subnode) */ |
| 2201 |
out.resultType = spgMatchNode; |
| 2202 |
} |
| 2203 |
|
| 2204 |
if (innerTuple->allTheSame) |
| 2205 |
{ |
| 2206 |
/* |
| 2207 |
* It's not allowed to do an AddNode at an allTheSame tuple. |
| 2208 |
* Opclass must say "match", in which case we choose a random |
| 2209 |
* one of the nodes to descend into, or "split". |
| 2210 |
*/ |
| 2211 |
if (out.resultType == spgAddNode) |
| 2212 |
elog(ERROR, "cannot add a node to an allTheSame inner tuple"); |
| 2213 |
else if (out.resultType == spgMatchNode) |
| 2214 |
out.result.matchNode.nodeN = |
| 2215 |
pg_prng_uint64_range(&pg_global_prng_state, |
| 2216 |
0, innerTuple->nNodes - 1); |
| 2217 |
} |
| 2218 |
|
| 2219 |
switch (out.resultType) |
| 2220 |
{ |
| 2221 |
case spgMatchNode: |
| 2222 |
/* Descend to N'th child node */ |
| 2223 |
spgMatchNodeAction(index, state, innerTuple, |
| 2224 |
¤t, &parent, |
| 2225 |
out.result.matchNode.nodeN); |
| 2226 |
/* Adjust level as per opclass request */ |
| 2227 |
level += out.result.matchNode.levelAdd; |
| 2228 |
/* Replace leafDatum and recompute leafSize */ |
| 2229 |
if (!isnull) |
| 2230 |
{ |
| 2231 |
leafDatums[spgKeyColumn] = out.result.matchNode.restDatum; |
| 2232 |
leafSize = SpGistGetLeafTupleSize(leafDescriptor, |
| 2233 |
leafDatums, isnulls); |
| 2234 |
leafSize += sizeof(ItemIdData); |
| 2235 |
} |
| 2236 |
|
| 2237 |
/* |
| 2238 |
* Check new tuple size; fail if it can't fit, unless the |
| 2239 |
* opclass says it can handle the situation by suffixing. |
| 2240 |
* |
| 2241 |
* However, the opclass can only shorten the leaf datum, |
| 2242 |
* which may not be enough to ever make the tuple fit, |
| 2243 |
* since INCLUDE columns might alone use more than a page. |
| 2244 |
* Depending on the opclass' behavior, that could lead to |
| 2245 |
* an infinite loop --- spgtextproc.c, for example, will |
| 2246 |
* just repeatedly generate an empty-string leaf datum |
| 2247 |
* once it runs out of data. Actual bugs in opclasses |
| 2248 |
* might cause infinite looping, too. To detect such a |
| 2249 |
* loop, check to see if we are making progress by |
| 2250 |
* reducing the leafSize in each pass. This is a bit |
| 2251 |
* tricky though. Because of alignment considerations, |
| 2252 |
* the total tuple size might not decrease on every pass. |
| 2253 |
* Also, there are edge cases where the choose method |
| 2254 |
* might seem to not make progress for a cycle or two. |
| 2255 |
* Somewhat arbitrarily, we allow up to 10 no-progress |
| 2256 |
* iterations before failing. (This limit should be more |
| 2257 |
* than MAXALIGN, to accommodate opclasses that trim one |
| 2258 |
* byte from the leaf datum per pass.) |
| 2259 |
*/ |
| 2260 |
if (leafSize > SPGIST_PAGE_CAPACITY) |
| 2261 |
{ |
| 2262 |
bool ok = false; |
| 2263 |
|
| 2264 |
if (state->config.longValuesOK && !isnull) |
| 2265 |
{ |
| 2266 |
if (leafSize < bestLeafSize) |
| 2267 |
{ |
| 2268 |
ok = true; |
| 2269 |
bestLeafSize = leafSize; |
| 2270 |
numNoProgressCycles = 0; |
| 2271 |
} |
| 2272 |
else if (++numNoProgressCycles < 10) |
| 2273 |
ok = true; |
| 2274 |
} |
| 2275 |
if (!ok) |
| 2276 |
ereport(ERROR, |
| 2277 |
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
| 2278 |
errmsg("index row size %zu exceeds maximum %zu for index \"%s\"", |
| 2279 |
leafSize - sizeof(ItemIdData), |
| 2280 |
SPGIST_PAGE_CAPACITY - sizeof(ItemIdData), |
| 2281 |
RelationGetRelationName(index)), |
| 2282 |
errhint("Values larger than a buffer page cannot be indexed."))); |
| 2283 |
} |
| 2284 |
|
| 2285 |
/* |
| 2286 |
* Loop around and attempt to insert the new leafDatum at |
| 2287 |
* "current" (which might reference an existing child |
| 2288 |
* tuple, or might be invalid to force us to find a new |
| 2289 |
* page for the tuple). |
| 2290 |
*/ |
| 2291 |
break; |
| 2292 |
case spgAddNode: |
| 2293 |
/* AddNode is not sensible if nodes don't have labels */ |
| 2294 |
if (in.nodeLabels == NULL) |
| 2295 |
elog(ERROR, "cannot add a node to an inner tuple without node labels"); |
| 2296 |
/* Add node to inner tuple, per request */ |
| 2297 |
spgAddNodeAction(index, state, innerTuple, |
| 2298 |
¤t, &parent, |
| 2299 |
out.result.addNode.nodeN, |
| 2300 |
out.result.addNode.nodeLabel); |
| 2301 |
|
| 2302 |
/* |
| 2303 |
* Retry insertion into the enlarged node. We assume that |
| 2304 |
* we'll get a MatchNode result this time. |
| 2305 |
*/ |
| 2306 |
goto process_inner_tuple; |
| 2307 |
break; |
| 2308 |
case spgSplitTuple: |
| 2309 |
/* Split inner tuple, per request */ |
| 2310 |
spgSplitNodeAction(index, state, innerTuple, |
| 2311 |
¤t, &out); |
| 2312 |
|
| 2313 |
/* Retry insertion into the split node */ |
| 2314 |
goto process_inner_tuple; |
| 2315 |
break; |
| 2316 |
default: |
| 2317 |
elog(ERROR, "unrecognized SPGiST choose result: %d", |
| 2318 |
(int) out.resultType); |
| 2319 |
break; |
| 2320 |
} |
| 2321 |
} |
| 2322 |
} /* end loop */ |
| 2323 |
|
| 2324 |
/* |
| 2325 |
* Release any buffers we're still holding. Beware of possibility that |
| 2326 |
* current and parent reference same buffer. |
| 2327 |
*/ |
| 2328 |
if (current.buffer != InvalidBuffer) |
| 2329 |
{ |
| 2330 |
SpGistSetLastUsedPage(index, current.buffer); |
| 2331 |
UnlockReleaseBuffer(current.buffer); |
| 2332 |
} |
| 2333 |
if (parent.buffer != InvalidBuffer && |
| 2334 |
parent.buffer != current.buffer) |
| 2335 |
{ |
| 2336 |
SpGistSetLastUsedPage(index, parent.buffer); |
| 2337 |
UnlockReleaseBuffer(parent.buffer); |
| 2338 |
} |
| 2339 |
|
| 2340 |
/* |
| 2341 |
* We do not support being called while some outer function is holding a |
| 2342 |
* buffer lock (or any other reason to postpone query cancels). If that |
| 2343 |
* were the case, telling the caller to retry would create an infinite |
| 2344 |
* loop. |
| 2345 |
*/ |
| 2346 |
Assert(INTERRUPTS_CAN_BE_PROCESSED()); |
| 2347 |
|
| 2348 |
/* |
| 2349 |
* Finally, check for interrupts again. If there was a query cancel, |
| 2350 |
* ProcessInterrupts() will be able to throw the error here. If it was |
| 2351 |
* some other kind of interrupt that can just be cleared, return false to |
| 2352 |
* tell our caller to retry. |
| 2353 |
*/ |
| 2354 |
CHECK_FOR_INTERRUPTS(); |
| 2355 |
|
| 2356 |
return result; |
| 2357 |
} |