76 lines
2.8 KiB
Diff
76 lines
2.8 KiB
Diff
|
From 4357f0dbb8668ac4090cd7070c2ea195e5683326 Mon Sep 17 00:00:00 2001
|
||
|
|
From: Damien Ciabrini <dciabrin@redhat.com>
|
||
|
|
Date: Wed, 24 Jan 2024 13:27:26 +0100
|
||
|
|
Subject: [PATCH] galera: allow joiner to report non-Primary during initial IST
|
||
|
|
|
||
|
|
It seems that with recent galera versions, when a galera node
|
||
|
|
joins a cluster, there is a small time window where the node is
|
||
|
|
connected to the primary component of the galera cluster, but it
|
||
|
|
might still be preparing its IST. During this time, it can report
|
||
|
|
itself as being 'not ready' and in 'non-primary' state.
|
||
|
|
|
||
|
|
Update the galera resource agent to allow the node to be in
|
||
|
|
non-primary state, but only if running a "promote" operation. Any
|
||
|
|
network partition during the promotion will be caught by the
|
||
|
|
promote timeout.
|
||
|
|
|
||
|
|
In reworking the promotion code, we move the check for primary
|
||
|
|
partition into the "galera_monitor" function. The check works
|
||
|
|
as before for regular "monitor" or "probe" operations.
|
||
|
|
|
||
|
|
Related-Bug: rhbz#2255414
|
||
|
|
---
|
||
|
|
heartbeat/galera.in | 25 +++++++++++++++++--------
|
||
|
|
1 file changed, 17 insertions(+), 8 deletions(-)
|
||
|
|
|
||
|
|
diff --git a/heartbeat/galera.in b/heartbeat/galera.in
|
||
|
|
index 6aed3e4b6d..b518595cb0 100755
|
||
|
|
--- a/heartbeat/galera.in
|
||
|
|
+++ b/heartbeat/galera.in
|
||
|
|
@@ -822,6 +822,11 @@ galera_promote()
|
||
|
|
return $rc
|
||
|
|
fi
|
||
|
|
|
||
|
|
+ # At this point, the mysql pidfile is created on disk and the
|
||
|
|
+ # mysql server is reacheable via its UNIX socket. If we are a
|
||
|
|
+ # joiner, SST transfers (rsync) have finished, but an IST may
|
||
|
|
+ # still be requested or ongoing
|
||
|
|
+
|
||
|
|
galera_monitor
|
||
|
|
rc=$?
|
||
|
|
if [ $rc != $OCF_SUCCESS -a $rc != $OCF_RUNNING_MASTER ]; then
|
||
|
|
@@ -835,12 +840,6 @@ galera_promote()
|
||
|
|
return $OCF_ERR_GENERIC
|
||
|
|
fi
|
||
|
|
|
||
|
|
- is_primary
|
||
|
|
- if [ $? -ne 0 ]; then
|
||
|
|
- ocf_exit_reason "Failure. Master instance started, but is not in Primary mode."
|
||
|
|
- return $OCF_ERR_GENERIC
|
||
|
|
- fi
|
||
|
|
-
|
||
|
|
if ocf_is_true $bootstrap; then
|
||
|
|
promote_everyone
|
||
|
|
clear_bootstrap_node
|
||
|
|
@@ -991,8 +990,18 @@ galera_monitor()
|
||
|
|
fi
|
||
|
|
rc=$OCF_RUNNING_MASTER
|
||
|
|
else
|
||
|
|
- ocf_exit_reason "local node <${NODENAME}> is started, but not in primary mode. Unknown state."
|
||
|
|
- rc=$OCF_ERR_GENERIC
|
||
|
|
+ # It seems that with recent galera (26.4+), a joiner that is
|
||
|
|
+ # connected to a Primary component and is preparing its IST
|
||
|
|
+ # request might still temporarily report its state as
|
||
|
|
+ # Non-Primary. Do not fail in this case as the promote
|
||
|
|
+ # operation will loop until the IST finishes or the promote
|
||
|
|
+ # times out.
|
||
|
|
+ if [ "$__OCF_ACTION" = "promote" ] && ! ocf_is_true $(is_bootstrap); then
|
||
|
|
+ ocf_log info "local node <${NODENAME}> is receiving a State Transfer."
|
||
|
|
+ else
|
||
|
|
+ ocf_exit_reason "local node <${NODENAME}> is started, but not in primary mode. Unknown state."
|
||
|
|
+ rc=$OCF_ERR_GENERIC
|
||
|
|
+ fi
|
||
|
|
fi
|
||
|
|
|
||
|
|
return $rc
|