Disable synchronous commits in pg_rewind.

hlinnaka · hlinnaka · commit aab809664813 · 2016-10-06T13:34:38.000+03:00
If you point pg_rewind to a server that is using synchronous replication,
with "pg_rewind --source-server=...", and the replication is not working
for some reason, pg_rewind will get stuck because it creates a temporary
table, which needs to be replicated. You could call broken replication a
pilot error, but pg_rewind is often used in special circumstances, when
there are changes to the replication setup.

We don't do any "real" updates, and we don't care about fsyncing or
replicating the operations on the temporary tables, so fix that by
setting synchronous_commit off.

Michael Banck, Michael Paquier. Backpatch to 9.5, where pg_rewind was
introduced.

Discussion: &lt;20161005143938.GA12247@nighthawk.caipicrew.dd-dns.de&gt;
diff --git a/src/bin/pg_rewind/libpq_fetch.c b/src/bin/pg_rewind/libpq_fetch.c
@@ -49,6 +49,7 @@ void
 libpqConnect(const char *connstr)
 {
 	char	   *str;
+	PGresult   *res;
 
 	conn = PQconnectdb(connstr);
 	if (PQstatus(conn) == CONNECTION_BAD)
@@ -77,6 +78,19 @@ libpqConnect(const char *connstr)
 	if (strcmp(str, "on") != 0)
 		pg_fatal("full_page_writes must be enabled in the source server\n");
 	pg_free(str);
+
+	/*
+	 * Although we don't do any "real" updates, we do work with a temporary
+	 * table. We don't care about synchronous commit for that. It doesn't
+	 * otherwise matter much, but if the server is using synchronous
+	 * replication, and replication isn't working for some reason, we don't
+	 * want to get stuck, waiting for it to start working again.
+	 */
+	res = PQexec(conn, "SET synchronous_commit = off");
+	if (PQresultStatus(res) != PGRES_COMMAND_OK)
+		pg_fatal("could not set up connection context: %s",
+				 PQresultErrorMessage(res));
+	PQclear(res);
 }
 
 /*
