- galera: fix issue where joiner promotion fails is the node reports
being in non-primary state Resolves: RHEL-31763
This commit is contained in:
Oyvind Albrigtsen
parent
d66a3df4d3
commit
580971cd28
75
RHEL-31763-galera-fix-joiner-promotion-fails-issue.patch
Normal file
75
RHEL-31763-galera-fix-joiner-promotion-fails-issue.patch
Normal file
@ -0,0 +1,75 @@
|
||||
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
|
||||
@ -45,7 +45,7 @@
|
||||
Name: resource-agents
|
||||
Summary: Open Source HA Reusable Cluster Resource Scripts
|
||||
Version: 4.10.0
|
||||
Release: 53%{?rcver:%{rcver}}%{?numcomm:.%{numcomm}}%{?alphatag:.%{alphatag}}%{?dirty:.%{dirty}}%{?dist}
|
||||
Release: 54%{?rcver:%{rcver}}%{?numcomm:.%{numcomm}}%{?alphatag:.%{alphatag}}%{?dirty:.%{dirty}}%{?dist}
|
||||
License: GPLv2+ and LGPLv2+
|
||||
URL: https://github.com/ClusterLabs/resource-agents
|
||||
Source0: %{upstream_prefix}-%{upstream_version}.tar.gz
|
||||
@ -119,6 +119,7 @@ Patch66: RHEL-16247-aws-vpc-move-ip-aws-vpc-route53-awseip-awsvip-auth_type-role
|
||||
Patch67: RHEL-17072-1-storage_mon-findif-leak-unitialized-values-EOS-fixes.patch
|
||||
Patch68: RHEL-17072-2-storage_mon-use-memset-to-fix-covscan-error.patch
|
||||
Patch69: RHEL-15304-2-findif.sh-dont-use-table-parameter.patch
|
||||
Patch70: RHEL-31763-galera-fix-joiner-promotion-fails-issue.patch
|
||||
|
||||
# bundled ha-cloud-support libs
|
||||
Patch500: ha-cloud-support-aws.patch
|
||||
@ -312,6 +313,7 @@ exit 1
|
||||
%patch -p1 -P 67
|
||||
%patch -p1 -P 68
|
||||
%patch -p1 -P 69
|
||||
%patch -p1 -P 70
|
||||
|
||||
# bundled ha-cloud-support libs
|
||||
%patch -p1 -P 500
|
||||
@ -633,6 +635,12 @@ rm -rf %{buildroot}/usr/share/doc/resource-agents
|
||||
%{_usr}/lib/ocf/lib/heartbeat/OCF_*.pm
|
||||
|
||||
%changelog
|
||||
* Wed Apr 10 2024 Oyvind Albrigtsen <oalbrigt@redhat.com> - 4.10.0-54
|
||||
- galera: fix issue where joiner promotion fails is the node reports
|
||||
being in non-primary state
|
||||
|
||||
Resolves: RHEL-31763
|
||||
|
||||
* Wed Mar 6 2024 Oyvind Albrigtsen <oalbrigt@redhat.com> - 4.10.0-53
|
||||
- aliyun-vpc-move-ip: use new aliyun-cli
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user