Restart on every update

[bcroq]

Is there a reason why the pods are re-created on every helm update?

I know that the date in date/deploy-date is the reason why k8s re-creates the pods, but why did you choose to ask for a new pod on each update?

[bokysan]

Because otherwise postfix won't get reloaded if any configs are changed. I do expect many more issues with people changing the config and not seeing it applied than knowing that they need to do a SIGHUP on all postfix and opendkim instances.

One option would be to send a SIGHUP to all running instances from a post-update job, but that would require quite a bit more coding. If you're willing to submit a patch, pull requests are welcome.

[bcroq]

Could there be a value that would disable this restart once you know your configuration is stable?

[bokysan]

Sure, there could be an option values.yml to disable injecting of date/deploy-date, but that would be at your own risk and I wouldn't be able to support support it.

May I ask what the use case it, as usually people don't need to redeploy the chart very often?

[bcroq]

This chart is used as a dependency in a umbrella chart so each time our application is upgraded its mail relay is restarted and is unavailable for a few seconds.

Now that the mail relay configuration is done, it would be great to keep it running when our application is upgraded.

[bokysan]

I will look into it but running at least two instances should resolve your issue -- kubernetes will do a rolling deployment and your mail relay will have 100% uptime.

[bcroq]

As proposed, I currently deploy using 2 replicas; rolling deployments make sure at least one pod is running.

I see you added recreateOnRedeploy in 3.3.0. I test it as soon as possible.

Thanks!

[bcroq]

Tested and works as expected.