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/gist.h"
18 #include "access/itup.h"
19 #include "access/xlogreader.h"
21 #include "lib/pairingheap.h"
22 #include "storage/bufmgr.h"
23 #include "storage/buffile.h"
24 #include "utils/hsearch.h"
25 #include "access/genam.h"
28 * Maximum number of "halves" a page can be split into in one operation.
29 * Typically a split produces 2 halves, but can be more if keys have very
30 * different lengths, or when inserting multiple keys in one operation (as
31 * when inserting downlinks to an internal node). There is no theoretical
32 * limit on this, but in practice if you get more than a handful page halves
33 * in one split, there's something wrong with the opclass implementation.
34 * GIST_MAX_SPLIT_PAGES is an arbitrary limit on that, used to size some
35 * local arrays used during split. Note that there is also a limit on the
36 * number of buffers that can be held locked at a time, MAX_SIMUL_LWLOCKS,
37 * so if you raise this higher than that limit, you'll just get a different
40 #define GIST_MAX_SPLIT_PAGES 75
42 /* Buffer lock modes */
43 #define GIST_SHARE BUFFER_LOCK_SHARE
44 #define GIST_EXCLUSIVE BUFFER_LOCK_EXCLUSIVE
45 #define GIST_UNLOCK BUFFER_LOCK_UNLOCK
51 char tupledata[FLEXIBLE_ARRAY_MEMBER];
54 #define BUFFER_PAGE_DATA_OFFSET MAXALIGN(offsetof(GISTNodeBufferPage, tupledata))
55 /* Returns free space in node buffer page */
56 #define PAGE_FREE_SPACE(nbp) (nbp->freespace)
57 /* Checks if node buffer page is empty */
58 #define PAGE_IS_EMPTY(nbp) (nbp->freespace == BLCKSZ - BUFFER_PAGE_DATA_OFFSET)
59 /* Checks if node buffers page don't contain sufficient space for index tuple */
60 #define PAGE_NO_SPACE(nbp, itup) (PAGE_FREE_SPACE(nbp) < \
61 MAXALIGN(IndexTupleSize(itup)))
64 * GISTSTATE: information needed for any GiST index operation
66 * This struct retains call info for the index's opclass-specific support
67 * functions (per index column), plus the index's tuple descriptor.
69 * scanCxt holds the GISTSTATE itself as well as any data that lives for the
70 * lifetime of the index operation. We pass this to the support functions
71 * via fn_mcxt, so that they can store scan-lifespan data in it. The
72 * functions are invoked in tempCxt, which is typically short-lifespan
73 * (that is, it's reset after each tuple). However, tempCxt can be the same
74 * as scanCxt if we're not bothering with per-tuple context resets.
76 typedef struct GISTSTATE
78 MemoryContext scanCxt; /* context for scan-lifespan data */
79 MemoryContext tempCxt; /* short-term context for calling functions */
81 TupleDesc tupdesc; /* index's tuple descriptor */
82 TupleDesc fetchTupdesc; /* tuple descriptor for tuples returned in an
85 FmgrInfo consistentFn[INDEX_MAX_KEYS];
86 FmgrInfo unionFn[INDEX_MAX_KEYS];
87 FmgrInfo compressFn[INDEX_MAX_KEYS];
88 FmgrInfo decompressFn[INDEX_MAX_KEYS];
89 FmgrInfo penaltyFn[INDEX_MAX_KEYS];
90 FmgrInfo picksplitFn[INDEX_MAX_KEYS];
91 FmgrInfo equalFn[INDEX_MAX_KEYS];
92 FmgrInfo distanceFn[INDEX_MAX_KEYS];
93 FmgrInfo fetchFn[INDEX_MAX_KEYS];
95 /* Collations to pass to the support functions */
96 Oid supportCollation[INDEX_MAX_KEYS];
101 * During a GiST index search, we must maintain a queue of unvisited items,
102 * which can be either individual heap tuples or whole index pages. If it
103 * is an ordered search, the unvisited items should be visited in distance
104 * order. Unvisited items at the same distance should be visited in
105 * depth-first order, that is heap items first, then lower index pages, then
106 * upper index pages; this rule avoids doing extra work during a search that
107 * ends early due to LIMIT.
109 * To perform an ordered search, we use an RBTree to manage the distance-order
110 * queue. Each GISTSearchTreeItem stores all unvisited items of the same
111 * distance; they are GISTSearchItems chained together via their next fields.
113 * In a non-ordered search (no order-by operators), the RBTree degenerates
114 * to a single item, which we use as a queue of unvisited index pages only.
115 * In this case matched heap items from the current index leaf page are
116 * remembered in GISTScanOpaqueData.pageData[] and returned directly from
117 * there, instead of building a separate GISTSearchItem for each one.
120 /* Individual heap tuple to be visited */
121 typedef struct GISTSearchHeapItem
123 ItemPointerData heapPtr;
124 bool recheck; /* T if quals must be rechecked */
125 bool recheckDistances; /* T if distances must be rechecked */
126 IndexTuple ftup; /* data fetched back from the index, used in
127 * index-only scans */
128 OffsetNumber offnum; /* track offset in page to mark tuple as
130 } GISTSearchHeapItem;
132 /* Unvisited item, either index page or heap tuple */
133 typedef struct GISTSearchItem
135 pairingheap_node phNode;
136 BlockNumber blkno; /* index page number, or InvalidBlockNumber */
139 GistNSN parentlsn; /* parent page's LSN, if index page */
140 /* we must store parentlsn to detect whether a split occurred */
141 GISTSearchHeapItem heap; /* heap info, if heap tuple */
143 double distances[FLEXIBLE_ARRAY_MEMBER]; /* numberOfOrderBys
147 #define GISTSearchItemIsHeap(item) ((item).blkno == InvalidBlockNumber)
149 #define SizeOfGISTSearchItem(n_distances) (offsetof(GISTSearchItem, distances) + sizeof(double) * (n_distances))
152 * GISTScanOpaqueData: private state for a scan of a GiST index
154 typedef struct GISTScanOpaqueData
156 GISTSTATE *giststate; /* index information, see above */
157 Oid *orderByTypes; /* datatypes of ORDER BY expressions */
159 pairingheap *queue; /* queue of unvisited items */
160 MemoryContext queueCxt; /* context holding the queue */
161 bool qual_ok; /* false if qual can never be satisfied */
162 bool firstCall; /* true until first gistgettuple call */
164 /* pre-allocated workspace arrays */
165 double *distances; /* output area for gistindex_keytest */
167 /* info about killed items if any (killedItems is NULL if never used) */
168 OffsetNumber *killedItems; /* offset numbers of killed items */
169 int numKilled; /* number of currently stored items */
170 BlockNumber curBlkno; /* current number of block */
171 GistNSN curPageLSN; /* pos in the WAL stream when page was read */
173 /* In a non-ordered search, returnable heap items are stored here: */
174 GISTSearchHeapItem pageData[BLCKSZ / sizeof(IndexTupleData)];
175 OffsetNumber nPageData; /* number of valid items in array */
176 OffsetNumber curPageData; /* next item to return */
177 MemoryContext pageDataCxt; /* context holding the fetched tuples, for
178 * index-only scans */
179 } GISTScanOpaqueData;
181 typedef GISTScanOpaqueData *GISTScanOpaque;
186 #define XLOG_GIST_PAGE_UPDATE 0x00
187 /* #define XLOG_GIST_NEW_ROOT 0x20 */ /* not used anymore */
188 #define XLOG_GIST_PAGE_SPLIT 0x30
189 /* #define XLOG_GIST_INSERT_COMPLETE 0x40 */ /* not used anymore */
190 #define XLOG_GIST_CREATE_INDEX 0x50
191 /* #define XLOG_GIST_PAGE_DELETE 0x60 */ /* not used anymore */
194 * Backup Blk 0: updated page.
195 * Backup Blk 1: If this operation completes a page split, by inserting a
196 * downlink for the split page, the left half of the split
198 typedef struct gistxlogPageUpdate
200 /* number of deleted offsets */
205 * In payload of blk 0 : 1. todelete OffsetNumbers 2. tuples to insert
207 } gistxlogPageUpdate;
210 * Backup Blk 0: If this operation completes a page split, by inserting a
211 * downlink for the split page, the left half of the split
212 * Backup Blk 1 - npage: split pages (1 is the original page)
214 typedef struct gistxlogPageSplit
216 BlockNumber origrlink; /* rightlink of the page before split */
217 GistNSN orignsn; /* NSN of the page before split */
218 bool origleaf; /* was splitted page a leaf page? */
220 uint16 npage; /* # of pages in the split */
221 bool markfollowright; /* set F_FOLLOW_RIGHT flags */
224 * follow: 1. gistxlogPage and array of IndexTupleData per page
228 typedef struct gistxlogPage
231 int num; /* number of index tuples following */
234 /* SplitedPageLayout - gistSplit function result */
235 typedef struct SplitedPageLayout
238 IndexTupleData *list;
240 IndexTuple itup; /* union key for page */
241 Page page; /* to operate */
242 Buffer buffer; /* to write after all proceed */
244 struct SplitedPageLayout *next;
248 * GISTInsertStack used for locking buffers and transfer arguments during
251 typedef struct GISTInsertStack
259 * log sequence number from page->lsn to recognize page update and compare
260 * it with page's nsn to recognize page split
264 /* offset of the downlink in the parent page, that points to this page */
265 OffsetNumber downlinkoffnum;
267 /* pointer to parent */
268 struct GISTInsertStack *parent;
271 /* Working state and results for multi-column split logic in gistsplit.c */
272 typedef struct GistSplitVector
274 GIST_SPLITVEC splitVector; /* passed to/from user PickSplit method */
276 Datum spl_lattr[INDEX_MAX_KEYS]; /* Union of subkeys in
277 * splitVector.spl_left */
278 bool spl_lisnull[INDEX_MAX_KEYS];
280 Datum spl_rattr[INDEX_MAX_KEYS]; /* Union of subkeys in
281 * splitVector.spl_right */
282 bool spl_risnull[INDEX_MAX_KEYS];
284 bool *spl_dontcare; /* flags tuples which could go to either side
285 * of the split for zero penalty */
291 Size freespace; /* free space to be left */
293 GISTInsertStack *stack;
296 /* root page of a gist index */
297 #define GIST_ROOT_BLKNO 0
300 * Before PostgreSQL 9.1, we used rely on so-called "invalid tuples" on inner
301 * pages to finish crash recovery of incomplete page splits. If a crash
302 * happened in the middle of a page split, so that the downlink pointers were
303 * not yet inserted, crash recovery inserted a special downlink pointer. The
304 * semantics of an invalid tuple was that it if you encounter one in a scan,
305 * it must always be followed, because we don't know if the tuples on the
306 * child page match or not.
308 * We no longer create such invalid tuples, we now mark the left-half of such
309 * an incomplete split with the F_FOLLOW_RIGHT flag instead, and finish the
310 * split properly the next time we need to insert on that page. To retain
311 * on-disk compatibility for the sake of pg_upgrade, we still store 0xffff as
312 * the offset number of all inner tuples. If we encounter any invalid tuples
313 * with 0xfffe during insertion, we throw an error, though scans still handle
314 * them. You should only encounter invalid tuples if you pg_upgrade a pre-9.1
315 * gist index which already has invalid tuples in it because of a crash. That
316 * should be rare, and you are recommended to REINDEX anyway if you have any
317 * invalid tuples in an index, so throwing an error is as far as we go with
320 #define TUPLE_IS_VALID 0xffff
321 #define TUPLE_IS_INVALID 0xfffe
323 #define GistTupleIsInvalid(itup) ( ItemPointerGetOffsetNumber( &((itup)->t_tid) ) == TUPLE_IS_INVALID )
324 #define GistTupleSetValid(itup) ItemPointerSetOffsetNumber( &((itup)->t_tid), TUPLE_IS_VALID )
330 * A buffer attached to an internal node, used when building an index in
335 BlockNumber nodeBlocknum; /* index block # this buffer is for */
336 int32 blocksCount; /* current # of blocks occupied by buffer */
338 BlockNumber pageBlocknum; /* temporary file block # */
339 GISTNodeBufferPage *pageBuffer; /* in-memory buffer page */
341 /* is this buffer queued for emptying? */
342 bool queuedForEmptying;
344 /* is this a temporary copy, not in the hash table? */
347 int level; /* 0 == leaf */
351 * Does specified level have buffers? (Beware of multiple evaluation of
354 #define LEVEL_HAS_BUFFERS(nlevel, gfbb) \
355 ((nlevel) != 0 && (nlevel) % (gfbb)->levelStep == 0 && \
356 (nlevel) != (gfbb)->rootlevel)
358 /* Is specified buffer at least half-filled (should be queued for emptying)? */
359 #define BUFFER_HALF_FILLED(nodeBuffer, gfbb) \
360 ((nodeBuffer)->blocksCount > (gfbb)->pagesPerBuffer / 2)
363 * Is specified buffer full? Our buffers can actually grow indefinitely,
364 * beyond the "maximum" size, so this just means whether the buffer has grown
365 * beyond the nominal maximum size.
367 #define BUFFER_OVERFLOWED(nodeBuffer, gfbb) \
368 ((nodeBuffer)->blocksCount > (gfbb)->pagesPerBuffer)
371 * Data structure with general information about build buffers.
373 typedef struct GISTBuildBuffers
375 /* Persistent memory context for the buffers and metadata. */
376 MemoryContext context;
378 BufFile *pfile; /* Temporary file to store buffers in */
379 long nFileBlocks; /* Current size of the temporary file */
382 * resizable array of free blocks.
385 int nFreeBlocks; /* # of currently free blocks in the array */
386 int freeBlocksLen; /* current allocated length of the array */
388 /* Hash for buffers by block number */
389 HTAB *nodeBuffersTab;
391 /* List of buffers scheduled for emptying */
392 List *bufferEmptyingQueue;
395 * Parameters to the buffering build algorithm. levelStep determines which
396 * levels in the tree have buffers, and pagesPerBuffer determines how
397 * large each buffer is.
402 /* Array of lists of buffers on each level, for final emptying */
403 List **buffersOnLevels;
404 int buffersOnLevelsLen;
407 * Dynamically-sized array of buffers that currently have their last page
408 * loaded in main memory.
410 GISTNodeBuffer **loadedBuffers;
411 int loadedBuffersCount; /* # of entries in loadedBuffers */
412 int loadedBuffersLen; /* allocated size of loadedBuffers */
414 /* Level of the current root node (= height of the index tree - 1) */
419 * Storage type for GiST's reloptions
421 typedef struct GiSTOptions
423 int32 vl_len_; /* varlena header (do not touch directly!) */
424 int fillfactor; /* page fill factor in percent (0..100) */
425 int bufferingModeOffset; /* use buffering build? */
429 extern Datum gistbuildempty(PG_FUNCTION_ARGS);
430 extern Datum gistinsert(PG_FUNCTION_ARGS);
431 extern Datum gistcanreturn(PG_FUNCTION_ARGS);
432 extern MemoryContext createTempGistContext(void);
433 extern GISTSTATE *initGISTstate(Relation index);
434 extern void freeGISTstate(GISTSTATE *giststate);
435 extern void gistdoinsert(Relation r,
438 GISTSTATE *GISTstate);
440 /* A List of these is returned from gistplacetopage() in *splitinfo */
443 Buffer buf; /* the split page "half" */
444 IndexTuple downlink; /* downlink for this half. */
447 extern bool gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
449 IndexTuple *itup, int ntup,
450 OffsetNumber oldoffnum, BlockNumber *newblkno,
455 extern SplitedPageLayout *gistSplit(Relation r, Page page, IndexTuple *itup,
456 int len, GISTSTATE *giststate);
459 extern void gist_redo(XLogReaderState *record);
460 extern void gist_desc(StringInfo buf, XLogReaderState *record);
461 extern const char *gist_identify(uint8 info);
462 extern void gist_xlog_startup(void);
463 extern void gist_xlog_cleanup(void);
465 extern XLogRecPtr gistXLogUpdate(RelFileNode node, Buffer buffer,
466 OffsetNumber *todelete, int ntodelete,
467 IndexTuple *itup, int ntup,
470 extern XLogRecPtr gistXLogSplit(RelFileNode node,
471 BlockNumber blkno, bool page_is_leaf,
472 SplitedPageLayout *dist,
473 BlockNumber origrlink, GistNSN oldnsn,
474 Buffer leftchild, bool markfollowright);
477 extern Datum gistgettuple(PG_FUNCTION_ARGS);
478 extern Datum gistgetbitmap(PG_FUNCTION_ARGS);
482 #define GiSTPageSize \
483 ( BLCKSZ - SizeOfPageHeaderData - MAXALIGN(sizeof(GISTPageOpaqueData)) )
485 #define GIST_MIN_FILLFACTOR 10
486 #define GIST_DEFAULT_FILLFACTOR 90
488 extern Datum gistoptions(PG_FUNCTION_ARGS);
489 extern bool gistfitpage(IndexTuple *itvec, int len);
490 extern bool gistnospace(Page page, IndexTuple *itvec, int len, OffsetNumber todelete, Size freespace);
491 extern void gistcheckpage(Relation rel, Buffer buf);
492 extern Buffer gistNewBuffer(Relation r);
493 extern void gistfillbuffer(Page page, IndexTuple *itup, int len,
495 extern IndexTuple *gistextractpage(Page page, int *len /* out */ );
496 extern IndexTuple *gistjoinvector(
497 IndexTuple *itvec, int *len,
498 IndexTuple *additvec, int addlen);
499 extern IndexTupleData *gistfillitupvec(IndexTuple *vec, int veclen, int *memlen);
501 extern IndexTuple gistunion(Relation r, IndexTuple *itvec,
502 int len, GISTSTATE *giststate);
503 extern IndexTuple gistgetadjusted(Relation r,
506 GISTSTATE *giststate);
507 extern IndexTuple gistFormTuple(GISTSTATE *giststate,
508 Relation r, Datum *attdata, bool *isnull, bool isleaf);
510 extern OffsetNumber gistchoose(Relation r, Page p,
512 GISTSTATE *giststate);
514 extern void GISTInitBuffer(Buffer b, uint32 f);
515 extern void gistdentryinit(GISTSTATE *giststate, int nkey, GISTENTRY *e,
516 Datum k, Relation r, Page pg, OffsetNumber o,
517 bool l, bool isNull);
519 extern float gistpenalty(GISTSTATE *giststate, int attno,
520 GISTENTRY *key1, bool isNull1,
521 GISTENTRY *key2, bool isNull2);
522 extern void gistMakeUnionItVec(GISTSTATE *giststate, IndexTuple *itvec, int len,
523 Datum *attr, bool *isnull);
524 extern bool gistKeyIsEQ(GISTSTATE *giststate, int attno, Datum a, Datum b);
525 extern void gistDeCompressAtt(GISTSTATE *giststate, Relation r, IndexTuple tuple, Page p,
526 OffsetNumber o, GISTENTRY *attdata, bool *isnull);
527 extern IndexTuple gistFetchTuple(GISTSTATE *giststate, Relation r,
529 extern void gistMakeUnionKey(GISTSTATE *giststate, int attno,
530 GISTENTRY *entry1, bool isnull1,
531 GISTENTRY *entry2, bool isnull2,
532 Datum *dst, bool *dstisnull);
534 extern XLogRecPtr gistGetFakeLSN(Relation rel);
537 extern Datum gistbulkdelete(PG_FUNCTION_ARGS);
538 extern Datum gistvacuumcleanup(PG_FUNCTION_ARGS);
541 extern void gistSplitByKey(Relation r, Page page, IndexTuple *itup,
542 int len, GISTSTATE *giststate,
547 extern Datum gistbuild(PG_FUNCTION_ARGS);
548 extern void gistValidateBufferingOption(char *value);
550 /* gistbuildbuffers.c */
551 extern GISTBuildBuffers *gistInitBuildBuffers(int pagesPerBuffer, int levelStep,
553 extern GISTNodeBuffer *gistGetNodeBuffer(GISTBuildBuffers *gfbb,
554 GISTSTATE *giststate,
555 BlockNumber blkno, int level);
556 extern void gistPushItupToNodeBuffer(GISTBuildBuffers *gfbb,
557 GISTNodeBuffer *nodeBuffer, IndexTuple item);
558 extern bool gistPopItupFromNodeBuffer(GISTBuildBuffers *gfbb,
559 GISTNodeBuffer *nodeBuffer, IndexTuple *item);
560 extern void gistFreeBuildBuffers(GISTBuildBuffers *gfbb);
561 extern void gistRelocateBuildBuffersOnSplit(GISTBuildBuffers *gfbb,
562 GISTSTATE *giststate, Relation r,
563 int level, Buffer buffer,
565 extern void gistUnloadNodeBuffers(GISTBuildBuffers *gfbb);
567 #endif /* GIST_PRIVATE_H */