Fix unconditional WAL receiver shutdown during stream-archive transition
authorMichael Paquier <[email protected]>
Tue, 4 Nov 2025 01:52:38 +0000 (10:52 +0900)
committerMichael Paquier <[email protected]>
Tue, 4 Nov 2025 01:52:38 +0000 (10:52 +0900)
Commit b4f584f9d2a1 (affecting v15~, later backpatched down to 13 as of
3635a0a35aaf) introduced an unconditional WAL receiver shutdown when
switching from streaming to archive WAL sources.  This causes problems
during a timeline switch, when a WAL receiver enters WALRCV_WAITING
state but remains alive, waiting for instructions.

The unconditional shutdown can break some monitoring scenarios as the
WAL receiver gets repeatedly terminated and re-spawned, causing
pg_stat_wal_receiver.status to show a "streaming" instead of "waiting"
status, masking the fact that the WAL receiver is waiting for a new TLI
and a new LSN to be able to continue streaming.

This commit changes the WAL receiver behavior so as the shutdown becomes
conditional, with InstallXLogFileSegmentActive being always reset to
prevent the regression fixed by b4f584f9d2a1: only terminate the WAL
receiver when it is actively streaming (WALRCV_STREAMING,
WALRCV_STARTING, or WALRCV_RESTARTING).  When in WALRCV_WAITING state,
just reset InstallXLogFileSegmentActive flag to allow archive
restoration without killing the process.  WALRCV_STOPPED and
WALRCV_STOPPING are not reachable states in this code path.  For the
latter, the startup process is the one in charge of setting
WALRCV_STOPPING via ShutdownWalRcv(), waiting for the WAL receiver to
reach a WALRCV_STOPPED state after switching walRcvState, so
WaitForWALToBecomeAvailable() cannot be reached while a WAL receiver is
in a WALRCV_STOPPING state.

A regression test is added to check that a WAL receiver is not stopped
on timeline jump, that fails when the fix of this commit is reverted.

Reported-by: Ryan Bird <[email protected]>
Author: Xuneng Zhou <[email protected]>
Reviewed-by: Noah Misch <[email protected]>
Reviewed-by: Michael Paquier <[email protected]>
Discussion: https://postgr.es/m/19093-c4fff49a608f82a0@postgresql.org
Backpatch-through: 13

src/backend/access/transam/xlog.c
src/backend/access/transam/xlogrecovery.c
src/include/access/xlog.h
src/test/recovery/t/004_timeline_switch.pl

index 196a49ccec7461f67709714eac0d3693d5e28a39..70cb88a7feead5e6a79df3c04b42b87298502c8f 100644 (file)
@@ -9015,10 +9015,7 @@ void
 XLogShutdownWalRcv(void)
 {
    ShutdownWalRcv();
-
-   LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
-   XLogCtl->InstallXLogFileSegmentActive = false;
-   LWLockRelease(ControlFileLock);
+   ResetInstallXLogFileSegmentActive();
 }
 
 /* Enable WAL file recycling and preallocation. */
@@ -9030,6 +9027,15 @@ SetInstallXLogFileSegmentActive(void)
    LWLockRelease(ControlFileLock);
 }
 
+/* Disable WAL file recycling and preallocation. */
+void
+ResetInstallXLogFileSegmentActive(void)
+{
+   LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
+   XLogCtl->InstallXLogFileSegmentActive = false;
+   LWLockRelease(ControlFileLock);
+}
+
 bool
 IsInstallXLogFileSegmentActive(void)
 {
index 0cc1ef22a663a8614d51965c9d653643c3c6f1bc..de49bd65c06362ba6cdcdf119365022d8b898d01 100644 (file)
@@ -3613,8 +3613,19 @@ WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
                     * Before we leave XLOG_FROM_STREAM state, make sure that
                     * walreceiver is not active, so that it won't overwrite
                     * WAL that we restore from archive.
+                    *
+                    * If walreceiver is actively streaming (or attempting to
+                    * connect), we must shut it down. However, if it's
+                    * already in WAITING state (e.g., due to timeline
+                    * divergence), we only need to reset the install flag to
+                    * allow archive restoration.
                     */
-                   XLogShutdownWalRcv();
+                   if (WalRcvStreaming())
+                       XLogShutdownWalRcv();
+                   else
+                   {
+                       ResetInstallXLogFileSegmentActive();
+                   }
 
                    /*
                     * Before we sleep, re-scan for possible new timelines if
index 48ca8523810ccf249215e633ae7dfec4ed7b10ed..ebb9eaade0ac9370de2960f018ec1b551ff57197 100644 (file)
@@ -257,6 +257,7 @@ extern void SwitchIntoArchiveRecovery(XLogRecPtr EndRecPtr, TimeLineID replayTLI
 extern void ReachedEndOfBackup(XLogRecPtr EndRecPtr, TimeLineID tli);
 extern void SetInstallXLogFileSegmentActive(void);
 extern bool IsInstallXLogFileSegmentActive(void);
+extern void ResetInstallXLogFileSegmentActive(void);
 extern void XLogShutdownWalRcv(void);
 
 /*
index edaef918454b489cf3f2535d78b3ac3450eb74ab..6702fd2d09e478eca87308b46ed47cdc409e91fd 100644 (file)
@@ -68,6 +68,14 @@ my $result =
   $node_standby_2->safe_psql('postgres', "SELECT count(*) FROM tab_int");
 is($result, qq(2000), 'check content of standby 2');
 
+# Check the logs, WAL receiver should not have been stopped while
+# transitioning to its new timeline.  There is no need to rely on an
+# offset in this check of the server logs: a new log file is used on
+# node restart when primary_conninfo is updated above.
+ok( !$node_standby_2->log_contains(
+       "FATAL: .* terminating walreceiver process due to administrator command"
+   ),
+   'WAL receiver should not be stopped across timeline jumps');
 
 # Ensure that a standby is able to follow a primary on a newer timeline
 # when WAL archiving is enabled.