1 /*-------------------------------------------------------------------------
4 * private declarations for GiST -- declarations related to the
5 * internal implementation of GiST, not the public API
7 * Portions Copyright (c) 1996-2016, 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/amapi.h"
18 #include "access/gist.h"
19 #include "access/itup.h"
20 #include "access/xlogreader.h"
22 #include "lib/pairingheap.h"
23 #include "storage/bufmgr.h"
24 #include "storage/buffile.h"
25 #include "utils/hsearch.h"
26 #include "access/genam.h"
29 * Maximum number of "halves" a page can be split into in one operation.
30 * Typically a split produces 2 halves, but can be more if keys have very
31 * different lengths, or when inserting multiple keys in one operation (as
32 * when inserting downlinks to an internal node). There is no theoretical
33 * limit on this, but in practice if you get more than a handful page halves
34 * in one split, there's something wrong with the opclass implementation.
35 * GIST_MAX_SPLIT_PAGES is an arbitrary limit on that, used to size some
36 * local arrays used during split. Note that there is also a limit on the
37 * number of buffers that can be held locked at a time, MAX_SIMUL_LWLOCKS,
38 * so if you raise this higher than that limit, you'll just get a different
41 #define GIST_MAX_SPLIT_PAGES 75
43 /* Buffer lock modes */
44 #define GIST_SHARE BUFFER_LOCK_SHARE
45 #define GIST_EXCLUSIVE BUFFER_LOCK_EXCLUSIVE
46 #define GIST_UNLOCK BUFFER_LOCK_UNLOCK
52 char tupledata[FLEXIBLE_ARRAY_MEMBER];
55 #define BUFFER_PAGE_DATA_OFFSET MAXALIGN(offsetof(GISTNodeBufferPage, tupledata))
56 /* Returns free space in node buffer page */
57 #define PAGE_FREE_SPACE(nbp) (nbp->freespace)
58 /* Checks if node buffer page is empty */
59 #define PAGE_IS_EMPTY(nbp) (nbp->freespace == BLCKSZ - BUFFER_PAGE_DATA_OFFSET)
60 /* Checks if node buffers page don't contain sufficient space for index tuple */
61 #define PAGE_NO_SPACE(nbp, itup) (PAGE_FREE_SPACE(nbp) < \
62 MAXALIGN(IndexTupleSize(itup)))
65 * GISTSTATE: information needed for any GiST index operation
67 * This struct retains call info for the index's opclass-specific support
68 * functions (per index column), plus the index's tuple descriptor.
70 * scanCxt holds the GISTSTATE itself as well as any data that lives for the
71 * lifetime of the index operation. We pass this to the support functions
72 * via fn_mcxt, so that they can store scan-lifespan data in it. The
73 * functions are invoked in tempCxt, which is typically short-lifespan
74 * (that is, it's reset after each tuple). However, tempCxt can be the same
75 * as scanCxt if we're not bothering with per-tuple context resets.
77 typedef struct GISTSTATE
79 MemoryContext scanCxt; /* context for scan-lifespan data */
80 MemoryContext tempCxt; /* short-term context for calling functions */
82 TupleDesc tupdesc; /* index's tuple descriptor */
83 TupleDesc fetchTupdesc; /* tuple descriptor for tuples returned in an
86 FmgrInfo consistentFn[INDEX_MAX_KEYS];
87 FmgrInfo unionFn[INDEX_MAX_KEYS];
88 FmgrInfo compressFn[INDEX_MAX_KEYS];
89 FmgrInfo decompressFn[INDEX_MAX_KEYS];
90 FmgrInfo penaltyFn[INDEX_MAX_KEYS];
91 FmgrInfo picksplitFn[INDEX_MAX_KEYS];
92 FmgrInfo equalFn[INDEX_MAX_KEYS];
93 FmgrInfo distanceFn[INDEX_MAX_KEYS];
94 FmgrInfo fetchFn[INDEX_MAX_KEYS];
96 /* Collations to pass to the support functions */
97 Oid supportCollation[INDEX_MAX_KEYS];
102 * During a GiST index search, we must maintain a queue of unvisited items,
103 * which can be either individual heap tuples or whole index pages. If it
104 * is an ordered search, the unvisited items should be visited in distance
105 * order. Unvisited items at the same distance should be visited in
106 * depth-first order, that is heap items first, then lower index pages, then
107 * upper index pages; this rule avoids doing extra work during a search that
108 * ends early due to LIMIT.
110 * To perform an ordered search, we use an RBTree to manage the distance-order
111 * queue. Each GISTSearchTreeItem stores all unvisited items of the same
112 * distance; they are GISTSearchItems chained together via their next fields.
114 * In a non-ordered search (no order-by operators), the RBTree degenerates
115 * to a single item, which we use as a queue of unvisited index pages only.
116 * In this case matched heap items from the current index leaf page are
117 * remembered in GISTScanOpaqueData.pageData[] and returned directly from
118 * there, instead of building a separate GISTSearchItem for each one.
121 /* Individual heap tuple to be visited */
122 typedef struct GISTSearchHeapItem
124 ItemPointerData heapPtr;
125 bool recheck; /* T if quals must be rechecked */
126 bool recheckDistances; /* T if distances must be rechecked */
127 IndexTuple ftup; /* data fetched back from the index, used in
128 * index-only scans */
129 OffsetNumber offnum; /* track offset in page to mark tuple as
131 } GISTSearchHeapItem;
133 /* Unvisited item, either index page or heap tuple */
134 typedef struct GISTSearchItem
136 pairingheap_node phNode;
137 BlockNumber blkno; /* index page number, or InvalidBlockNumber */
140 GistNSN parentlsn; /* parent page's LSN, if index page */
141 /* we must store parentlsn to detect whether a split occurred */
142 GISTSearchHeapItem heap; /* heap info, if heap tuple */
144 double distances[FLEXIBLE_ARRAY_MEMBER]; /* numberOfOrderBys
148 #define GISTSearchItemIsHeap(item) ((item).blkno == InvalidBlockNumber)
150 #define SizeOfGISTSearchItem(n_distances) (offsetof(GISTSearchItem, distances) + sizeof(double) * (n_distances))
153 * GISTScanOpaqueData: private state for a scan of a GiST index
155 typedef struct GISTScanOpaqueData
157 GISTSTATE *giststate; /* index information, see above */
158 Oid *orderByTypes; /* datatypes of ORDER BY expressions */
160 pairingheap *queue; /* queue of unvisited items */
161 MemoryContext queueCxt; /* context holding the queue */
162 bool qual_ok; /* false if qual can never be satisfied */
163 bool firstCall; /* true until first gistgettuple call */
165 /* pre-allocated workspace arrays */
166 double *distances; /* output area for gistindex_keytest */
168 /* info about killed items if any (killedItems is NULL if never used) */
169 OffsetNumber *killedItems; /* offset numbers of killed items */
170 int numKilled; /* number of currently stored items */
171 BlockNumber curBlkno; /* current number of block */
172 GistNSN curPageLSN; /* pos in the WAL stream when page was read */
174 /* In a non-ordered search, returnable heap items are stored here: */
175 GISTSearchHeapItem pageData[BLCKSZ / sizeof(IndexTupleData)];
176 OffsetNumber nPageData; /* number of valid items in array */
177 OffsetNumber curPageData; /* next item to return */
178 MemoryContext pageDataCxt; /* context holding the fetched tuples, for
179 * index-only scans */
180 } GISTScanOpaqueData;
182 typedef GISTScanOpaqueData *GISTScanOpaque;
187 #define XLOG_GIST_PAGE_UPDATE 0x00
188 /* #define XLOG_GIST_NEW_ROOT 0x20 */ /* not used anymore */
189 #define XLOG_GIST_PAGE_SPLIT 0x30
190 /* #define XLOG_GIST_INSERT_COMPLETE 0x40 */ /* not used anymore */
191 #define XLOG_GIST_CREATE_INDEX 0x50
192 /* #define XLOG_GIST_PAGE_DELETE 0x60 */ /* not used anymore */
195 * Backup Blk 0: updated page.
196 * Backup Blk 1: If this operation completes a page split, by inserting a
197 * downlink for the split page, the left half of the split
199 typedef struct gistxlogPageUpdate
201 /* number of deleted offsets */
206 * In payload of blk 0 : 1. todelete OffsetNumbers 2. tuples to insert
208 } gistxlogPageUpdate;
211 * Backup Blk 0: If this operation completes a page split, by inserting a
212 * downlink for the split page, the left half of the split
213 * Backup Blk 1 - npage: split pages (1 is the original page)
215 typedef struct gistxlogPageSplit
217 BlockNumber origrlink; /* rightlink of the page before split */
218 GistNSN orignsn; /* NSN of the page before split */
219 bool origleaf; /* was splitted page a leaf page? */
221 uint16 npage; /* # of pages in the split */
222 bool markfollowright; /* set F_FOLLOW_RIGHT flags */
225 * follow: 1. gistxlogPage and array of IndexTupleData per page
229 typedef struct gistxlogPage
232 int num; /* number of index tuples following */
235 /* SplitedPageLayout - gistSplit function result */
236 typedef struct SplitedPageLayout
239 IndexTupleData *list;
241 IndexTuple itup; /* union key for page */
242 Page page; /* to operate */
243 Buffer buffer; /* to write after all proceed */
245 struct SplitedPageLayout *next;
249 * GISTInsertStack used for locking buffers and transfer arguments during
252 typedef struct GISTInsertStack
260 * log sequence number from page->lsn to recognize page update and compare
261 * it with page's nsn to recognize page split
265 /* offset of the downlink in the parent page, that points to this page */
266 OffsetNumber downlinkoffnum;
268 /* pointer to parent */
269 struct GISTInsertStack *parent;
272 /* Working state and results for multi-column split logic in gistsplit.c */
273 typedef struct GistSplitVector
275 GIST_SPLITVEC splitVector; /* passed to/from user PickSplit method */
277 Datum spl_lattr[INDEX_MAX_KEYS]; /* Union of subkeys in
278 * splitVector.spl_left */
279 bool spl_lisnull[INDEX_MAX_KEYS];
281 Datum spl_rattr[INDEX_MAX_KEYS]; /* Union of subkeys in
282 * splitVector.spl_right */
283 bool spl_risnull[INDEX_MAX_KEYS];
285 bool *spl_dontcare; /* flags tuples which could go to either side
286 * of the split for zero penalty */
292 Size freespace; /* free space to be left */
294 GISTInsertStack *stack;
297 /* root page of a gist index */
298 #define GIST_ROOT_BLKNO 0
301 * Before PostgreSQL 9.1, we used rely on so-called "invalid tuples" on inner
302 * pages to finish crash recovery of incomplete page splits. If a crash
303 * happened in the middle of a page split, so that the downlink pointers were
304 * not yet inserted, crash recovery inserted a special downlink pointer. The
305 * semantics of an invalid tuple was that it if you encounter one in a scan,
306 * it must always be followed, because we don't know if the tuples on the
307 * child page match or not.
309 * We no longer create such invalid tuples, we now mark the left-half of such
310 * an incomplete split with the F_FOLLOW_RIGHT flag instead, and finish the
311 * split properly the next time we need to insert on that page. To retain
312 * on-disk compatibility for the sake of pg_upgrade, we still store 0xffff as
313 * the offset number of all inner tuples. If we encounter any invalid tuples
314 * with 0xfffe during insertion, we throw an error, though scans still handle
315 * them. You should only encounter invalid tuples if you pg_upgrade a pre-9.1
316 * gist index which already has invalid tuples in it because of a crash. That
317 * should be rare, and you are recommended to REINDEX anyway if you have any
318 * invalid tuples in an index, so throwing an error is as far as we go with
321 #define TUPLE_IS_VALID 0xffff
322 #define TUPLE_IS_INVALID 0xfffe
324 #define GistTupleIsInvalid(itup) ( ItemPointerGetOffsetNumber( &((itup)->t_tid) ) == TUPLE_IS_INVALID )
325 #define GistTupleSetValid(itup) ItemPointerSetOffsetNumber( &((itup)->t_tid), TUPLE_IS_VALID )
331 * A buffer attached to an internal node, used when building an index in
336 BlockNumber nodeBlocknum; /* index block # this buffer is for */
337 int32 blocksCount; /* current # of blocks occupied by buffer */
339 BlockNumber pageBlocknum; /* temporary file block # */
340 GISTNodeBufferPage *pageBuffer; /* in-memory buffer page */
342 /* is this buffer queued for emptying? */
343 bool queuedForEmptying;
345 /* is this a temporary copy, not in the hash table? */
348 int level; /* 0 == leaf */
352 * Does specified level have buffers? (Beware of multiple evaluation of
355 #define LEVEL_HAS_BUFFERS(nlevel, gfbb) \
356 ((nlevel) != 0 && (nlevel) % (gfbb)->levelStep == 0 && \
357 (nlevel) != (gfbb)->rootlevel)
359 /* Is specified buffer at least half-filled (should be queued for emptying)? */
360 #define BUFFER_HALF_FILLED(nodeBuffer, gfbb) \
361 ((nodeBuffer)->blocksCount > (gfbb)->pagesPerBuffer / 2)
364 * Is specified buffer full? Our buffers can actually grow indefinitely,
365 * beyond the "maximum" size, so this just means whether the buffer has grown
366 * beyond the nominal maximum size.
368 #define BUFFER_OVERFLOWED(nodeBuffer, gfbb) \
369 ((nodeBuffer)->blocksCount > (gfbb)->pagesPerBuffer)
372 * Data structure with general information about build buffers.
374 typedef struct GISTBuildBuffers
376 /* Persistent memory context for the buffers and metadata. */
377 MemoryContext context;
379 BufFile *pfile; /* Temporary file to store buffers in */
380 long nFileBlocks; /* Current size of the temporary file */
383 * resizable array of free blocks.
386 int nFreeBlocks; /* # of currently free blocks in the array */
387 int freeBlocksLen; /* current allocated length of the array */
389 /* Hash for buffers by block number */
390 HTAB *nodeBuffersTab;
392 /* List of buffers scheduled for emptying */
393 List *bufferEmptyingQueue;
396 * Parameters to the buffering build algorithm. levelStep determines which
397 * levels in the tree have buffers, and pagesPerBuffer determines how
398 * large each buffer is.
403 /* Array of lists of buffers on each level, for final emptying */
404 List **buffersOnLevels;
405 int buffersOnLevelsLen;
408 * Dynamically-sized array of buffers that currently have their last page
409 * loaded in main memory.
411 GISTNodeBuffer **loadedBuffers;
412 int loadedBuffersCount; /* # of entries in loadedBuffers */
413 int loadedBuffersLen; /* allocated size of loadedBuffers */
415 /* Level of the current root node (= height of the index tree - 1) */
420 * Storage type for GiST's reloptions
422 typedef struct GiSTOptions
424 int32 vl_len_; /* varlena header (do not touch directly!) */
425 int fillfactor; /* page fill factor in percent (0..100) */
426 int bufferingModeOffset; /* use buffering build? */
430 extern Datum gisthandler(PG_FUNCTION_ARGS);
431 extern void gistbuildempty(Relation index);
432 extern bool gistinsert(Relation r, Datum *values, bool *isnull,
433 ItemPointer ht_ctid, Relation heapRel,
434 IndexUniqueCheck checkUnique);
435 extern MemoryContext createTempGistContext(void);
436 extern GISTSTATE *initGISTstate(Relation index);
437 extern void freeGISTstate(GISTSTATE *giststate);
438 extern void gistdoinsert(Relation r,
441 GISTSTATE *GISTstate);
443 /* A List of these is returned from gistplacetopage() in *splitinfo */
446 Buffer buf; /* the split page "half" */
447 IndexTuple downlink; /* downlink for this half. */
450 extern bool gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
452 IndexTuple *itup, int ntup,
453 OffsetNumber oldoffnum, BlockNumber *newblkno,
458 extern SplitedPageLayout *gistSplit(Relation r, Page page, IndexTuple *itup,
459 int len, GISTSTATE *giststate);
462 extern void gist_redo(XLogReaderState *record);
463 extern void gist_desc(StringInfo buf, XLogReaderState *record);
464 extern const char *gist_identify(uint8 info);
465 extern void gist_xlog_startup(void);
466 extern void gist_xlog_cleanup(void);
468 extern XLogRecPtr gistXLogUpdate(Buffer buffer,
469 OffsetNumber *todelete, int ntodelete,
470 IndexTuple *itup, int ntup,
473 extern XLogRecPtr gistXLogSplit(bool page_is_leaf,
474 SplitedPageLayout *dist,
475 BlockNumber origrlink, GistNSN oldnsn,
476 Buffer leftchild, bool markfollowright);
479 extern bool gistgettuple(IndexScanDesc scan, ScanDirection dir);
480 extern int64 gistgetbitmap(IndexScanDesc scan, TIDBitmap *tbm);
481 extern bool gistcanreturn(Relation index, int attno);
484 extern bool gistvalidate(Oid opclassoid);
488 #define GiSTPageSize \
489 ( BLCKSZ - SizeOfPageHeaderData - MAXALIGN(sizeof(GISTPageOpaqueData)) )
491 #define GIST_MIN_FILLFACTOR 10
492 #define GIST_DEFAULT_FILLFACTOR 90
494 extern bytea *gistoptions(Datum reloptions, bool validate);
495 extern bool gistfitpage(IndexTuple *itvec, int len);
496 extern bool gistnospace(Page page, IndexTuple *itvec, int len, OffsetNumber todelete, Size freespace);
497 extern void gistcheckpage(Relation rel, Buffer buf);
498 extern Buffer gistNewBuffer(Relation r);
499 extern void gistfillbuffer(Page page, IndexTuple *itup, int len,
501 extern IndexTuple *gistextractpage(Page page, int *len /* out */ );
502 extern IndexTuple *gistjoinvector(
503 IndexTuple *itvec, int *len,
504 IndexTuple *additvec, int addlen);
505 extern IndexTupleData *gistfillitupvec(IndexTuple *vec, int veclen, int *memlen);
507 extern IndexTuple gistunion(Relation r, IndexTuple *itvec,
508 int len, GISTSTATE *giststate);
509 extern IndexTuple gistgetadjusted(Relation r,
512 GISTSTATE *giststate);
513 extern IndexTuple gistFormTuple(GISTSTATE *giststate,
514 Relation r, Datum *attdata, bool *isnull, bool isleaf);
516 extern OffsetNumber gistchoose(Relation r, Page p,
518 GISTSTATE *giststate);
520 extern void GISTInitBuffer(Buffer b, uint32 f);
521 extern void gistdentryinit(GISTSTATE *giststate, int nkey, GISTENTRY *e,
522 Datum k, Relation r, Page pg, OffsetNumber o,
523 bool l, bool isNull);
525 extern float gistpenalty(GISTSTATE *giststate, int attno,
526 GISTENTRY *key1, bool isNull1,
527 GISTENTRY *key2, bool isNull2);
528 extern void gistMakeUnionItVec(GISTSTATE *giststate, IndexTuple *itvec, int len,
529 Datum *attr, bool *isnull);
530 extern bool gistKeyIsEQ(GISTSTATE *giststate, int attno, Datum a, Datum b);
531 extern void gistDeCompressAtt(GISTSTATE *giststate, Relation r, IndexTuple tuple, Page p,
532 OffsetNumber o, GISTENTRY *attdata, bool *isnull);
533 extern IndexTuple gistFetchTuple(GISTSTATE *giststate, Relation r,
535 extern void gistMakeUnionKey(GISTSTATE *giststate, int attno,
536 GISTENTRY *entry1, bool isnull1,
537 GISTENTRY *entry2, bool isnull2,
538 Datum *dst, bool *dstisnull);
540 extern XLogRecPtr gistGetFakeLSN(Relation rel);
543 extern IndexBulkDeleteResult *gistbulkdelete(IndexVacuumInfo *info,
544 IndexBulkDeleteResult *stats,
545 IndexBulkDeleteCallback callback,
546 void *callback_state);
547 extern IndexBulkDeleteResult *gistvacuumcleanup(IndexVacuumInfo *info,
548 IndexBulkDeleteResult *stats);
551 extern void gistSplitByKey(Relation r, Page page, IndexTuple *itup,
552 int len, GISTSTATE *giststate,
557 extern IndexBuildResult *gistbuild(Relation heap, Relation index,
558 struct IndexInfo *indexInfo);
559 extern void gistValidateBufferingOption(char *value);
561 /* gistbuildbuffers.c */
562 extern GISTBuildBuffers *gistInitBuildBuffers(int pagesPerBuffer, int levelStep,
564 extern GISTNodeBuffer *gistGetNodeBuffer(GISTBuildBuffers *gfbb,
565 GISTSTATE *giststate,
566 BlockNumber blkno, int level);
567 extern void gistPushItupToNodeBuffer(GISTBuildBuffers *gfbb,
568 GISTNodeBuffer *nodeBuffer, IndexTuple item);
569 extern bool gistPopItupFromNodeBuffer(GISTBuildBuffers *gfbb,
570 GISTNodeBuffer *nodeBuffer, IndexTuple *item);
571 extern void gistFreeBuildBuffers(GISTBuildBuffers *gfbb);
572 extern void gistRelocateBuildBuffersOnSplit(GISTBuildBuffers *gfbb,
573 GISTSTATE *giststate, Relation r,
574 int level, Buffer buffer,
576 extern void gistUnloadNodeBuffers(GISTBuildBuffers *gfbb);
578 #endif /* GIST_PRIVATE_H */