From: Simon Riggs <simon@2ndQuadrant.com>
Date: Sun, 23 Apr 2017 20:35:41 +0000 (+0100)
Subject: Fix LagTrackerRead() for timeline increments
X-Git-Tag: REL_10_BETA1~201
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=8463880872cf8a6fc6bb9a4ad34b84aded7a33e2;p=postgresql

Fix LagTrackerRead() for timeline increments

Bug was masked by error in running 004_timeline_switch.pl that was
fixed recently in 7d68f2281a.

Detective work by Alvaro Herrera and Tom Lane

Author: Thomas Munro
---

diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index 26090738fc..064cf5ee28 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -3328,7 +3328,16 @@ LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now)
 			WalTimeSample prev = LagTracker.last_read[head];
 			WalTimeSample next = LagTracker.buffer[LagTracker.read_heads[head]];
 
-			Assert(lsn >= prev.lsn);
+			if (lsn < prev.lsn)
+			{
+				/*
+				 * Reported LSNs shouldn't normally go backwards, but it's
+				 * possible when there is a timeline change.  Treat as not
+				 * found.
+				 */
+				return -1;
+			}
+
 			Assert(prev.lsn < next.lsn);
 
 			if (prev.time > next.time)