1/*-------------------------------------------------------------------------
4 * header file for postgres btree access method implementation.
7 * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1994, Regents of the University of California
10 * src/include/access/nbtree.h
12 *-------------------------------------------------------------------------
20#include "catalog/pg_am_d.h"
29/* There's room for a 16-bit vacuum cycle ID in BTPageOpaqueData */
33 * BTPageOpaqueData -- At the end of every page, we store a pointer
34 * to both siblings in the tree. This is used to do forward/backward
35 * index scans. The next-page link is also critical for recovery when
36 * a search has navigated to the wrong page due to concurrent page splits
37 * or deletions; see src/backend/access/nbtree/README for more info.
39 * In addition, we store the page's btree level (counting upwards from
40 * zero at a leaf page) as well as some flag bits indicating the page type
41 * and status. If the page is deleted, a BTDeletedPageData struct is stored
42 * in the page's tuple area, while a standard BTPageOpaqueData struct is
43 * stored in the page special area.
45 * We also store a "vacuum cycle ID". When a page is split while VACUUM is
46 * processing the index, a nonzero value associated with the VACUUM run is
47 * stored into both halves of the split page. (If VACUUM is not running,
48 * both pages receive zero cycleids.) This allows VACUUM to detect whether
49 * a page was split since it started, with a small probability of false match
50 * if the page was last split some exact multiple of MAX_BT_CYCLE_ID VACUUMs
51 * ago. Also, during a split, the BTP_SPLIT_END flag is cleared in the left
52 * (original) page, and set in the right page, but only if the next page
53 * to its right has a different cycleid.
55 * NOTE: the BTP_LEAF flag bit is redundant since level==0 could be tested
58 * NOTE: the btpo_level field used to be a union type in order to allow
59 * deleted pages to store a 32-bit safexid in the same field. We now store
60 * 64-bit/full safexid values using BTDeletedPageData instead.
74 #define BTPageGetOpaque(page) ((BTPageOpaque) PageGetSpecialPointer(page))
76/* Bits defined in btpo_flags */
77 #define BTP_LEAF (1 << 0) /* leaf page, i.e. not internal page */
78 #define BTP_ROOT (1 << 1) /* root page (has no parent) */
79 #define BTP_DELETED (1 << 2) /* page has been deleted from tree */
80 #define BTP_META (1 << 3) /* meta-page */
81 #define BTP_HALF_DEAD (1 << 4) /* empty, but still in tree */
82 #define BTP_SPLIT_END (1 << 5) /* rightmost page of split group */
83 #define BTP_HAS_GARBAGE (1 << 6) /* page has LP_DEAD tuples (deprecated) */
84 #define BTP_INCOMPLETE_SPLIT (1 << 7) /* right sibling's downlink is missing */
85 #define BTP_HAS_FULLXID (1 << 8) /* contains BTDeletedPageData */
88 * The max allowed value of a cycle ID is a bit less than 64K. This is
89 * for convenience of pg_filedump and similar utilities: we want to use
90 * the last 2 bytes of special space as an index type indicator, and
91 * restricting cycle ID lets btree use that space for vacuum cycle IDs
92 * while still allowing index type to be identified.
94 #define MAX_BT_CYCLE_ID 0xFF7F
98 * The Meta page is always the first page in the btree index.
99 * Its primary purpose is to point to the location of the btree root page.
100 * We also point to the "fast" root, which is the current effective root;
101 * see README for discussion.
112 /* remaining fields only valid when btm_version >= BTREE_NOVAC_VERSION */
114 /* number of deleted, non-recyclable pages during last cleanup */
116 /* number of heap tuples during last cleanup (deprecated) */
122 #define BTPageGetMeta(p) \
123 ((BTMetaPageData *) PageGetContents(p))
126 * The current Btree version is 4. That's what you'll get when you create
129 * Btree version 3 was used in PostgreSQL v11. It is mostly the same as
130 * version 4, but heap TIDs were not part of the keyspace. Index tuples
131 * with duplicate keys could be stored in any order. We continue to
132 * support reading and writing Btree versions 2 and 3, so that they don't
133 * need to be immediately re-indexed at pg_upgrade. In order to get the
134 * new heapkeyspace semantics, however, a REINDEX is needed.
136 * Deduplication is safe to use when the btm_allequalimage field is set to
137 * true. It's safe to read the btm_allequalimage field on version 3, but
138 * only version 4 indexes make use of deduplication. Even version 4
139 * indexes created on PostgreSQL v12 will need a REINDEX to make use of
140 * deduplication, though, since there is no other way to set
141 * btm_allequalimage to true (pg_upgrade hasn't been taught to set the
144 * Btree version 2 is mostly the same as version 3. There are two new
145 * fields in the metapage that were introduced in version 3. A version 2
146 * metapage will be automatically upgraded to version 3 on the first
147 * insert to it. INCLUDE indexes cannot use version 2.
149 #define BTREE_METAPAGE 0 /* first page is meta */
150 #define BTREE_MAGIC 0x053162 /* magic number in metapage */
151 #define BTREE_VERSION 4 /* current version number */
152 #define BTREE_MIN_VERSION 2 /* minimum supported version */
153 #define BTREE_NOVAC_VERSION 3 /* version with all meta fields set */
156 * Maximum size of a btree index entry, including its tuple header.
158 * We actually need to be able to fit three items on every page,
159 * so restrict any one item to 1/3 the per-page available space.
161 * There are rare cases where _bt_truncate() will need to enlarge
162 * a heap index tuple to make space for a tiebreaker heap TID
163 * attribute, which we account for here.
165 #define BTMaxItemSize \
166 (MAXALIGN_DOWN((BLCKSZ - \
167 MAXALIGN(SizeOfPageHeaderData + 3*sizeof(ItemIdData)) - \
168 MAXALIGN(sizeof(BTPageOpaqueData))) / 3) - \
169 MAXALIGN(sizeof(ItemPointerData)))
170 #define BTMaxItemSizeNoHeapTid \
171 MAXALIGN_DOWN((BLCKSZ - \
172 MAXALIGN(SizeOfPageHeaderData + 3*sizeof(ItemIdData)) - \
173 MAXALIGN(sizeof(BTPageOpaqueData))) / 3)
176 * MaxTIDsPerBTreePage is an upper bound on the number of heap TIDs tuples
177 * that may be stored on a btree leaf page. It is used to size the
178 * per-page temporary buffers.
180 * Note: we don't bother considering per-tuple overheads here to keep
181 * things simple (value is based on how many elements a single array of
182 * heap TIDs must have to fill the space between the page header and
183 * special area). The value is slightly higher (i.e. more conservative)
184 * than necessary as a result, which is considered acceptable.
186 #define MaxTIDsPerBTreePage \
187 (int) ((BLCKSZ - SizeOfPageHeaderData - sizeof(BTPageOpaqueData)) / \
188 sizeof(ItemPointerData))
191 * The leaf-page fillfactor defaults to 90% but is user-adjustable.
192 * For pages above the leaf level, we use a fixed 70% fillfactor.
193 * The fillfactor is applied during index build and when splitting
194 * a rightmost page; when splitting non-rightmost pages we try to
195 * divide the data equally. When splitting a page that's entirely
196 * filled with a single value (duplicates), the effective leaf-page
197 * fillfactor is 96%, regardless of whether the page is a rightmost
200 #define BTREE_MIN_FILLFACTOR 10
201 #define BTREE_DEFAULT_FILLFACTOR 90
202 #define BTREE_NONLEAF_FILLFACTOR 70
203 #define BTREE_SINGLEVAL_FILLFACTOR 96
206 * In general, the btree code tries to localize its knowledge about
207 * page layout to a couple of routines. However, we need a special
208 * value to indicate "no page number" in those places where we expect
209 * page numbers. We can use zero for this because we never need to
210 * make a pointer to the metadata page.
216 * Macros to test whether a page is leftmost or rightmost on its tree level,
217 * as well as other state info kept in the opaque data.
219 #define P_LEFTMOST(opaque) ((opaque)->btpo_prev == P_NONE)
220 #define P_RIGHTMOST(opaque) ((opaque)->btpo_next == P_NONE)
221 #define P_ISLEAF(opaque) (((opaque)->btpo_flags & BTP_LEAF) != 0)
222 #define P_ISROOT(opaque) (((opaque)->btpo_flags & BTP_ROOT) != 0)
223 #define P_ISDELETED(opaque) (((opaque)->btpo_flags & BTP_DELETED) != 0)
224 #define P_ISMETA(opaque) (((opaque)->btpo_flags & BTP_META) != 0)
225 #define P_ISHALFDEAD(opaque) (((opaque)->btpo_flags & BTP_HALF_DEAD) != 0)
226 #define P_IGNORE(opaque) (((opaque)->btpo_flags & (BTP_DELETED|BTP_HALF_DEAD)) != 0)
227 #define P_HAS_GARBAGE(opaque) (((opaque)->btpo_flags & BTP_HAS_GARBAGE) != 0)
228 #define P_INCOMPLETE_SPLIT(opaque) (((opaque)->btpo_flags & BTP_INCOMPLETE_SPLIT) != 0)
229 #define P_HAS_FULLXID(opaque) (((opaque)->btpo_flags & BTP_HAS_FULLXID) != 0)
232 * BTDeletedPageData is the page contents of a deleted page
255 /* Set safexid in deleted page */
266 /* We only expect to be called with a deleted page */
271 /* pg_upgrade'd deleted page -- must be safe to recycle now */
275 /* Get safexid from deleted page */
281 * Is an existing page recyclable?
283 * This exists to centralize the policy on which deleted pages are now safe to
284 * re-use. However, _bt_pendingfsm_finalize() duplicates some of the same
285 * logic because it doesn't work directly with pages -- keep the two in sync.
287 * Note: PageIsNew() pages are always safe to recycle, but we can't deal with
288 * them here (caller is responsible for that case themselves). Caller might
289 * well need special handling for new pages anyway.
299 /* Recycling okay iff page is deleted and safexid is old enough */
306 * The page was deleted, but when? If it was just deleted, a scan
307 * might have seen the downlink to it, and will read the page later.
308 * As long as that can happen, we must keep the deleted page around as
311 * For that check if the deletion XID could still be visible to
312 * anyone. If not, then no scan that's still in progress could have
313 * seen its downlink, and we can recycle it.
322 * BTVacState and BTPendingFSM are private nbtree.c state used during VACUUM.
323 * They are exported for use by page deletion related code in nbtpage.c.
341 * _bt_pendingfsm_finalize() state
343 int bufsize;
/* pendingpages space (in # elements) */
350 * Lehman and Yao's algorithm requires a ``high key'' on every non-rightmost
351 * page. The high key is not a tuple that is used to visit the heap. It is
352 * a pivot tuple (see "Notes on B-Tree tuple format" below for definition).
353 * The high key on a page is required to be greater than or equal to any
354 * other key that appears on the page. If we find ourselves trying to
355 * insert a key that is strictly > high key, we know we need to move right
356 * (this should only happen if the page was split since we examined the
359 * Our insertion algorithm guarantees that we can use the initial least key
360 * on our right sibling as the high key. Once a page is created, its high
361 * key changes only if the page is split.
363 * On a non-rightmost page, the high key lives in item 1 and data items
364 * start in item 2. Rightmost pages have no high key, so we store data
365 * items beginning in item 1.
368 #define P_HIKEY ((OffsetNumber) 1)
369 #define P_FIRSTKEY ((OffsetNumber) 2)
370 #define P_FIRSTDATAKEY(opaque) (P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY)
373 * Notes on B-Tree tuple format, and key and non-key attributes:
375 * INCLUDE B-Tree indexes have non-key attributes. These are extra
376 * attributes that may be returned by index-only scans, but do not influence
377 * the order of items in the index (formally, non-key attributes are not
378 * considered to be part of the key space). Non-key attributes are only
379 * present in leaf index tuples whose item pointers actually point to heap
380 * tuples (non-pivot tuples). _bt_check_natts() enforces the rules
383 * Non-pivot tuple format (plain/non-posting variant):
385 * t_tid | t_info | key values | INCLUDE columns, if any
387 * t_tid points to the heap TID, which is a tiebreaker key column as of
390 * Non-pivot tuples complement pivot tuples, which only have key columns.
391 * The sole purpose of pivot tuples is to represent how the key space is
392 * separated. In general, any B-Tree index that has more than one level
393 * (i.e. any index that does not just consist of a metapage and a single
394 * leaf root page) must have some number of pivot tuples, since pivot
395 * tuples are used for traversing the tree. Suffix truncation can omit
396 * trailing key columns when a new pivot is formed, which makes minus
397 * infinity their logical value. Since BTREE_VERSION 4 indexes treat heap
398 * TID as a trailing key column that ensures that all index tuples are
399 * physically unique, it is necessary to represent heap TID as a trailing
400 * key column in pivot tuples, though very often this can be truncated
401 * away, just like any other key column. (Actually, the heap TID is
402 * omitted rather than truncated, since its representation is different to
403 * the non-pivot representation.)
405 * Pivot tuple format:
407 * t_tid | t_info | key values | [heap TID]
409 * We store the number of columns present inside pivot tuples by abusing
410 * their t_tid offset field, since pivot tuples never need to store a real
411 * offset (pivot tuples generally store a downlink in t_tid, though). The
412 * offset field only stores the number of columns/attributes when the
413 * INDEX_ALT_TID_MASK bit is set, which doesn't count the trailing heap
414 * TID column sometimes stored in pivot tuples -- that's represented by
415 * the presence of BT_PIVOT_HEAP_TID_ATTR. The INDEX_ALT_TID_MASK bit in
416 * t_info is always set on BTREE_VERSION 4 pivot tuples, since
417 * BTreeTupleIsPivot() must work reliably on heapkeyspace versions.
419 * In version 2 or version 3 (!heapkeyspace) indexes, INDEX_ALT_TID_MASK
420 * might not be set in pivot tuples. BTreeTupleIsPivot() won't work
421 * reliably as a result. The number of columns stored is implicitly the
422 * same as the number of columns in the index, just like any non-pivot
423 * tuple. (The number of columns stored should not vary, since suffix
424 * truncation of key columns is unsafe within any !heapkeyspace index.)
426 * The 12 least significant bits from t_tid's offset number are used to
427 * represent the number of key columns within a pivot tuple. This leaves 4
428 * status bits (BT_STATUS_OFFSET_MASK bits), which are shared by all tuples
429 * that have the INDEX_ALT_TID_MASK bit set (set in t_info) to store basic
430 * tuple metadata. BTreeTupleIsPivot() and BTreeTupleIsPosting() use the
431 * BT_STATUS_OFFSET_MASK bits.
433 * Sometimes non-pivot tuples also use a representation that repurposes
434 * t_tid to store metadata rather than a TID. PostgreSQL v13 introduced a
435 * new non-pivot tuple format to support deduplication: posting list
436 * tuples. Deduplication merges together multiple equal non-pivot tuples
437 * into a logically equivalent, space efficient representation. A posting
438 * list is an array of ItemPointerData elements. Non-pivot tuples are
439 * merged together to form posting list tuples lazily, at the point where
440 * we'd otherwise have to split a leaf page.
442 * Posting tuple format (alternative non-pivot tuple representation):
444 * t_tid | t_info | key values | posting list (TID array)
446 * Posting list tuples are recognized as such by having the
447 * INDEX_ALT_TID_MASK status bit set in t_info and the BT_IS_POSTING status
448 * bit set in t_tid's offset number. These flags redefine the content of
449 * the posting tuple's t_tid to store the location of the posting list
450 * (instead of a block number), as well as the total number of heap TIDs
451 * present in the tuple (instead of a real offset number).
453 * The 12 least significant bits from t_tid's offset number are used to
454 * represent the number of heap TIDs present in the tuple, leaving 4 status
455 * bits (the BT_STATUS_OFFSET_MASK bits). Like any non-pivot tuple, the
456 * number of columns stored is always implicitly the total number in the
457 * index (in practice there can never be non-key columns stored, since
458 * deduplication is not supported with INCLUDE indexes).
460 #define INDEX_ALT_TID_MASK INDEX_AM_RESERVED_BIT
462/* Item pointer offset bit masks */
463 #define BT_OFFSET_MASK 0x0FFF
464 #define BT_STATUS_OFFSET_MASK 0xF000
465/* BT_STATUS_OFFSET_MASK status bits */
466 #define BT_PIVOT_HEAP_TID_ATTR 0x1000
467 #define BT_IS_POSTING 0x2000
470 * Mask allocated for number of keys in index tuple must be able to fit
471 * maximum possible number of index attributes
474 "BT_OFFSET_MASK can't fit INDEX_MAX_KEYS");
477 * Note: BTreeTupleIsPivot() can have false negatives (but not false
478 * positives) when used with !heapkeyspace indexes
485 /* absence of BT_IS_POSTING in offset number indicates pivot tuple */
497 /* presence of BT_IS_POSTING in offset number indicates posting tuple */
551 * Get/set downlink block number in pivot tuple.
553 * Note: Cannot assert that tuple is a pivot tuple. If we did so then
554 * !heapkeyspace indexes would exhibit false positive assertion failures.
569 * Get number of attributes within tuple.
571 * Note that this does not include an implicit tiebreaker heap TID
572 * attribute, if any. Note also that the number of key attributes must be
573 * explicitly represented in all heapkeyspace pivot tuples.
575 * Note: This is defined as a macro rather than an inline function to
576 * avoid including rel.h.
578 #define BTreeTupleGetNAtts(itup, rel) \
580 (BTreeTupleIsPivot(itup)) ? \
582 ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) & BT_OFFSET_MASK \
585 IndexRelationGetNumberOfAttributes(rel) \
589 * Set number of key attributes in tuple.
591 * The heap TID tiebreaker attribute bit may also be set here, indicating that
592 * a heap TID value will be stored at the end of the tuple (i.e. using the
593 * special pivot tuple representation).
600 Assert(!heaptid || nkeyatts > 0);
608 /* BT_IS_POSTING bit is deliberately unset here */
614 * Get/set leaf page's "top parent" link from its high key. Used during page
617 * Note: Cannot assert that tuple is a pivot tuple. If we did so then
618 * !heapkeyspace indexes would exhibit false positive assertion failures.
634 * Get tiebreaker heap TID attribute, if any.
636 * This returns the first/lowest heap TID in the case of a posting list tuple.
643 /* Pivot tuple heap TID representation? */
649 /* Heap TID attribute was truncated */
659 * Get maximum heap TID attribute, which could be the only TID in the case of
660 * a non-pivot tuple that does not have a posting list.
662 * Works with non-pivot tuples only.
680 * Operator strategy numbers for B-tree have been moved to access/stratnum.h,
681 * because many places need to use them in ScanKeyInit() calls.
683 * The strategy numbers are chosen so that we can commute them by
686 #define BTCommuteStrategyNumber(strat) (BTMaxStrategyNumber + 1 - (strat))
689 * When a new operator class is declared, we require that the user
690 * supply us with an amproc procedure (BTORDER_PROC) for determining
691 * whether, for two keys a and b, a < b, a = b, or a > b. This routine
692 * must return < 0, 0, > 0, respectively, in these three cases.
694 * To facilitate accelerated sorting, an operator class may choose to
695 * offer a second procedure (BTSORTSUPPORT_PROC). For full details, see
696 * src/include/utils/sortsupport.h.
698 * To support window frames defined by "RANGE offset PRECEDING/FOLLOWING",
699 * an operator class may choose to offer a third amproc procedure
700 * (BTINRANGE_PROC), independently of whether it offers sortsupport.
701 * For full details, see doc/src/sgml/btree.sgml.
703 * To facilitate B-Tree deduplication, an operator class may choose to
704 * offer a forth amproc procedure (BTEQUALIMAGE_PROC). For full details,
705 * see doc/src/sgml/btree.sgml.
707 * An operator class may choose to offer a fifth amproc procedure
708 * (BTOPTIONS_PROC). These procedures define a set of user-visible
709 * parameters that can be used to control operator class behavior. None of
710 * the built-in B-Tree operator classes currently register an "options" proc.
712 * To facilitate more efficient B-Tree skip scans, an operator class may
713 * choose to offer a sixth amproc procedure (BTSKIPSUPPORT_PROC). For full
714 * details, see src/include/utils/skipsupport.h.
717 #define BTORDER_PROC 1
718 #define BTSORTSUPPORT_PROC 2
719 #define BTINRANGE_PROC 3
720 #define BTEQUALIMAGE_PROC 4
721 #define BTOPTIONS_PROC 5
722 #define BTSKIPSUPPORT_PROC 6
726 * We need to be able to tell the difference between read and write
727 * requests for pages, in order to do locking correctly.
730 #define BT_READ BUFFER_LOCK_SHARE
731 #define BT_WRITE BUFFER_LOCK_EXCLUSIVE
734 * BTStackData -- As we descend a tree, we push the location of pivot
735 * tuples whose downlink we are about to follow onto a private stack. If
736 * we split a leaf, we use this stack to walk back up the tree and insert
737 * data into its parent page at the correct location. We also have to
738 * recursively insert into the grandparent page if and when the parent page
739 * splits. Our private stack can become stale due to concurrent page
740 * splits and page deletions, but it should never give us an irredeemably
753 * BTScanInsertData is the btree-private state needed to find an initial
754 * position for an indexscan, or to insert new tuples -- an "insertion
755 * scankey" (not to be confused with a search scankey). It's used to descend
756 * a B-Tree using _bt_search.
758 * heapkeyspace indicates if we expect all keys in the index to be physically
759 * unique because heap TID is used as a tiebreaker attribute, and if index may
760 * have truncated key attributes in pivot tuples. This is actually a property
761 * of the index relation itself (not an indexscan). heapkeyspace indexes are
762 * indexes whose version is >= version 4. It's convenient to keep this close
763 * by, rather than accessing the metapage repeatedly.
765 * allequalimage is set to indicate that deduplication is safe for the index.
766 * This is also a property of the index relation rather than an indexscan.
768 * anynullkeys indicates if any of the keys had NULL value when scankey was
769 * built from index tuple (note that already-truncated tuple key attributes
770 * set NULL as a placeholder key value, which also affects value of
771 * anynullkeys). This is a convenience for unique index non-pivot tuple
772 * insertion, which usually temporarily unsets scantid, but shouldn't iff
773 * anynullkeys is true. Value generally matches non-pivot tuple's HasNulls
774 * bit, but may not when inserting into an INCLUDE index (tuple header value
775 * is affected by the NULL-ness of both key and non-key attributes).
777 * See comments in _bt_first for an explanation of the nextkey and backward
780 * scantid is the heap TID that is used as a final tiebreaker attribute. It
781 * is set to NULL when index scan doesn't need to find a position for a
782 * specific physical tuple. Must be set when inserting new tuples into
783 * heapkeyspace indexes, since every tuple in the tree unambiguously belongs
784 * in one exact position (it's never set with !heapkeyspace indexes, though).
785 * Despite the representational difference, nbtree search code considers
786 * scantid to be just another insertion scankey attribute.
788 * scankeys is an array of scan key entries for attributes that are compared
789 * before scantid (user-visible attributes). keysz is the size of the array.
790 * During insertion, there must be a scan key for every attribute, but when
791 * starting a regular index scan some can be omitted. The array is used as a
792 * flexible array member, though it's sized in a way that makes it possible to
793 * use stack allocations. See nbtree/README for full details.
803 int keysz;
/* Size of scankeys array */
810 * BTInsertStateData is a working area used during insertion.
812 * This is filled in after descending the tree to the first leaf page the new
813 * tuple might belong on. Tracks the current position while performing
814 * uniqueness check, before we have determined which exact page to insert
817 * (This should be private to nbtinsert.c, but it's also used by
818 * _bt_binsrch_insert)
826 /* Buffer containing leaf page we're likely to insert itup on */
830 * Cache of bounds within the current buffer. Only used for insertions
831 * where _bt_check_unique is called. See _bt_binsrch_insert and
832 * _bt_findinsertloc for details.
839 * if _bt_binsrch_insert found the location inside existing posting list,
840 * save the position inside the list. -1 sentinel value indicates overlap
841 * with an existing posting list tuple that has its LP_DEAD bit set.
849 * State used to representing an individual pending tuple during
859 * BTDedupStateData is a working area used during deduplication.
861 * The status info fields track the state of a whole-page deduplication pass.
862 * State about the current pending posting list is also tracked.
864 * A pending posting list is comprised of a contiguous group of equal items
865 * from the page, starting from page offset number 'baseoff'. This is the
866 * offset number of the "base" tuple for new posting list. 'nitems' is the
867 * current total number of existing items from the page that will be merged to
868 * make a new posting list tuple, including the base tuple item. (Existing
869 * items may themselves be posting list tuples, or regular non-pivot tuples.)
871 * The total size of the existing tuples to be freed when pending posting list
872 * is processed gets tracked by 'phystupsize'. This information allows
873 * deduplication to calculate the space saving for each new posting list
874 * tuple, and for the entire pass over the page as a whole.
878 /* Deduplication status info for entire pass over page */
883 /* Metadata about base tuple of current pending posting list */
888 /* Other metadata about pending posting list */
890 int nhtids;
/* Number of heap TIDs in htids array */
891 int nitems;
/* Number of existing tuples/line pointers */
895 * Array of tuples to go on new version of the page. Contains one entry
896 * for each group of consecutive items. Note that existing tuples that
897 * will not become posting list tuples do not appear in the array (they
898 * are implicitly unchanged by deduplication pass).
907 * BTVacuumPostingData is state that represents how to VACUUM (or delete) a
908 * posting list tuple when some (though not all) of its TIDs are to be
911 * Convention is that itup field is the original posting list tuple on input,
912 * and palloc()'d final tuple used to overwrite existing tuple on output.
916 /* Tuple that will be/was updated */
920 /* State needed to describe final itup in WAL */
928 * BTScanOpaqueData is the btree-private state needed for an indexscan.
929 * This consists of preprocessed scan keys (see _bt_preprocess_keys() for
930 * details of the preprocessing), information about the current location
931 * of the scan, and information about the marked location, if any. (We use
932 * BTScanPosData to represent the data needed for each of current and marked
933 * locations.) In addition we can remember some known-killed index entries
934 * that must be marked before we can move off the current page.
936 * Index scans work a page at a time: we pin and read-lock the page, identify
937 * all the matching items on the page and save them in BTScanPosData, then
938 * release the read-lock while returning the items to the caller for
939 * processing. This approach minimizes lock/unlock traffic. We must always
940 * drop the lock to make it okay for caller to process the returned items.
941 * Whether or not we can also release the pin during this window will vary.
942 * We drop the pin (when so->dropPin) to avoid blocking progress by VACUUM
943 * (see nbtree/README section about making concurrent TID recycling safe).
944 * We'll always release both the lock and the pin on the current page before
945 * moving on to its sibling page.
947 * If we are doing an index-only scan, we save the entire IndexTuple for each
948 * matched item, otherwise only its heap TID and offset. The IndexTuples go
949 * into a separate workspace array; each BTScanPosItem stores its tuple's
950 * offset within that array. Posting list tuples store a "base" tuple once,
951 * allowing the same key to be returned for each TID in the posting list
966 /* page details as of the saved position's call to _bt_readpage */
972 /* scan direction for the saved position's call to _bt_readpage */
976 * If we are doing an index-only scan, nextTupleOffset is the first free
977 * location in the associated tuple storage workspace.
982 * moreLeft and moreRight track whether we think there may be matching
983 * index entries to the left and right of the current page, respectively.
989 * The items array is always ordered in index order (ie, increasing
990 * indexoffset). When scanning backwards it is convenient to fill the
991 * array back-to-front, so we start at the last slot and fill downwards.
992 * Hence we need both a first-valid-entry and a last-valid-entry counter.
993 * itemIndex is a cursor showing which entry was last returned to caller.
1004 #define BTScanPosIsPinned(scanpos) \
1006 AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
1007 !BufferIsValid((scanpos).buf)), \
1008 BufferIsValid((scanpos).buf) \
1010 #define BTScanPosUnpin(scanpos) \
1012 ReleaseBuffer((scanpos).buf); \
1013 (scanpos).buf = InvalidBuffer; \
1015 #define BTScanPosUnpinIfPinned(scanpos) \
1017 if (BTScanPosIsPinned(scanpos)) \
1018 BTScanPosUnpin(scanpos); \
1021 #define BTScanPosIsValid(scanpos) \
1023 AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
1024 !BufferIsValid((scanpos).buf)), \
1025 BlockNumberIsValid((scanpos).currPage) \
1027 #define BTScanPosInvalidate(scanpos) \
1029 (scanpos).buf = InvalidBuffer; \
1030 (scanpos).currPage = InvalidBlockNumber; \
1033/* We need one of these for each equality-type SK_SEARCHARRAY scan key */
1036 /* fields set for both kinds of array (SAOP arrays and skip arrays) */
1040 /* fields set for ScalarArrayOpExpr arrays only */
1042 int cur_elem;
/* index of current element in elem_values */
1044 /* fields set for skip arrays only */
1046 bool attbyval;
/* attr's FormData_pg_attribute.attbyval */
1055 /* these fields are set by _bt_preprocess_keys(): */
1056 bool qual_ok;
/* false if qual can never be satisfied */
1060 /* workspace for SK_SEARCHARRAY support */
1062 bool skipScan;
/* At least one skip array in arrayKeys[]? */
1070 /* info about killed items if any (killedItems is NULL if never used) */
1073 bool dropPin;
/* drop leaf pin before btgettuple returns? */
1076 * If we are doing an index-only scan, these are the tuple storage
1077 * workspaces for the currPos and markPos respectively. Each is of size
1078 * BLCKSZ, so it can hold as much as a full page's worth of tuples.
1084 * If the marked position is on the same page as current position, we
1085 * don't use markPos, but just keep the marked itemIndex in markItemIndex
1086 * (all the rest of currPos is valid for the mark position). Hence, to
1087 * determine if there is a mark, first look at markItemIndex, then at
1092 /* keep these last in struct for efficiency */
1100 * _bt_readpage state used across _bt_checkkeys calls for a page
1104 /* Input parameters, set by _bt_readpage for _bt_checkkeys */
1113 /* Per-tuple input parameters, set by _bt_readpage for _bt_checkkeys */
1116 /* Output parameters, set by _bt_checkkeys for _bt_readpage */
1121 * Private _bt_checkkeys state used to manage "look ahead" optimization
1122 * and primscan scheduling (only used during scans with array keys)
1131 * We use some private sk_flags bits in preprocessed scan keys. We're allowed
1132 * to use bits 16-31 (see skey.h). The uppermost bits are copied from the
1133 * index's indoption[] array entry for the index attribute.
1135 #define SK_BT_REQFWD 0x00010000 /* required to continue forward scan */
1136 #define SK_BT_REQBKWD 0x00020000 /* required to continue backward scan */
1137 #define SK_BT_SKIP 0x00040000 /* skip array on column without input = */
1139/* SK_BT_SKIP-only flags (set and unset by array advancement) */
1140 #define SK_BT_MINVAL 0x00080000 /* invalid sk_argument, use low_compare */
1141 #define SK_BT_MAXVAL 0x00100000 /* invalid sk_argument, use high_compare */
1142 #define SK_BT_NEXT 0x00200000 /* positions the scan > sk_argument */
1143 #define SK_BT_PRIOR 0x00400000 /* positions the scan < sk_argument */
1145/* Remaps pg_index flag bits to uppermost SK_BT_* byte */
1146 #define SK_BT_INDOPTION_SHIFT 24 /* must clear the above bits */
1147 #define SK_BT_DESC (INDOPTION_DESC << SK_BT_INDOPTION_SHIFT)
1148 #define SK_BT_NULLS_FIRST (INDOPTION_NULLS_FIRST << SK_BT_INDOPTION_SHIFT)
1158 #define BTGetFillFactor(relation) \
1159 (AssertMacro(relation->rd_rel->relkind == RELKIND_INDEX && \
1160 relation->rd_rel->relam == BTREE_AM_OID), \
1161 (relation)->rd_options ? \
1162 ((BTOptions *) (relation)->rd_options)->fillfactor : \
1163 BTREE_DEFAULT_FILLFACTOR)
1164 #define BTGetTargetPageFreeSpace(relation) \
1165 (BLCKSZ * (100 - BTGetFillFactor(relation)) / 100)
1166 #define BTGetDeduplicateItems(relation) \
1167 (AssertMacro(relation->rd_rel->relkind == RELKIND_INDEX && \
1168 relation->rd_rel->relam == BTREE_AM_OID), \
1169 ((relation)->rd_options ? \
1170 ((BTOptions *) (relation)->rd_options)->deduplicate_items : true))
1173 * Constant definition for progress reporting. Phase numbers must match
1176/* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 (see progress.h) */
1177 #define PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN 2
1178 #define PROGRESS_BTREE_PHASE_PERFORMSORT_1 3
1179 #define PROGRESS_BTREE_PHASE_PERFORMSORT_2 4
1180 #define PROGRESS_BTREE_PHASE_LEAF_LOAD 5
1183 * external entry points for btree, in nbtree.c
1189 bool indexUnchanged,
1197 ScanKey orderbys,
int norderbys);
1205 void *callback_state);
1215 * prototypes for internal functions in nbtree.c
1227 * prototypes for functions in nbtdedup.c
1230 Size newitemsz,
bool bottomupdedup);
1244 * prototypes for functions in nbtinsert.c
1255 * prototypes for functions in nbtsplitloc.c
1259 bool *newitemonleft);
1262 * prototypes for functions in nbtpage.c
1265 bool allequalimage);
1273 bool *allequalimage);
1298 * prototypes for functions in nbtpreprocesskeys.c
1303 * prototypes for functions in nbtsearch.c
1314 * prototypes for functions in nbtutils.c
1321 Datum tupdatum,
bool tupnull,
1323 int32 *set_elem_result);
1340 bool *res,
bool *isnull);
1353 * prototypes for functions in nbtvalidate.c
1362 * prototypes for functions in nbtsort.c
1368#endif /* NBTREE_H */
static bool validate(Port *port, const char *auth)
static Datum values[MAXATTR]
PageHeaderData * PageHeader
static bool PageIsNew(const PageData *page)
#define SizeOfPageHeaderData
static char * PageGetContents(Page page)
#define FLEXIBLE_ARRAY_MEMBER
bool(* IndexBulkDeleteCallback)(ItemPointer itemptr, void *state)
Assert(PointerIsAligned(start, uint64))
static void ItemPointerSetOffsetNumber(ItemPointerData *pointer, OffsetNumber offsetNumber)
static void ItemPointerSetBlockNumber(ItemPointerData *pointer, BlockNumber blockNumber)
static OffsetNumber ItemPointerGetOffsetNumberNoCheck(const ItemPointerData *pointer)
static BlockNumber ItemPointerGetBlockNumberNoCheck(const ItemPointerData *pointer)
struct ItemPointerData ItemPointerData
static Size IndexTupleSize(const IndexTupleData *itup)
#define MaxIndexTuplesPerPage
#define P_HAS_FULLXID(opaque)
bool btinsert(Relation rel, Datum *values, bool *isnull, ItemPointer ht_ctid, Relation heapRel, IndexUniqueCheck checkUnique, bool indexUnchanged, struct IndexInfo *indexInfo)
Buffer _bt_get_endpoint(Relation rel, uint32 level, bool rightmost)
Buffer _bt_relandgetbuf(Relation rel, Buffer obuf, BlockNumber blkno, int access)
void _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace, Page page, IndexTuple newtup)
void _bt_parallel_primscan_schedule(IndexScanDesc scan, BlockNumber curr_page)
bool btcanreturn(Relation index, int attno)
bool _bt_scanbehind_checkkeys(IndexScanDesc scan, ScanDirection dir, IndexTuple finaltup)
BTPageOpaqueData * BTPageOpaque
#define BT_PIVOT_HEAP_TID_ATTR
static uint16 BTreeTupleGetNPosting(IndexTuple posting)
void _bt_upgrademetapage(Page page)
void _bt_relbuf(Relation rel, Buffer buf)
IndexTuple _bt_swap_posting(IndexTuple newitem, IndexTuple oposting, int postingoff)
static bool BTreeTupleIsPivot(IndexTuple itup)
Buffer _bt_gettrueroot(Relation rel)
int _bt_getrootheight(Relation rel)
void _bt_end_vacuum(Relation rel)
bool btvalidate(Oid opclassoid)
void _bt_pageinit(Page page, Size size)
bool _bt_first(IndexScanDesc scan, ScanDirection dir)
void _bt_dedup_pass(Relation rel, Buffer buf, IndexTuple newitem, Size newitemsz, bool bottomupdedup)
OffsetNumber _bt_findsplitloc(Relation rel, Page origpage, OffsetNumber newitemoff, Size newitemsz, IndexTuple newitem, bool *newitemonleft)
static FullTransactionId BTPageGetDeleteXid(Page page)
bool _bt_parallel_seize(IndexScanDesc scan, BlockNumber *next_scan_page, BlockNumber *last_curr_page, bool first)
void _bt_update_posting(BTVacuumPosting vacposting)
bool _bt_bottomupdel_pass(Relation rel, Buffer buf, Relation heapRel, Size newitemsz)
bool _bt_checkkeys(IndexScanDesc scan, BTReadPageState *pstate, bool arrayKeys, IndexTuple tuple, int tupnatts)
bool _bt_dedup_save_htid(BTDedupState state, IndexTuple itup)
void _bt_end_vacuum_callback(int code, Datum arg)
void _bt_pagedel(Relation rel, Buffer leafbuf, BTVacState *vstate)
Buffer _bt_allocbuf(Relation rel, Relation heaprel)
struct BTPageOpaqueData BTPageOpaqueData
StrategyNumber bttranslatecmptype(CompareType cmptype, Oid opfamily)
IndexScanDesc btbeginscan(Relation rel, int nkeys, int norderbys)
void _bt_delitems_vacuum(Relation rel, Buffer buf, OffsetNumber *deletable, int ndeletable, BTVacuumPosting *updatable, int nupdatable)
struct BTReadPageState BTReadPageState
int _bt_binsrch_array_skey(FmgrInfo *orderproc, bool cur_elem_trig, ScanDirection dir, Datum tupdatum, bool tupnull, BTArrayKeyInfo *array, ScanKey cur, int32 *set_elem_result)
static void BTreeTupleSetTopParent(IndexTuple leafhikey, BlockNumber blkno)
void _bt_freestack(BTStack stack)
static void BTreeTupleSetPosting(IndexTuple itup, uint16 nhtids, int postingoffset)
Buffer _bt_getstackbuf(Relation rel, Relation heaprel, BTStack stack, BlockNumber child)
Size btestimateparallelscan(Relation rel, int nkeys, int norderbys)
void BTreeShmemInit(void)
void _bt_parallel_done(IndexScanDesc scan)
void _bt_dedup_start_pending(BTDedupState state, IndexTuple base, OffsetNumber baseoff)
#define BTPageGetOpaque(page)
#define P_ISDELETED(opaque)
static ItemPointer BTreeTupleGetPosting(IndexTuple posting)
void _bt_checkpage(Relation rel, Buffer buf)
IndexBulkDeleteResult * btvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
BTCycleId _bt_vacuum_cycleid(Relation rel)
void _bt_metaversion(Relation rel, bool *heapkeyspace, bool *allequalimage)
void _bt_delitems_delete_check(Relation rel, Buffer buf, Relation heapRel, struct TM_IndexDeleteOp *delstate)
static BlockNumber BTreeTupleGetTopParent(IndexTuple leafhikey)
struct BTArrayKeyInfo BTArrayKeyInfo
void btadjustmembers(Oid opfamilyoid, Oid opclassoid, List *operators, List *functions)
BTScanInsert _bt_mkscankey(Relation rel, IndexTuple itup)
struct BTPendingFSM BTPendingFSM
void _bt_killitems(IndexScanDesc scan)
IndexTuple _bt_form_posting(IndexTuple base, ItemPointer htids, int nhtids)
bool _bt_doinsert(Relation rel, IndexTuple itup, IndexUniqueCheck checkUnique, bool indexUnchanged, Relation heapRel)
CompareType bttranslatestrategy(StrategyNumber strategy, Oid opfamily)
#define MaxTIDsPerBTreePage
static void BTreeTupleSetDownLink(IndexTuple pivot, BlockNumber blkno)
bool btgettuple(IndexScanDesc scan, ScanDirection dir)
void btparallelrescan(IndexScanDesc scan)
bool _bt_start_prim_scan(IndexScanDesc scan, ScanDirection dir)
struct BTVacState BTVacState
bool _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
IndexTuple _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright, BTScanInsert itup_key)
void _bt_set_cleanup_info(Relation rel, BlockNumber num_delpages)
#define BT_STATUS_OFFSET_MASK
static uint32 BTreeTupleGetPostingOffset(IndexTuple posting)
struct BTScanInsertData BTScanInsertData
void btbuildempty(Relation index)
struct BTMetaPageData BTMetaPageData
bool _bt_conditionallockbuf(Relation rel, Buffer buf)
Buffer _bt_getbuf(Relation rel, BlockNumber blkno, int access)
BTScanInsertData * BTScanInsert
int _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
static ItemPointer BTreeTupleGetPostingN(IndexTuple posting, int n)
int btgettreeheight(Relation rel)
void _bt_finish_split(Relation rel, Relation heaprel, Buffer lbuf, BTStack stack)
void btinitparallelscan(void *target)
IndexBulkDeleteResult * btbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats, IndexBulkDeleteCallback callback, void *callback_state)
void _bt_parallel_build_main(dsm_segment *seg, shm_toc *toc)
static bool BTPageIsRecyclable(Page page, Relation heaprel)
void _bt_unlockbuf(Relation rel, Buffer buf)
static BlockNumber BTreeTupleGetDownLink(IndexTuple pivot)
#define INDEX_ALT_TID_MASK
BTStack _bt_search(Relation rel, Relation heaprel, BTScanInsert key, Buffer *bufP, int access)
void _bt_upgradelockbufcleanup(Relation rel, Buffer buf)
void _bt_initmetapage(Page page, BlockNumber rootbknum, uint32 level, bool allequalimage)
BTVacuumPostingData * BTVacuumPosting
void _bt_set_startikey(IndexScanDesc scan, BTReadPageState *pstate)
IndexBuildResult * btbuild(Relation heap, Relation index, struct IndexInfo *indexInfo)
struct BTDeletedPageData BTDeletedPageData
bool _bt_vacuum_needs_cleanup(Relation rel)
OffsetNumber _bt_binsrch_insert(Relation rel, BTInsertState insertstate)
bool _bt_next(IndexScanDesc scan, ScanDirection dir)
char * btbuildphasename(int64 phasenum)
struct BTDedupInterval BTDedupInterval
int32 _bt_compare(Relation rel, BTScanInsert key, Page page, OffsetNumber offnum)
struct BTScanPosItem BTScanPosItem
bytea * btoptions(Datum reloptions, bool validate)
static ItemPointer BTreeTupleGetMaxHeapTID(IndexTuple itup)
static bool BTreeTupleIsPosting(IndexTuple itup)
Size BTreeShmemSize(void)
static void BTPageSetDeleted(Page page, FullTransactionId safexid)
int64 btgetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
void btmarkpos(IndexScanDesc scan)
BTDedupStateData * BTDedupState
void btendscan(IndexScanDesc scan)
void _bt_pendingfsm_finalize(Relation rel, BTVacState *vstate)
void _bt_lockbuf(Relation rel, Buffer buf, int access)
struct BTOptions BTOptions
void btrescan(IndexScanDesc scan, ScanKey scankey, int nscankeys, ScanKey orderbys, int norderbys)
bool btproperty(Oid index_oid, int attno, IndexAMProperty prop, const char *propname, bool *res, bool *isnull)
Buffer _bt_getroot(Relation rel, Relation heaprel, int access)
struct BTScanPosData BTScanPosData
static ItemPointer BTreeTupleGetHeapTID(IndexTuple itup)
void _bt_parallel_release(IndexScanDesc scan, BlockNumber next_scan_page, BlockNumber curr_page)
void _bt_pendingfsm_init(Relation rel, BTVacState *vstate, bool cleanuponly)
Size _bt_dedup_finish_pending(Page newpage, BTDedupState state)
struct BTScanOpaqueData BTScanOpaqueData
BTInsertStateData * BTInsertState
static void BTreeTupleSetNAtts(IndexTuple itup, uint16 nkeyatts, bool heaptid)
void btrestrpos(IndexScanDesc scan)
bool _bt_allequalimage(Relation rel, bool debugmessage)
struct BTDedupStateData BTDedupStateData
StaticAssertDecl(BT_OFFSET_MASK >=INDEX_MAX_KEYS, "BT_OFFSET_MASK can't fit INDEX_MAX_KEYS")
void _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir)
void _bt_preprocess_keys(IndexScanDesc scan)
struct BTStackData BTStackData
BTCycleId _bt_start_vacuum(Relation rel)
BTScanPosData * BTScanPos
struct BTInsertStateData BTInsertStateData
struct BTVacuumPostingData BTVacuumPostingData
BTScanOpaqueData * BTScanOpaque
bool GlobalVisCheckRemovableFullXid(Relation rel, FullTransactionId fxid)
static const struct fns functions
BTDedupInterval intervals[MaxIndexTuplesPerPage]
FullTransactionId safexid
uint32 btm_last_cleanup_num_delpages
float8 btm_last_cleanup_num_heap_tuples
float8 vacuum_cleanup_index_scale_factor
FullTransactionId safexid
ScanKeyData scankeys[INDEX_MAX_KEYS]
BTArrayKeyInfo * arrayKeys
MemoryContext arrayContext
BTScanPosItem items[MaxTIDsPerBTreePage]
LocationIndex tupleOffset
struct BTStackData * bts_parent
IndexBulkDeleteResult * stats
BTPendingFSM * pendingpages
IndexBulkDeleteCallback callback
MemoryContext pagedelcontext
uint16 deletetids[FLEXIBLE_ARRAY_MEMBER]
OffsetNumber updatedoffset
static void callback(struct sockaddr *addr, struct sockaddr *mask, void *unused)
#define FirstNormalFullTransactionId