v3-0001-pg_rewind-use-flush-lsn-to-set-minRecoveryPoint.patch

From 29de0c65cca4111c49357f61d06e7d2f50a97a2b Mon Sep 17 00:00:00 2001
From: ChangAo Chen <cca5507@qq.com>
Date: Tue, 16 Jun 2026 11:22:13 +0800
Subject: [PATCH v3] pg_rewind: use flush lsn to set minRecoveryPoint.

Currently pg_rewind uses pg_current_wal_insert_lsn() to set
minRecoveryPoint. The minRecoveryPoint is expected to be a
lsn points to a xlog record end lsn + 1, but the return value
of pg_current_wal_insert_lsn() is the current wal insert lsn.
They are different when at a xlog page boundary, e.g., when
the former points to the begin of a xlog page, the latter
will skip the xlog page header. This can lead to a standby
cannot reach consistent when the primary is idle.

To fix it, use pg_current_wal_flush_lsn() instead. It's safe
because we query flush lsn at the end of rewind, so all copied
pages should have lsn <= flush lsn. Using the flush lsn is also
crash-safe with respect to the source: the insert lsn lives only
in shared memory and can be lost on a source crash, leaving the
target's minRecoveryPoint ahead of any lsn the source can
subsequently reach.
---
 src/bin/pg_rewind/libpq_source.c  | 12 ++++++------
 src/bin/pg_rewind/local_source.c  |  2 +-
 src/bin/pg_rewind/pg_rewind.c     | 12 ++++++++++--
 src/bin/pg_rewind/rewind_source.h |  5 +++--
 4 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/src/bin/pg_rewind/libpq_source.c b/src/bin/pg_rewind/libpq_source.c
index 6955bc575ea..97435427803 100644
--- a/src/bin/pg_rewind/libpq_source.c
+++ b/src/bin/pg_rewind/libpq_source.c
@@ -68,7 +68,7 @@ static void libpq_queue_fetch_range(rewind_source *source, const char *path,
 static void libpq_finish_fetch(rewind_source *source);
 static char *libpq_fetch_file(rewind_source *source, const char *path,
 							  size_t *filesize);
-static XLogRecPtr libpq_get_current_wal_insert_lsn(rewind_source *source);
+static XLogRecPtr libpq_get_current_wal_flush_lsn(rewind_source *source);
 static void libpq_destroy(rewind_source *source);
 
 /*
@@ -91,7 +91,7 @@ init_libpq_source(PGconn *conn)
 	src->common.queue_fetch_file = libpq_queue_fetch_file;
 	src->common.queue_fetch_range = libpq_queue_fetch_range;
 	src->common.finish_fetch = libpq_finish_fetch;
-	src->common.get_current_wal_insert_lsn = libpq_get_current_wal_insert_lsn;
+	src->common.get_current_wal_flush_lsn = libpq_get_current_wal_flush_lsn;
 	src->common.destroy = libpq_destroy;
 
 	src->conn = conn;
@@ -202,10 +202,10 @@ run_simple_command(PGconn *conn, const char *sql)
 }
 
 /*
- * Call the pg_current_wal_insert_lsn() function in the remote system.
+ * Call the pg_current_wal_flush_lsn() function in the remote system.
  */
 static XLogRecPtr
-libpq_get_current_wal_insert_lsn(rewind_source *source)
+libpq_get_current_wal_flush_lsn(rewind_source *source)
 {
 	PGconn	   *conn = ((libpq_source *) source)->conn;
 	XLogRecPtr	result;
@@ -213,10 +213,10 @@ libpq_get_current_wal_insert_lsn(rewind_source *source)
 	uint32		lo;
 	char	   *val;
 
-	val = run_simple_query(conn, "SELECT pg_current_wal_insert_lsn()");
+	val = run_simple_query(conn, "SELECT pg_current_wal_flush_lsn()");
 
 	if (sscanf(val, "%X/%08X", &hi, &lo) != 2)
-		pg_fatal("unrecognized result \"%s\" for current WAL insert location", val);
+		pg_fatal("unrecognized result \"%s\" for current WAL flush location", val);
 
 	result = ((uint64) hi) << 32 | lo;
 
diff --git a/src/bin/pg_rewind/local_source.c b/src/bin/pg_rewind/local_source.c
index 4841cf01fb7..029f13a7396 100644
--- a/src/bin/pg_rewind/local_source.c
+++ b/src/bin/pg_rewind/local_source.c
@@ -46,7 +46,7 @@ init_local_source(const char *datadir)
 	src->common.queue_fetch_file = local_queue_fetch_file;
 	src->common.queue_fetch_range = local_queue_fetch_range;
 	src->common.finish_fetch = local_finish_fetch;
-	src->common.get_current_wal_insert_lsn = NULL;
+	src->common.get_current_wal_flush_lsn = NULL;
 	src->common.destroy = local_destroy;
 
 	src->datadir = datadir;
diff --git a/src/bin/pg_rewind/pg_rewind.c b/src/bin/pg_rewind/pg_rewind.c
index 9d745d4b25b..4867d3abf00 100644
--- a/src/bin/pg_rewind/pg_rewind.c
+++ b/src/bin/pg_rewind/pg_rewind.c
@@ -712,12 +712,20 @@ perform_rewind(filemap_t *filemap, rewind_source *source,
 		{
 			/*
 			 * Source is a production, non-standby, server. We must replay to
-			 * the last WAL insert location.
+			 * the last WAL flush location.
+			 *
+			 * Use the source's flush LSN as the target's minRecoveryPoint: every
+			 * WAL-logged page we copied has page-LSN <= source's flush LSN at
+			 * copy time (WAL-before-data), and flush LSN is monotonic. We avoid
+			 * using the insert LSN because it can sit one page-header past a
+			 * record's end at page boundaries (where no record will end), and
+			 * it is not durable, a source crash can leave flush LSN behind an
+			 * insert LSN we already pinned.
 			 */
 			if (ControlFile_source_after.state != DB_IN_PRODUCTION)
 				pg_fatal("source system was in unexpected state at end of rewind");
 
-			endrec = source->get_current_wal_insert_lsn(source);
+			endrec = source->get_current_wal_flush_lsn(source);
 			endtli = Max(ControlFile_source_after.checkPointCopy.ThisTimeLineID,
 						 ControlFile_source_after.minRecoveryPointTLI);
 		}
diff --git a/src/bin/pg_rewind/rewind_source.h b/src/bin/pg_rewind/rewind_source.h
index f2e9b84a732..0b9f643e22b 100644
--- a/src/bin/pg_rewind/rewind_source.h
+++ b/src/bin/pg_rewind/rewind_source.h
@@ -66,9 +66,10 @@ typedef struct rewind_source
 	void		(*finish_fetch) (struct rewind_source *);
 
 	/*
-	 * Get the current WAL insert position in the source system.
+	 * Get the current WAL flush position in the source system. To use
+	 * this callback, the source must be a live primary server.
 	 */
-	XLogRecPtr	(*get_current_wal_insert_lsn) (struct rewind_source *);
+	XLogRecPtr	(*get_current_wal_flush_lsn) (struct rewind_source *);
 
 	/*
 	 * Free this rewind_source object.
-- 
2.34.1