Skip to content

Commit 6655a72

Browse files
hlinnakahlinnaka
committed
Use full 64-bit XID for checking if a deleted GiST page is old enough.
Otherwise, after a deleted page gets even older, it becomes unrecyclable again. B-tree has the same problem, and has had since time immemorial, but let's at least fix this in GiST, where this is new. Backpatch to v12, where GiST page deletion was introduced. Reviewed-by: Andrey Borodin Discussion: https://www.postgresql.org/message-id/835A15A5-F1B4-4446-A711-BF48357EB602%40yandex-team.ru
1 parent 9eb5607 commit 6655a72

File tree

9 files changed

+134
-28
lines changed

9 files changed

+134
-28
lines changed

‎src/backend/access/gist/gistutil.c

Lines changed: 21 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -882,9 +882,27 @@ gistNewBuffer(Relation r)
882882
bool
883883
gistPageRecyclable(Page page)
884884
{
885-
return PageIsNew(page) ||
886-
(GistPageIsDeleted(page) &&
887-
TransactionIdPrecedes(GistPageGetDeleteXid(page), RecentGlobalXmin));
885+
if (PageIsNew(page))
886+
return true;
887+
if (GistPageIsDeleted(page))
888+
{
889+
/*
890+
* The page was deleted, but when? If it was just deleted, a scan
891+
* might have seen the downlink to it, and will read the page later.
892+
* As long as that can happen, we must keep the deleted page around as
893+
* a tombstone.
894+
*
895+
* Compare the deletion XID with RecentGlobalXmin. If deleteXid <
896+
* RecentGlobalXmin, then no scan that's still in progress could have
897+
* seen its downlink, and we can recycle it.
898+
*/
899+
FullTransactionId deletexid_full = GistPageGetDeleteXid(page);
900+
FullTransactionId recentxmin_full = GetFullRecentGlobalXmin();
901+
902+
if (FullTransactionIdPrecedes(deletexid_full, recentxmin_full))
903+
return true;
904+
}
905+
return false;
888906
}
889907

890908
bytea *

‎src/backend/access/gist/gistvacuum.c

Lines changed: 3 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -595,7 +595,7 @@ gistdeletepage(IndexVacuumInfo *info, GistBulkDeleteResult *stats,
595595
ItemId iid;
596596
IndexTuple idxtuple;
597597
XLogRecPtr recptr;
598-
TransactionId txid;
598+
FullTransactionId txid;
599599

600600
/*
601601
* Check that the leaf is still empty and deletable.
@@ -648,14 +648,13 @@ gistdeletepage(IndexVacuumInfo *info, GistBulkDeleteResult *stats,
648648
* currently in progress must have ended. (That's much more conservative
649649
* than needed, but let's keep it safe and simple.)
650650
*/
651-
txid = ReadNewTransactionId();
651+
txid = ReadNextFullTransactionId();
652652

653653
START_CRIT_SECTION();
654654

655655
/* mark the page as deleted */
656656
MarkBufferDirty(leafBuffer);
657-
GistPageSetDeleteXid(leafPage, txid);
658-
GistPageSetDeleted(leafPage);
657+
GistPageSetDeleted(leafPage, txid);
659658
stats->stats.pages_deleted++;
660659

661660
/* remove the downlink from the parent */

‎src/backend/access/gist/gistxlog.c

Lines changed: 25 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -356,8 +356,7 @@ gistRedoPageDelete(XLogReaderState *record)
356356
{
357357
Page page = (Page) BufferGetPage(leafBuffer);
358358

359-
GistPageSetDeleteXid(page, xldata->deleteXid);
360-
GistPageSetDeleted(page);
359+
GistPageSetDeleted(page, xldata->deleteXid);
361360

362361
PageSetLSN(page, lsn);
363362
MarkBufferDirty(leafBuffer);
@@ -396,8 +395,27 @@ gistRedoPageReuse(XLogReaderState *record)
396395
*/
397396
if (InHotStandby)
398397
{
399-
ResolveRecoveryConflictWithSnapshot(xlrec->latestRemovedXid,
400-
xlrec->node);
398+
FullTransactionId latestRemovedFullXid = xlrec->latestRemovedFullXid;
399+
FullTransactionId nextFullXid = ReadNextFullTransactionId();
400+
uint64 diff;
401+
402+
/*
403+
* ResolveRecoveryConflictWithSnapshot operates on 32-bit
404+
* TransactionIds, so truncate the logged FullTransactionId. If the
405+
* logged value is very old, so that XID wrap-around already happened
406+
* on it, there can't be any snapshots that still see it.
407+
*/
408+
nextFullXid = ReadNextFullTransactionId();
409+
diff = U64FromFullTransactionId(nextFullXid) -
410+
U64FromFullTransactionId(latestRemovedFullXid);
411+
if (diff < MaxTransactionId / 2)
412+
{
413+
TransactionId latestRemovedXid;
414+
415+
latestRemovedXid = XidFromFullTransactionId(latestRemovedFullXid);
416+
ResolveRecoveryConflictWithSnapshot(latestRemovedXid,
417+
xlrec->node);
418+
}
401419
}
402420
}
403421

@@ -554,7 +572,7 @@ gistXLogSplit(bool page_is_leaf,
554572
* downlink from the parent page.
555573
*/
556574
XLogRecPtr
557-
gistXLogPageDelete(Buffer buffer, TransactionId xid,
575+
gistXLogPageDelete(Buffer buffer, FullTransactionId xid,
558576
Buffer parentBuffer, OffsetNumber downlinkOffset)
559577
{
560578
gistxlogPageDelete xlrec;
@@ -578,7 +596,7 @@ gistXLogPageDelete(Buffer buffer, TransactionId xid,
578596
* Write XLOG record about reuse of a deleted page.
579597
*/
580598
void
581-
gistXLogPageReuse(Relation rel, BlockNumber blkno, TransactionId latestRemovedXid)
599+
gistXLogPageReuse(Relation rel, BlockNumber blkno, FullTransactionId latestRemovedXid)
582600
{
583601
gistxlogPageReuse xlrec_reuse;
584602

@@ -591,7 +609,7 @@ gistXLogPageReuse(Relation rel, BlockNumber blkno, TransactionId latestRemovedXi
591609
/* XLOG stuff */
592610
xlrec_reuse.node = rel->rd_node;
593611
xlrec_reuse.block = blkno;
594-
xlrec_reuse.latestRemovedXid = latestRemovedXid;
612+
xlrec_reuse.latestRemovedFullXid = latestRemovedXid;
595613

596614
XLogBeginInsert();
597615
XLogRegisterData((char *) &xlrec_reuse, SizeOfGistxlogPageReuse);

‎src/backend/access/rmgrdesc/gistdesc.c

Lines changed: 7 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -26,10 +26,11 @@ out_gistxlogPageUpdate(StringInfo buf, gistxlogPageUpdate *xlrec)
2626
static void
2727
out_gistxlogPageReuse(StringInfo buf, gistxlogPageReuse *xlrec)
2828
{
29-
appendStringInfo(buf, "rel %u/%u/%u; blk %u; latestRemovedXid %u",
29+
appendStringInfo(buf, "rel %u/%u/%u; blk %u; latestRemovedXid %u:%u",
3030
xlrec->node.spcNode, xlrec->node.dbNode,
3131
xlrec->node.relNode, xlrec->block,
32-
xlrec->latestRemovedXid);
32+
EpochFromFullTransactionId(xlrec->latestRemovedFullXid),
33+
XidFromFullTransactionId(xlrec->latestRemovedFullXid));
3334
}
3435

3536
static void
@@ -50,8 +51,10 @@ out_gistxlogPageSplit(StringInfo buf, gistxlogPageSplit *xlrec)
5051
static void
5152
out_gistxlogPageDelete(StringInfo buf, gistxlogPageDelete *xlrec)
5253
{
53-
appendStringInfo(buf, "deleteXid %u; downlink %u",
54-
xlrec->deleteXid, xlrec->downlinkOffset);
54+
appendStringInfo(buf, "deleteXid %u:%u; downlink %u",
55+
EpochFromFullTransactionId(xlrec->deleteXid),
56+
XidFromFullTransactionId(xlrec->deleteXid),
57+
xlrec->downlinkOffset);
5558
}
5659

5760
void

‎src/backend/utils/time/snapmgr.c

Lines changed: 30 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -956,6 +956,36 @@ xmin_cmp(const pairingheap_node *a, const pairingheap_node *b, void *arg)
956956
return 0;
957957
}
958958

959+
/*
960+
* Get current RecentGlobalXmin value, as a FullTransactionId.
961+
*/
962+
FullTransactionId
963+
GetFullRecentGlobalXmin(void)
964+
{
965+
FullTransactionId nextxid_full;
966+
uint32 nextxid_epoch;
967+
TransactionId nextxid_xid;
968+
uint32 epoch;
969+
970+
Assert(TransactionIdIsNormal(RecentGlobalXmin));
971+
972+
/*
973+
* Compute the epoch from the next XID's epoch. This relies on the fact
974+
* that RecentGlobalXmin must be within the 2 billion XID horizon from the
975+
* next XID.
976+
*/
977+
nextxid_full = ReadNextFullTransactionId();
978+
nextxid_epoch = EpochFromFullTransactionId(nextxid_full);
979+
nextxid_xid = XidFromFullTransactionId(nextxid_full);
980+
981+
if (RecentGlobalXmin > nextxid_xid)
982+
epoch = nextxid_epoch - 1;
983+
else
984+
epoch = nextxid_epoch;
985+
986+
return FullTransactionIdFromEpochAndXid(epoch, RecentGlobalXmin);
987+
}
988+
959989
/*
960990
* SnapshotResetXmin
961991
*

‎src/include/access/gist.h

Lines changed: 40 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -16,6 +16,7 @@
1616
#ifndef GIST_H
1717
#define GIST_H
1818

19+
#include "access/transam.h"
1920
#include "access/xlog.h"
2021
#include "access/xlogdefs.h"
2122
#include "storage/block.h"
@@ -140,8 +141,6 @@ typedef struct GISTENTRY
140141
#define GIST_LEAF(entry) (GistPageIsLeaf((entry)->page))
141142

142143
#define GistPageIsDeleted(page) ( GistPageGetOpaque(page)->flags & F_DELETED)
143-
#define GistPageSetDeleted(page) ( GistPageGetOpaque(page)->flags |= F_DELETED)
144-
#define GistPageSetNonDeleted(page) ( GistPageGetOpaque(page)->flags &= ~F_DELETED)
145144

146145
#define GistTuplesDeleted(page) ( GistPageGetOpaque(page)->flags & F_TUPLES_DELETED)
147146
#define GistMarkTuplesDeleted(page) ( GistPageGetOpaque(page)->flags |= F_TUPLES_DELETED)
@@ -158,9 +157,45 @@ typedef struct GISTENTRY
158157
#define GistPageGetNSN(page) ( PageXLogRecPtrGet(GistPageGetOpaque(page)->nsn))
159158
#define GistPageSetNSN(page, val) ( PageXLogRecPtrSet(GistPageGetOpaque(page)->nsn, val))
160159

161-
/* For deleted pages we store last xid which could see the page in scan */
162-
#define GistPageGetDeleteXid(page) ( ((PageHeader) (page))->pd_prune_xid )
163-
#define GistPageSetDeleteXid(page, val) ( ((PageHeader) (page))->pd_prune_xid = val)
160+
161+
/*
162+
* On a deleted page, we store this struct. A deleted page doesn't contain any
163+
* tuples, so we don't use the normal page layout with line pointers. Instead,
164+
* this struct is stored right after the standard page header. pd_lower points
165+
* to the end of this struct. If we add fields to this struct in the future, we
166+
* can distinguish the old and new formats by pd_lower.
167+
*/
168+
typedef struct GISTDeletedPageContents
169+
{
170+
/* last xid which could see the page in a scan */
171+
FullTransactionId deleteXid;
172+
} GISTDeletedPageContents;
173+
174+
static inline void
175+
GistPageSetDeleted(Page page, FullTransactionId deletexid)
176+
{
177+
Assert(PageIsEmpty(page));
178+
179+
GistPageGetOpaque(page)->flags |= F_DELETED;
180+
((PageHeader) page)->pd_lower = MAXALIGN(SizeOfPageHeaderData) + sizeof(GISTDeletedPageContents);
181+
182+
((GISTDeletedPageContents *) PageGetContents(page))->deleteXid = deletexid;
183+
}
184+
185+
static inline FullTransactionId
186+
GistPageGetDeleteXid(Page page)
187+
{
188+
Assert(GistPageIsDeleted(page));
189+
190+
/* Is the deleteXid field present? */
191+
if (((PageHeader) page)->pd_lower >= MAXALIGN(SizeOfPageHeaderData) +
192+
offsetof(GISTDeletedPageContents, deleteXid) + sizeof(FullTransactionId))
193+
{
194+
return ((GISTDeletedPageContents *) PageGetContents(page))->deleteXid;
195+
}
196+
else
197+
return FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
198+
}
164199

165200
/*
166201
* Vector of GISTENTRY structs; user-defined methods union and picksplit

‎src/include/access/gist_private.h

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -426,11 +426,11 @@ extern SplitedPageLayout *gistSplit(Relation r, Page page, IndexTuple *itup,
426426

427427
/* gistxlog.c */
428428
extern XLogRecPtr gistXLogPageDelete(Buffer buffer,
429-
TransactionId xid, Buffer parentBuffer,
429+
FullTransactionId xid, Buffer parentBuffer,
430430
OffsetNumber downlinkOffset);
431431

432432
extern void gistXLogPageReuse(Relation rel, BlockNumber blkno,
433-
TransactionId latestRemovedXid);
433+
FullTransactionId latestRemovedXid);
434434

435435
extern XLogRecPtr gistXLogUpdate(Buffer buffer,
436436
OffsetNumber *todelete, int ntodelete,

‎src/include/access/gistxlog.h

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -83,7 +83,7 @@ typedef struct gistxlogPageSplit
8383
*/
8484
typedef struct gistxlogPageDelete
8585
{
86-
TransactionId deleteXid; /* last Xid which could see page in scan */
86+
FullTransactionId deleteXid; /* last Xid which could see page in scan */
8787
OffsetNumber downlinkOffset; /* Offset of downlink referencing this
8888
* page */
8989
} gistxlogPageDelete;
@@ -98,10 +98,10 @@ typedef struct gistxlogPageReuse
9898
{
9999
RelFileNode node;
100100
BlockNumber block;
101-
TransactionId latestRemovedXid;
101+
FullTransactionId latestRemovedFullXid;
102102
} gistxlogPageReuse;
103103

104-
#define SizeOfGistxlogPageReuse (offsetof(gistxlogPageReuse, latestRemovedXid) + sizeof(TransactionId))
104+
#define SizeOfGistxlogPageReuse (offsetof(gistxlogPageReuse, latestRemovedFullXid) + sizeof(FullTransactionId))
105105

106106
extern void gist_redo(XLogReaderState *record);
107107
extern void gist_desc(StringInfo buf, XLogReaderState *record);

‎src/include/utils/snapmgr.h

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -13,6 +13,7 @@
1313
#ifndef SNAPMGR_H
1414
#define SNAPMGR_H
1515

16+
#include "access/transam.h"
1617
#include "fmgr.h"
1718
#include "utils/relcache.h"
1819
#include "utils/resowner.h"
@@ -122,6 +123,8 @@ extern void UnregisterSnapshot(Snapshot snapshot);
122123
extern Snapshot RegisterSnapshotOnOwner(Snapshot snapshot, ResourceOwner owner);
123124
extern void UnregisterSnapshotFromOwner(Snapshot snapshot, ResourceOwner owner);
124125

126+
extern FullTransactionId GetFullRecentGlobalXmin(void);
127+
125128
extern void AtSubCommit_Snapshot(int level);
126129
extern void AtSubAbort_Snapshot(int level);
127130
extern void AtEOXact_Snapshot(bool isCommit, bool resetXmin);

0 commit comments

Comments
 (0)