OCF Pacemaker resources for DB/MQ clusters should run with requires=nothing
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Fuel for OpenStack |
Fix Committed
|
Medium
|
Bogdan Dobrelya | ||
Mitaka |
Fix Released
|
Medium
|
Bogdan Dobrelya | ||
Newton |
Fix Committed
|
Medium
|
Bogdan Dobrelya |
Bug Description
Jepsen tests shown [0] a Galera cluster may handle network partitions on its own, without quorum policy enforced by Pacemaker, and maintain serializable transactions in multi-master mode w/o consistency issues. This makes the quorum requirement for it an overkill. Same for MQ cluster (although should be postponed until jepsen tests TBD).
Note, that this change shall not apply for fencing. But Fuel configures quorum policy, no fencing. This cannot cover cases when a node become unresponsive. An unresponsive node would behave the same way and will fail to stop resources with requires=
UX: this issue impacts only AV and self-healing (recovery) of clusters: with requires=nothing, cluster members will not be stopped by a Pacemaker, which is
expected to bring less issues with self-healing, like this one https:/
Changed in fuel: | |
milestone: | none → 10.0 |
importance: | Undecided → High |
tags: | added: galera pacemaker rabbitmq tech-debt |
description: | updated |
description: | updated |
description: | updated |
Changed in fuel: | |
assignee: | nobody → Fuel Sustaining (fuel-sustaining-team) |
status: | New → Confirmed |
Changed in fuel: | |
assignee: | Fuel Sustaining (fuel-sustaining-team) → Bogdan Dobrelya (bogdando) |
Fix proposed to branch: master /review. openstack. org/314031
Review: https:/