Do not re-elect RabbitMQ multistate resource master
Current monitor action for multistate RabbitMQ resource
in Pacemaker may re-elect existing master when there is
no need to do so. That is a problem as master elections
introduce full cluster downtime and we don't want
additional elections.
The fix is to check whether the given node uptime is equal
to the other nodes uptime and drop this node from the list
of candidates if there is a master exists. Note, there is a
special 'rabbit-master' attribute in CIB for this.
Reviewed: https:/ /review. openstack. org/171192 /git.openstack. org/cgit/ stackforge/ fuel-library/ commit/ ?id=78edb8cf6f0 524a9618d4e1e45 790369f8dda16e
Committed: https:/
Submitter: Jenkins
Branch: stable/6.0
commit 78edb8cf6f0524a 9618d4e1e457903 69f8dda16e
Author: Bogdan Dobrelya <email address hidden>
Date: Mon Mar 30 17:27:51 2015 +0200
Do not re-elect RabbitMQ multistate resource master
Current monitor action for multistate RabbitMQ resource
in Pacemaker may re-elect existing master when there is
no need to do so. That is a problem as master elections
introduce full cluster downtime and we don't want
additional elections.
The fix is to check whether the given node uptime is equal
to the other nodes uptime and drop this node from the list
of candidates if there is a master exists. Note, there is a
special 'rabbit-master' attribute in CIB for this.
Closes-bug: #1435254
Change-Id: I6f2484fb8f3284 f76461c184148d2 7aa86a5d4b6
Signed-off-by: Bogdan Dobrelya <email address hidden>