Fix unlogged sequence corruption after standby promotion

Previously, if an unlogged sequence was created on the primary and
replicated to a standby, reading the sequence after promoting the
standby (for example, with nextval()) could trigger the following
assertion failure:

    TRAP: failed Assert("((const PageHeaderData *) page)->pd_special >= SizeOfPageHeaderData")

In non-assert builds, the same operation could instead fail with an
error such as:

    ERROR:  bad magic number in sequence

The problem was that seq_redo() updated the init fork page in shared
buffers but did not flush it to disk. During promotion,
ResetUnloggedRelations() recreates the main fork of unlogged
relations by copying the init fork from disk, bypassing shared
buffers. As a result, the main fork could be recreated from a stale
init fork instead of the WAL-replayed page.

Fix this by introducing a helper to flush init fork buffers
immediately, and make seq_redo() use it. As a result, the main fork
of an unlogged sequence is recreated from the up-to-date init fork on
disk, allowing the unlogged sequence to be read successfully after
standby promotion.

Backpatch to v15, where unlogged sequences were introduced.

Author: Fujii Masao <masao.fujii@gmail.com>
Reviewed-by: vignesh C <vignesh21@gmail.com>
Discussion: https://postgr.es/m/CAHGQGwH1Ssze3XM6wjoTjSLVOR041c6xP+vsdLP951=w8oG8bA@mail.gmail.com
Backpatch-through: 15

diff --git a/src/backend/access/hash/hash_xlog.c b/src/backend/access/hash/hash_xlog.c
index 4e1ec468569e183bb3f10e08da24fdfd8c5b8abf..de88697f191b9fc311636c714133b4a92ed8274a 100644 (file)
--- 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 = (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 0bd82da4bb8516a5d8d8562c482b0953ed63be70..325e3d63042798ef3cb08605abc84508fed6ccf8 100644 (file)
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -332,6 +332,28 @@ 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.
@@ -409,7 +431,9 @@ 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.c b/src/backend/commands/sequence.c
index 2b64a480e161ef7ad53268b4ecd609cc6777d235..d5aff5b5b7ca7ae24c6741523a8edf12cec9fb04 100644 (file)
--- a/src/backend/commands/sequence.c
+++ b/src/backend/commands/sequence.c
@@ -1875,6 +1875,7 @@ 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 e24613e8f8174a9b7ba0393753dcdaecd58bb389..14d4f77da28f9dd501509f1fa1ad87b0fe42299a 100644 (file)
--- a/src/include/access/xlogutils.h
+++ b/src/include/access/xlogutils.h
@@ -84,6 +84,8 @@ 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,

diff --git a/src/test/recovery/meson.build b/src/test/recovery/meson.build
index 7a36de5eea4cb9713b8d1eec11157e1616a51930..fc7d21b6a5bcd28ca57b8421dc48c789e25b8050 100644 (file)
--- a/src/test/recovery/meson.build
+++ b/src/test/recovery/meson.build
@@ -57,6 +57,7 @@ tests += {
       't/047_checkpoint_physical_slot.pl',
       't/048_vacuum_horizon_floor.pl',
       't/053_standby_login_event_trigger.pl',
+      't/054_unlogged_sequence_promotion.pl',
     ],
   },
 }

diff --git a/src/test/recovery/t/054_unlogged_sequence_promotion.pl b/src/test/recovery/t/054_unlogged_sequence_promotion.pl
new file mode 100644 (file)
index 0000000..96d1e4b
--- /dev/null
+++ b/src/test/recovery/t/054_unlogged_sequence_promotion.pl
@@ -0,0 +1,34 @@
+# Copyright (c) 2026, PostgreSQL Global Development Group
+
+# Test that unlogged sequences created on a primary can be read after
+# promotion of a standby that replayed their init fork.
+
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $node_primary = PostgreSQL::Test::Cluster->new('primary');
+$node_primary->init(allows_streaming => 1);
+$node_primary->start;
+
+my $backup_name = 'my_backup';
+$node_primary->backup($backup_name);
+
+my $node_standby = PostgreSQL::Test::Cluster->new('standby');
+$node_standby->init_from_backup($node_primary, $backup_name,
+       has_streaming => 1);
+$node_standby->start;
+
+# Create the unlogged sequence after the standby has started, so its init fork
+# is generated by WAL replay on the standby.
+$node_primary->safe_psql('postgres', "CREATE UNLOGGED SEQUENCE ulseq");
+$node_primary->wait_for_replay_catchup($node_standby);
+
+$node_standby->promote;
+
+is($node_standby->safe_psql('postgres', "SELECT nextval('ulseq')"),
+       1, 'unlogged sequence can be read after standby promotion');
+
+done_testing();