Before exiting walreceiver, fsync() all the WAL received.

Otherwise WAL recovery will replay the un-flushed WAL after walreceiver has
exited, which can lead to a non-recoverable standby if the system crashes hard
at that point.

diff --git a/src/backend/replication/walreceiver.c b/src/backend/replication/walreceiver.c
index 06646ead94a468524bede8ba79590aafcdf03580..df4610732b1179304c9e5615771b47a2796ab010 100644 (file)
--- a/src/backend/replication/walreceiver.c
+++ b/src/backend/replication/walreceiver.c
@@ -323,6 +323,9 @@ WalRcvDie(int code, Datum arg)
    /* use volatile pointer to prevent code rearrangement */
    volatile WalRcvData *walrcv = WalRcv;
 
+   /* Ensure that all WAL records received are flushed to disk */
+   XLogWalRcvFlush();
+
    SpinLockAcquire(&walrcv->mutex);
    Assert(walrcv->walRcvState == WALRCV_RUNNING ||
           walrcv->walRcvState == WALRCV_STOPPING);