1 /*-------------------------------------------------------------------------
4 * private declarations for GiST -- declarations related to the
5 * internal implementation of GiST, not the public API
7 * Portions Copyright (c) 1996-2014, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1994, Regents of the University of California
10 * src/include/access/gist_private.h
12 *-------------------------------------------------------------------------
14 #ifndef GIST_PRIVATE_H
15 #define GIST_PRIVATE_H
17 #include "access/gist.h"
18 #include "access/itup.h"
20 #include "storage/bufmgr.h"
21 #include "storage/buffile.h"
22 #include "utils/rbtree.h"
23 #include "utils/hsearch.h"
26 * Maximum number of "halves" a page can be split into in one operation.
27 * Typically a split produces 2 halves, but can be more if keys have very
28 * different lengths, or when inserting multiple keys in one operation (as
29 * when inserting downlinks to an internal node). There is no theoretical
30 * limit on this, but in practice if you get more than a handful page halves
31 * in one split, there's something wrong with the opclass implementation.
32 * GIST_MAX_SPLIT_PAGES is an arbitrary limit on that, used to size some
33 * local arrays used during split. Note that there is also a limit on the
34 * number of buffers that can be held locked at a time, MAX_SIMUL_LWLOCKS,
35 * so if you raise this higher than that limit, you'll just get a different
38 #define GIST_MAX_SPLIT_PAGES 75
40 /* Buffer lock modes */
41 #define GIST_SHARE BUFFER_LOCK_SHARE
42 #define GIST_EXCLUSIVE BUFFER_LOCK_EXCLUSIVE
43 #define GIST_UNLOCK BUFFER_LOCK_UNLOCK
52 #define BUFFER_PAGE_DATA_OFFSET MAXALIGN(offsetof(GISTNodeBufferPage, tupledata))
53 /* Returns free space in node buffer page */
54 #define PAGE_FREE_SPACE(nbp) (nbp->freespace)
55 /* Checks if node buffer page is empty */
56 #define PAGE_IS_EMPTY(nbp) (nbp->freespace == BLCKSZ - BUFFER_PAGE_DATA_OFFSET)
57 /* Checks if node buffers page don't contain sufficient space for index tuple */
58 #define PAGE_NO_SPACE(nbp, itup) (PAGE_FREE_SPACE(nbp) < \
59 MAXALIGN(IndexTupleSize(itup)))
62 * GISTSTATE: information needed for any GiST index operation
64 * This struct retains call info for the index's opclass-specific support
65 * functions (per index column), plus the index's tuple descriptor.
67 * scanCxt holds the GISTSTATE itself as well as any data that lives for the
68 * lifetime of the index operation. We pass this to the support functions
69 * via fn_mcxt, so that they can store scan-lifespan data in it. The
70 * functions are invoked in tempCxt, which is typically short-lifespan
71 * (that is, it's reset after each tuple). However, tempCxt can be the same
72 * as scanCxt if we're not bothering with per-tuple context resets.
74 typedef struct GISTSTATE
76 MemoryContext scanCxt; /* context for scan-lifespan data */
77 MemoryContext tempCxt; /* short-term context for calling functions */
79 TupleDesc tupdesc; /* index's tuple descriptor */
81 FmgrInfo consistentFn[INDEX_MAX_KEYS];
82 FmgrInfo unionFn[INDEX_MAX_KEYS];
83 FmgrInfo compressFn[INDEX_MAX_KEYS];
84 FmgrInfo decompressFn[INDEX_MAX_KEYS];
85 FmgrInfo penaltyFn[INDEX_MAX_KEYS];
86 FmgrInfo picksplitFn[INDEX_MAX_KEYS];
87 FmgrInfo equalFn[INDEX_MAX_KEYS];
88 FmgrInfo distanceFn[INDEX_MAX_KEYS];
90 /* Collations to pass to the support functions */
91 Oid supportCollation[INDEX_MAX_KEYS];
96 * During a GiST index search, we must maintain a queue of unvisited items,
97 * which can be either individual heap tuples or whole index pages. If it
98 * is an ordered search, the unvisited items should be visited in distance
99 * order. Unvisited items at the same distance should be visited in
100 * depth-first order, that is heap items first, then lower index pages, then
101 * upper index pages; this rule avoids doing extra work during a search that
102 * ends early due to LIMIT.
104 * To perform an ordered search, we use an RBTree to manage the distance-order
105 * queue. Each GISTSearchTreeItem stores all unvisited items of the same
106 * distance; they are GISTSearchItems chained together via their next fields.
108 * In a non-ordered search (no order-by operators), the RBTree degenerates
109 * to a single item, which we use as a queue of unvisited index pages only.
110 * In this case matched heap items from the current index leaf page are
111 * remembered in GISTScanOpaqueData.pageData[] and returned directly from
112 * there, instead of building a separate GISTSearchItem for each one.
115 /* Individual heap tuple to be visited */
116 typedef struct GISTSearchHeapItem
118 ItemPointerData heapPtr;
119 bool recheck; /* T if quals must be rechecked */
120 } GISTSearchHeapItem;
122 /* Unvisited item, either index page or heap tuple */
123 typedef struct GISTSearchItem
125 struct GISTSearchItem *next; /* list link */
126 BlockNumber blkno; /* index page number, or InvalidBlockNumber */
129 GistNSN parentlsn; /* parent page's LSN, if index page */
130 /* we must store parentlsn to detect whether a split occurred */
131 GISTSearchHeapItem heap; /* heap info, if heap tuple */
135 #define GISTSearchItemIsHeap(item) ((item).blkno == InvalidBlockNumber)
138 * Within a GISTSearchTreeItem's chain, heap items always appear before
139 * index-page items, since we want to visit heap items first. lastHeap points
140 * to the last heap item in the chain, or is NULL if there are none.
142 typedef struct GISTSearchTreeItem
144 RBNode rbnode; /* this is an RBTree item */
145 GISTSearchItem *head; /* first chain member */
146 GISTSearchItem *lastHeap; /* last heap-tuple member, if any */
147 double distances[1]; /* array with numberOfOrderBys entries */
148 } GISTSearchTreeItem;
150 #define GSTIHDRSZ offsetof(GISTSearchTreeItem, distances)
153 * GISTScanOpaqueData: private state for a scan of a GiST index
155 typedef struct GISTScanOpaqueData
157 GISTSTATE *giststate; /* index information, see above */
158 RBTree *queue; /* queue of unvisited items */
159 MemoryContext queueCxt; /* context holding the queue */
160 bool qual_ok; /* false if qual can never be satisfied */
161 bool firstCall; /* true until first gistgettuple call */
163 GISTSearchTreeItem *curTreeItem; /* current queue item, if any */
165 /* pre-allocated workspace arrays */
166 GISTSearchTreeItem *tmpTreeItem; /* workspace to pass to rb_insert */
167 double *distances; /* output area for gistindex_keytest */
169 /* In a non-ordered search, returnable heap items are stored here: */
170 GISTSearchHeapItem pageData[BLCKSZ / sizeof(IndexTupleData)];
171 OffsetNumber nPageData; /* number of valid items in array */
172 OffsetNumber curPageData; /* next item to return */
173 } GISTScanOpaqueData;
175 typedef GISTScanOpaqueData *GISTScanOpaque;
180 #define XLOG_GIST_PAGE_UPDATE 0x00
181 /* #define XLOG_GIST_NEW_ROOT 0x20 */ /* not used anymore */
182 #define XLOG_GIST_PAGE_SPLIT 0x30
183 /* #define XLOG_GIST_INSERT_COMPLETE 0x40 */ /* not used anymore */
184 #define XLOG_GIST_CREATE_INDEX 0x50
185 /* #define XLOG_GIST_PAGE_DELETE 0x60 */ /* not used anymore */
187 typedef struct gistxlogPageUpdate
193 * If this operation completes a page split, by inserting a downlink for
194 * the split page, leftchild points to the left half of the split.
196 BlockNumber leftchild;
198 /* number of deleted offsets */
202 * follow: 1. todelete OffsetNumbers 2. tuples to insert
204 } gistxlogPageUpdate;
206 typedef struct gistxlogPageSplit
209 BlockNumber origblkno; /* splitted page */
210 BlockNumber origrlink; /* rightlink of the page before split */
211 GistNSN orignsn; /* NSN of the page before split */
212 bool origleaf; /* was splitted page a leaf page? */
214 BlockNumber leftchild; /* like in gistxlogPageUpdate */
215 uint16 npage; /* # of pages in the split */
216 bool markfollowright; /* set F_FOLLOW_RIGHT flags */
219 * follow: 1. gistxlogPage and array of IndexTupleData per page
223 typedef struct gistxlogPage
226 int num; /* number of index tuples following */
229 /* SplitedPageLayout - gistSplit function result */
230 typedef struct SplitedPageLayout
233 IndexTupleData *list;
235 IndexTuple itup; /* union key for page */
236 Page page; /* to operate */
237 Buffer buffer; /* to write after all proceed */
239 struct SplitedPageLayout *next;
243 * GISTInsertStack used for locking buffers and transfer arguments during
246 typedef struct GISTInsertStack
254 * log sequence number from page->lsn to recognize page update and compare
255 * it with page's nsn to recognize page split
259 /* offset of the downlink in the parent page, that points to this page */
260 OffsetNumber downlinkoffnum;
262 /* pointer to parent */
263 struct GISTInsertStack *parent;
266 /* Working state and results for multi-column split logic in gistsplit.c */
267 typedef struct GistSplitVector
269 GIST_SPLITVEC splitVector; /* passed to/from user PickSplit method */
271 Datum spl_lattr[INDEX_MAX_KEYS]; /* Union of subkeys in
272 * splitVector.spl_left */
273 bool spl_lisnull[INDEX_MAX_KEYS];
275 Datum spl_rattr[INDEX_MAX_KEYS]; /* Union of subkeys in
276 * splitVector.spl_right */
277 bool spl_risnull[INDEX_MAX_KEYS];
279 bool *spl_dontcare; /* flags tuples which could go to either side
280 * of the split for zero penalty */
286 Size freespace; /* free space to be left */
288 GISTInsertStack *stack;
291 /* root page of a gist index */
292 #define GIST_ROOT_BLKNO 0
295 * Before PostgreSQL 9.1, we used rely on so-called "invalid tuples" on inner
296 * pages to finish crash recovery of incomplete page splits. If a crash
297 * happened in the middle of a page split, so that the downlink pointers were
298 * not yet inserted, crash recovery inserted a special downlink pointer. The
299 * semantics of an invalid tuple was that it if you encounter one in a scan,
300 * it must always be followed, because we don't know if the tuples on the
301 * child page match or not.
303 * We no longer create such invalid tuples, we now mark the left-half of such
304 * an incomplete split with the F_FOLLOW_RIGHT flag instead, and finish the
305 * split properly the next time we need to insert on that page. To retain
306 * on-disk compatibility for the sake of pg_upgrade, we still store 0xffff as
307 * the offset number of all inner tuples. If we encounter any invalid tuples
308 * with 0xfffe during insertion, we throw an error, though scans still handle
309 * them. You should only encounter invalid tuples if you pg_upgrade a pre-9.1
310 * gist index which already has invalid tuples in it because of a crash. That
311 * should be rare, and you are recommended to REINDEX anyway if you have any
312 * invalid tuples in an index, so throwing an error is as far as we go with
315 #define TUPLE_IS_VALID 0xffff
316 #define TUPLE_IS_INVALID 0xfffe
318 #define GistTupleIsInvalid(itup) ( ItemPointerGetOffsetNumber( &((itup)->t_tid) ) == TUPLE_IS_INVALID )
319 #define GistTupleSetValid(itup) ItemPointerSetOffsetNumber( &((itup)->t_tid), TUPLE_IS_VALID )
325 * A buffer attached to an internal node, used when building an index in
330 BlockNumber nodeBlocknum; /* index block # this buffer is for */
331 int32 blocksCount; /* current # of blocks occupied by buffer */
333 BlockNumber pageBlocknum; /* temporary file block # */
334 GISTNodeBufferPage *pageBuffer; /* in-memory buffer page */
336 /* is this buffer queued for emptying? */
337 bool queuedForEmptying;
339 /* is this a temporary copy, not in the hash table? */
342 int level; /* 0 == leaf */
346 * Does specified level have buffers? (Beware of multiple evaluation of
349 #define LEVEL_HAS_BUFFERS(nlevel, gfbb) \
350 ((nlevel) != 0 && (nlevel) % (gfbb)->levelStep == 0 && \
351 (nlevel) != (gfbb)->rootlevel)
353 /* Is specified buffer at least half-filled (should be queued for emptying)? */
354 #define BUFFER_HALF_FILLED(nodeBuffer, gfbb) \
355 ((nodeBuffer)->blocksCount > (gfbb)->pagesPerBuffer / 2)
358 * Is specified buffer full? Our buffers can actually grow indefinitely,
359 * beyond the "maximum" size, so this just means whether the buffer has grown
360 * beyond the nominal maximum size.
362 #define BUFFER_OVERFLOWED(nodeBuffer, gfbb) \
363 ((nodeBuffer)->blocksCount > (gfbb)->pagesPerBuffer)
366 * Data structure with general information about build buffers.
368 typedef struct GISTBuildBuffers
370 /* Persistent memory context for the buffers and metadata. */
371 MemoryContext context;
373 BufFile *pfile; /* Temporary file to store buffers in */
374 long nFileBlocks; /* Current size of the temporary file */
377 * resizable array of free blocks.
380 int nFreeBlocks; /* # of currently free blocks in the array */
381 int freeBlocksLen; /* current allocated length of the array */
383 /* Hash for buffers by block number */
384 HTAB *nodeBuffersTab;
386 /* List of buffers scheduled for emptying */
387 List *bufferEmptyingQueue;
390 * Parameters to the buffering build algorithm. levelStep determines which
391 * levels in the tree have buffers, and pagesPerBuffer determines how
392 * large each buffer is.
397 /* Array of lists of buffers on each level, for final emptying */
398 List **buffersOnLevels;
399 int buffersOnLevelsLen;
402 * Dynamically-sized array of buffers that currently have their last page
403 * loaded in main memory.
405 GISTNodeBuffer **loadedBuffers;
406 int loadedBuffersCount; /* # of entries in loadedBuffers */
407 int loadedBuffersLen; /* allocated size of loadedBuffers */
409 /* Level of the current root node (= height of the index tree - 1) */
414 * Storage type for GiST's reloptions
416 typedef struct GiSTOptions
418 int32 vl_len_; /* varlena header (do not touch directly!) */
419 int fillfactor; /* page fill factor in percent (0..100) */
420 int bufferingModeOffset; /* use buffering build? */
424 extern Datum gistbuildempty(PG_FUNCTION_ARGS);
425 extern Datum gistinsert(PG_FUNCTION_ARGS);
426 extern MemoryContext createTempGistContext(void);
427 extern GISTSTATE *initGISTstate(Relation index);
428 extern void freeGISTstate(GISTSTATE *giststate);
429 extern void gistdoinsert(Relation r,
432 GISTSTATE *GISTstate);
434 /* A List of these is returned from gistplacetopage() in *splitinfo */
437 Buffer buf; /* the split page "half" */
438 IndexTuple downlink; /* downlink for this half. */
441 extern bool gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
443 IndexTuple *itup, int ntup,
444 OffsetNumber oldoffnum, BlockNumber *newblkno,
449 extern SplitedPageLayout *gistSplit(Relation r, Page page, IndexTuple *itup,
450 int len, GISTSTATE *giststate);
453 extern void gist_redo(XLogRecPtr lsn, XLogRecord *record);
454 extern void gist_desc(StringInfo buf, XLogRecord *record);
455 extern void gist_xlog_startup(void);
456 extern void gist_xlog_cleanup(void);
458 extern XLogRecPtr gistXLogUpdate(RelFileNode node, Buffer buffer,
459 OffsetNumber *todelete, int ntodelete,
460 IndexTuple *itup, int ntup,
463 extern XLogRecPtr gistXLogSplit(RelFileNode node,
464 BlockNumber blkno, bool page_is_leaf,
465 SplitedPageLayout *dist,
466 BlockNumber origrlink, GistNSN oldnsn,
467 Buffer leftchild, bool markfollowright);
470 extern Datum gistgettuple(PG_FUNCTION_ARGS);
471 extern Datum gistgetbitmap(PG_FUNCTION_ARGS);
475 #define GiSTPageSize \
476 ( BLCKSZ - SizeOfPageHeaderData - MAXALIGN(sizeof(GISTPageOpaqueData)) )
478 #define GIST_MIN_FILLFACTOR 10
479 #define GIST_DEFAULT_FILLFACTOR 90
481 extern Datum gistoptions(PG_FUNCTION_ARGS);
482 extern bool gistfitpage(IndexTuple *itvec, int len);
483 extern bool gistnospace(Page page, IndexTuple *itvec, int len, OffsetNumber todelete, Size freespace);
484 extern void gistcheckpage(Relation rel, Buffer buf);
485 extern Buffer gistNewBuffer(Relation r);
486 extern void gistfillbuffer(Page page, IndexTuple *itup, int len,
488 extern IndexTuple *gistextractpage(Page page, int *len /* out */ );
489 extern IndexTuple *gistjoinvector(
490 IndexTuple *itvec, int *len,
491 IndexTuple *additvec, int addlen);
492 extern IndexTupleData *gistfillitupvec(IndexTuple *vec, int veclen, int *memlen);
494 extern IndexTuple gistunion(Relation r, IndexTuple *itvec,
495 int len, GISTSTATE *giststate);
496 extern IndexTuple gistgetadjusted(Relation r,
499 GISTSTATE *giststate);
500 extern IndexTuple gistFormTuple(GISTSTATE *giststate,
501 Relation r, Datum *attdata, bool *isnull, bool newValues);
503 extern OffsetNumber gistchoose(Relation r, Page p,
505 GISTSTATE *giststate);
506 extern void gistcentryinit(GISTSTATE *giststate, int nkey,
507 GISTENTRY *e, Datum k,
509 OffsetNumber o, bool l, bool isNull);
511 extern void GISTInitBuffer(Buffer b, uint32 f);
512 extern void gistdentryinit(GISTSTATE *giststate, int nkey, GISTENTRY *e,
513 Datum k, Relation r, Page pg, OffsetNumber o,
514 bool l, bool isNull);
516 extern float gistpenalty(GISTSTATE *giststate, int attno,
517 GISTENTRY *key1, bool isNull1,
518 GISTENTRY *key2, bool isNull2);
519 extern void gistMakeUnionItVec(GISTSTATE *giststate, IndexTuple *itvec, int len,
520 Datum *attr, bool *isnull);
521 extern bool gistKeyIsEQ(GISTSTATE *giststate, int attno, Datum a, Datum b);
522 extern void gistDeCompressAtt(GISTSTATE *giststate, Relation r, IndexTuple tuple, Page p,
523 OffsetNumber o, GISTENTRY *attdata, bool *isnull);
525 extern void gistMakeUnionKey(GISTSTATE *giststate, int attno,
526 GISTENTRY *entry1, bool isnull1,
527 GISTENTRY *entry2, bool isnull2,
528 Datum *dst, bool *dstisnull);
530 extern XLogRecPtr gistGetFakeLSN(Relation rel);
533 extern Datum gistbulkdelete(PG_FUNCTION_ARGS);
534 extern Datum gistvacuumcleanup(PG_FUNCTION_ARGS);
537 extern void gistSplitByKey(Relation r, Page page, IndexTuple *itup,
538 int len, GISTSTATE *giststate,
543 extern Datum gistbuild(PG_FUNCTION_ARGS);
544 extern void gistValidateBufferingOption(char *value);
546 /* gistbuildbuffers.c */
547 extern GISTBuildBuffers *gistInitBuildBuffers(int pagesPerBuffer, int levelStep,
549 extern GISTNodeBuffer *gistGetNodeBuffer(GISTBuildBuffers *gfbb,
550 GISTSTATE *giststate,
551 BlockNumber blkno, int level);
552 extern void gistPushItupToNodeBuffer(GISTBuildBuffers *gfbb,
553 GISTNodeBuffer *nodeBuffer, IndexTuple item);
554 extern bool gistPopItupFromNodeBuffer(GISTBuildBuffers *gfbb,
555 GISTNodeBuffer *nodeBuffer, IndexTuple *item);
556 extern void gistFreeBuildBuffers(GISTBuildBuffers *gfbb);
557 extern void gistRelocateBuildBuffersOnSplit(GISTBuildBuffers *gfbb,
558 GISTSTATE *giststate, Relation r,
559 int level, Buffer buffer,
561 extern void gistUnloadNodeBuffers(GISTBuildBuffers *gfbb);
563 #endif /* GIST_PRIVATE_H */