From: Frédéric Lécaille <flecaille@haproxy.com>
Date: Mon, 18 Jan 2021 14:14:39 +0000 (+0100)
Subject: BUG/MINOR: peers: Wrong "new_conn" value for "show peers" CLI command.
X-Git-Tag: v2.4-dev6~46
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=2b0ba54ddb7d555bd0116eb8278b37ac409e7779;p=thirdparty%2Fhaproxy.git

BUG/MINOR: peers: Wrong "new_conn" value for "show peers" CLI command.

This counter could be hugely incremented by the peer task responsible of managing
peer synchronizations and reconnections, for instance when a peer is not reachable
there is a period where the appctx is not created. If we receive  stick-table
updates before the peer session (appctx) is instantiated, we reach the code
responsible of incrementing the "new_conn" counter.
With this patch we increment this counter only when we really instantiate a new
peer session thanks to peer_session_create().

May be backported as far as 2.0.
---

diff --git a/src/peers.c b/src/peers.c
index 754d73a8a8..f4553bc8de 100644
--- a/src/peers.c
+++ b/src/peers.c
@@ -2807,6 +2807,7 @@ static struct appctx *peer_session_create(struct peers *peers, struct peer *peer
 	struct session *sess;
 	struct stream *s;
 
+	peer->new_conn++;
 	peer->reconnect = tick_add(now_ms, MS_TO_TICKS(PEER_RECONNECT_TIMEOUT));
 	peer->heartbeat = TICK_ETERNITY;
 	peer->statuscode = PEER_SESS_SC_CONNECTCODE;
@@ -2931,7 +2932,6 @@ static struct task *process_peer_sync(struct task * task, void *context, unsigne
 
 						/* reschedule task for reconnect */
 						task->expire = tick_first(task->expire, ps->reconnect);
-						ps->new_conn++;
 					}
 					/* else do nothing */
 				} /* !ps->appctx */