Are the upgrades seamless?
I'm new to openshift and I heard some stories that for instance, when you upgrade Openshift Container Platform, there are usually issues with the EFK stack that is deployed on top.
This comes from the fact that the OCP upgrade procedure is not aware of the running application on top.
(I didn't verify it myself, but this is just to give a context to my question, and if it is not the case, then the question is still relevant)
My question is then, what about CNS? Do we have successful stories of customers upgrading from say OCP 3.7 to 3.9 with more than 5 glusterfs nodes and with many pods using CNS without any downtime? And if yes, what are the recommendations in the planning to avoid downtime during the upgrade.
Responses
Active Contributor
107 points
Hey there, in fact there have been some issues with OCP Upgrades. We also hit the EFK Stack Bug, which was an error with the provided Upgrade Ansible Playbook (all Elasticsearch master have been restarted simultaneously). It is fiexd now by the way ;-) If you plan to upgrade there should be a few things too keep in Mind: 1. The first step is to upgrade the controll plane (masters and etcds) 2. After that you start to upgrade the Metrics and the logging stack (which should normaly run fine) 3. When all these steps were successfully you can start to upgrade the nodes one by one (drain, openshift docker unexclude, playbook run, reboot, make node schedulable) Until now I have no experience with CNS, but as GFS is a HA FileService, I think it should cause no trouble when some bricks reboot, as long as they have a quorum. It should be als flawlessly as updating the etcds.
Active Contributor
107 points
And in addition to that, your own applications should always be designed to handle sudden pod restarts, which always can occur. Wether due to some maintenance by a sysadmin or due to node failures. This means you should have a concept regarding stateful pods. Stateless Pods should always run on more than one replica, so they won't care if they get evicted. Since CNS is a thing which redhat supports (like the Platform EFK Stack, Metrics and master etcds) they have custom upgrade Playbooks which take care of the overall health while upgrading. So you should be perfectly fine.
Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.