Rabbitmq Statefulset podManagementPolicy
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| openstack-helm-infra |
New
|
Undecided
|
yann degat | ||
Bug Description
Recently, the podManagementPolicy in statefulSets have been set to "Parallel" mode
instead of the default "OrderedReady" mode, in rabbitmq & mariadb helm packages.
(see commit: https:/
It can leads to race conditions issues, especially in the rabbitmq case where, if ever nodes are launched in parallel but havent reached a certain state, they can discover each other but fail to form a cluster.
(doc: https:/
As this "cluster forming" step isn't retried, the pods then run in standalone mode.
It happened to me a lot when i tried to launched 2 nodes clusters for tests purposes.
Dont know why it doesn't happened more with 3 nodes clusters.
Switching back explicitly to the default "OrderedReady" value of podManagementPolicy solved the issue as the second nodes starts once the first one is up & ready.
| Changed in openstack-helm-infra: | |
| assignee: | nobody → yann degat (yann-degat-z) |
