]> granicus.if.org Git - postgresql/commitdiff
Make logical WAL sender report streaming state appropriately
authorMichael Paquier <michael@paquier.xyz>
Thu, 12 Jul 2018 01:20:14 +0000 (10:20 +0900)
committerMichael Paquier <michael@paquier.xyz>
Thu, 12 Jul 2018 01:20:14 +0000 (10:20 +0900)
WAL senders sending logically-decoded data fail to properly report in
"streaming" state when starting up, hence as long as one extra record is
not replayed, such WAL senders would remain in a "catchup" state, which
is inconsistent with the physical cousin.

This can be easily reproduced by for example using pg_recvlogical and
restarting the upstream server.  The TAP tests have been slightly
modified to detect the failure and strengthened so as future tests also
make sure that a node is in streaming state when waiting for its
catchup.

Backpatch down to 9.4 where this code has been introduced.

Reported-by: Sawada Masahiko
Author: Simon Riggs, Sawada Masahiko
Reviewed-by: Petr Jelinek, Michael Paquier, Vaishnavi Prabakaran
Discussion: https://postgr.es/m/CAD21AoB2ZbCCqOx=bgKMcLrAvs1V0ZMqzs7wBTuDySezTGtMZA@mail.gmail.com

src/backend/replication/walsender.c

index 4e06ffc6862d23ed7a7d4195671c488c591c6514..ac11bf572442307cc94526b41a7a59185e537d97 100644 (file)
@@ -1926,8 +1926,8 @@ WalSndLoop(WalSndSendDataCallback send_data)
                        if (MyWalSnd->state == WALSNDSTATE_CATCHUP)
                        {
                                ereport(DEBUG1,
-                                        (errmsg("standby \"%s\" has now caught up with primary",
-                                                        application_name)));
+                                               (errmsg("\"%s\" has now caught up with upstream server",
+                                                               application_name)));
                                WalSndSetState(WALSNDSTATE_STREAMING);
                        }
 
@@ -2483,10 +2483,10 @@ XLogSendLogical(void)
        char       *errm;
 
        /*
-        * Don't know whether we've caught up yet. We'll set it to true in
-        * WalSndWaitForWal, if we're actually waiting. We also set to true if
-        * XLogReadRecord() had to stop reading but WalSndWaitForWal didn't wait -
-        * i.e. when we're shutting down.
+        * Don't know whether we've caught up yet. We'll set WalSndCaughtUp to
+        * true in WalSndWaitForWal, if we're actually waiting. We also set to
+        * true if XLogReadRecord() had to stop reading but WalSndWaitForWal
+        * didn't wait - i.e. when we're shutting down.
         */
        WalSndCaughtUp = false;
 
@@ -2499,9 +2499,19 @@ XLogSendLogical(void)
 
        if (record != NULL)
        {
+               /* XXX: Note that logical decoding cannot be used while in recovery */
+               XLogRecPtr      flushPtr = GetFlushRecPtr();
+
                LogicalDecodingProcessRecord(logical_decoding_ctx, logical_decoding_ctx->reader);
 
                sentPtr = logical_decoding_ctx->reader->EndRecPtr;
+
+               /*
+                * If we have sent a record that is at or beyond the flushed point, we
+                * have caught up.
+                */
+               if (sentPtr >= flushPtr)
+                       WalSndCaughtUp = true;
        }
        else
        {