From b91ca1dfef3b6b7f4fbbe87ddd86d1f2939ef7f8 Mon Sep 17 00:00:00 2001 From: Fujii Masao Date: Wed, 24 Jun 2026 18:42:43 +0900 Subject: [PATCH v1] Fix promoted-standby reads of unlogged sequences --- src/backend/access/hash/hash_xlog.c | 29 +++----------------------- src/backend/access/transam/xlogutils.c | 26 ++++++++++++++++++++++- src/backend/commands/sequence_xlog.c | 1 + src/include/access/xlogutils.h | 2 ++ 4 files changed, 31 insertions(+), 27 deletions(-) diff --git a/src/backend/access/hash/hash_xlog.c b/src/backend/access/hash/hash_xlog.c index 2060620c7de..e9a2b9aa9a7 100644 --- a/src/backend/access/hash/hash_xlog.c +++ b/src/backend/access/hash/hash_xlog.c @@ -29,7 +29,6 @@ hash_xlog_init_meta_page(XLogReaderState *record) XLogRecPtr lsn = record->EndRecPtr; Page page; Buffer metabuf; - ForkNumber forknum; xl_hash_init_meta_page *xlrec = (xl_hash_init_meta_page *) XLogRecGetData(record); @@ -41,16 +40,7 @@ hash_xlog_init_meta_page(XLogReaderState *record) page = BufferGetPage(metabuf); PageSetLSN(page, lsn); MarkBufferDirty(metabuf); - - /* - * Force the on-disk state of init forks to always be in sync with the - * state in shared buffers. See XLogReadBufferForRedoExtended. We need - * special handling for init forks as create index operations don't log a - * full page image of the metapage. - */ - XLogRecGetBlockTag(record, 0, NULL, &forknum, NULL); - if (forknum == INIT_FORKNUM) - FlushOneBuffer(metabuf); + XLogFlushBufferForRedoIfInit(record, 0, metabuf); /* all done */ UnlockReleaseBuffer(metabuf); @@ -68,7 +58,6 @@ hash_xlog_init_bitmap_page(XLogReaderState *record) Page page; HashMetaPage metap; uint32 num_buckets; - ForkNumber forknum; xl_hash_init_bitmap_page *xlrec = (xl_hash_init_bitmap_page *) XLogRecGetData(record); @@ -79,16 +68,7 @@ hash_xlog_init_bitmap_page(XLogReaderState *record) _hash_initbitmapbuffer(bitmapbuf, xlrec->bmsize, true); PageSetLSN(BufferGetPage(bitmapbuf), lsn); MarkBufferDirty(bitmapbuf); - - /* - * Force the on-disk state of init forks to always be in sync with the - * state in shared buffers. See XLogReadBufferForRedoExtended. We need - * special handling for init forks as create index operations don't log a - * full page image of the metapage. - */ - XLogRecGetBlockTag(record, 0, NULL, &forknum, NULL); - if (forknum == INIT_FORKNUM) - FlushOneBuffer(bitmapbuf); + XLogFlushBufferForRedoIfInit(record, 0, bitmapbuf); UnlockReleaseBuffer(bitmapbuf); /* add the new bitmap page to the metapage's list of bitmaps */ @@ -109,10 +89,7 @@ hash_xlog_init_bitmap_page(XLogReaderState *record) PageSetLSN(page, lsn); MarkBufferDirty(metabuf); - - XLogRecGetBlockTag(record, 1, NULL, &forknum, NULL); - if (forknum == INIT_FORKNUM) - FlushOneBuffer(metabuf); + XLogFlushBufferForRedoIfInit(record, 1, metabuf); } if (BufferIsValid(metabuf)) UnlockReleaseBuffer(metabuf); diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c index fdc341d8fa4..d8c179c5dcc 100644 --- a/src/backend/access/transam/xlogutils.c +++ b/src/backend/access/transam/xlogutils.c @@ -321,6 +321,28 @@ XLogInitBufferForRedo(XLogReaderState *record, uint8 block_id) return buf; } +/* + * If a redo routine modified an init fork, flush the buffer immediately. + * + * At the end of crash recovery the init forks of unlogged relations are + * copied to the main fork directly from disk, without going through shared + * buffers. Therefore, redo routines that update init forks without + * restoring a full-page image must call this after setting the page LSN and + * marking the buffer dirty. + */ +void +XLogFlushBufferForRedoIfInit(XLogReaderState *record, uint8 block_id, + Buffer buffer) +{ + ForkNumber forknum; + + Assert(BufferIsValid(buffer)); + + XLogRecGetBlockTag(record, block_id, NULL, &forknum, NULL); + if (forknum == INIT_FORKNUM) + FlushOneBuffer(buffer); +} + /* * XLogReadBufferForRedoExtended * Like XLogReadBufferForRedo, but with extra options. @@ -398,7 +420,9 @@ XLogReadBufferForRedoExtended(XLogReaderState *record, * At the end of crash recovery the init forks of unlogged relations * are copied, without going through shared buffers. So we need to * force the on-disk state of init forks to always be in sync with the - * state in shared buffers. + * state in shared buffers. Use XLogFlushBufferForRedoIfInit() for + * redo routines that dirty init-fork buffers without restoring a + * full-page image. */ if (forknum == INIT_FORKNUM) FlushOneBuffer(*buf); diff --git a/src/backend/commands/sequence_xlog.c b/src/backend/commands/sequence_xlog.c index d0aed48e268..fcb3230cf3b 100644 --- a/src/backend/commands/sequence_xlog.c +++ b/src/backend/commands/sequence_xlog.c @@ -63,6 +63,7 @@ seq_redo(XLogReaderState *record) memcpy(page, localpage, BufferGetPageSize(buffer)); MarkBufferDirty(buffer); + XLogFlushBufferForRedoIfInit(record, 0, buffer); UnlockReleaseBuffer(buffer); pfree(localpage); diff --git a/src/include/access/xlogutils.h b/src/include/access/xlogutils.h index b97387c6d4c..0c6c7410069 100644 --- a/src/include/access/xlogutils.h +++ b/src/include/access/xlogutils.h @@ -87,6 +87,8 @@ typedef struct ReadLocalXLogPageNoWaitPrivate extern XLogRedoAction XLogReadBufferForRedo(XLogReaderState *record, uint8 block_id, Buffer *buf); extern Buffer XLogInitBufferForRedo(XLogReaderState *record, uint8 block_id); +extern void XLogFlushBufferForRedoIfInit(XLogReaderState *record, + uint8 block_id, Buffer buffer); extern XLogRedoAction XLogReadBufferForRedoExtended(XLogReaderState *record, uint8 block_id, ReadBufferMode mode, bool get_cleanup_lock, -- 2.53.0