feat: Enable pod anti-affinity by default for Modelmesh Controller #272
Conversation
|
cc @njhill |
|
@anishasthana sorry for the delay. I'm unsure of whether we want to do this by default since the controller doesn't really require HA. If it is restarted / temporarily down there should not be much noticeable impact to the running services since its only part of the control plane and the api server will still accept requests. Do you know whether it's typical practice for kubernetes controllers to have multiple replicas, e.g. in OpenShift? I'm fine with adding the anti-affinity rule, and users are free to increase the number of replicas, just not sure of setting the default to > 1 (and thereby increasing the baseline footprint). |
The pod anti-affinity will tell the Kubernetes schedule to schedule pods on different nodes if possible. This will allow users to enable HA by setting the number of replicas to a number greater than 1 Signed-off-by: Anish Asthana <[email protected]>
|
@njhill makes sense! I've added the pod anti-affinity as well as a comment letting users know it's safe to increase the replicas. I don't know if it's "typical" practice, but I know a lot of SRE teams and the like greatly prefer having multiple pods where possible to minimize impact, even if the impact itself is barely noticeable. |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: anishasthana, njhill The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
|
/lgtm |
njhill
changed the title
The pod anti-affinity will tell the Kubernetes schedule to schedule pods on different nodes if possible. This will allow users to enable HA by setting the number of replicas to a number greater than 1.