piraeus-controller goes into NotReady constantly
alemonmk opened this issue · 3 comments
readinessProbe settings are too tight that it went into NotReady then Ready loop constantly.
The default timeoutSeconds of 1s (probably injected by kubelet) caused so many of "Timeout exceeded while awaiting headers" since I noticed it took a while to respond to linstor controller list-properties.
I increased the timeoutSeconds to 5s then it stayed at ready. I think we should define it in the deployment file. I also increased the periodSeconds so it does not spam the rest-access.log.
By the way, why are we mounting /etc/localtime? It caused me headaches because /etc/localtime on Fedora CoreOS for some reason is a empty directory. I don't see the reason since other deployments of Linstor does not do this.
Good catch. Previously, I was trying to get the entire cluster up as fast as possible. So will follow the readinessProbe change.
Mounting /etc/localtime is meant to set the pod timezone to that of the host. I have noticed it not working on some hosts. Let me try some other methods for this.
changes have been made accordingly.
relent readinessProbe to 5 sec
change timezone to /usr/share/zoneinfo/Etc/UTC format
Hey, I looked into your commit, actually the culprit is timeoutSeconds not periodSeconds (though it's probably better to change it as well), I think 3 seconds is enough, 5 would be a safer bet. Also only controller need to be changed, other things are perfectly fine with old settings.