Redeploys or restarts never complete if ngrok sessions are limited to 1 about kubernetes-ingress-controller

[alex-bezek]

Forcing the old one to close before the new one can start will result in a blip in uptime, so we still want to wait on the session to fully start if we can, but we still shouldn't block forever in the event that it never will.

Is it possible to wait for the new session to start before closing the old one if there is a limit of 1 agent session for free accounts? If you are limited to a single active session, it seems we'll always have to tear one down before creating the new one which will always have a small blip of downtime for free accounts.

We technically could make a helm configuration that allows free accounts to specify a Recreate Rollout Strategy instead of a Rolling Update, but the DevX for free accounts to remember to add a specific helm override isn't the greatest.

[jrobsonchase]

Is it possible to wait for the new session to start before closing the old one if there is a limit of 1 agent session for free accounts? If you are limited to a single active session, it seems we'll always have to tear one down before creating the new one which will always have a small blip of downtime for free accounts.

That's basically what we're doing already, but there's no way to tell the first session that it's time to shut down while the new one is coming up. The new instance can start, but it'll never get marked Ready until its tunnel session starts, which means the original will never get told to shut down.

We could go ahead and mark it "Ready" without tunnel session readiness, which will shut down the original, and cause the unavoidable blip for free accounts, and a racey blip for accounts with higher limits but only a single controller instance. If there are multiple controller pods, and they don't all go out simultaneously, it should be fine. That's probably the simplest fix.

The other is to check the error that comes back from a failure to start the tunnel session, and go ahead and mark the pod as Ready if it's a limit error.

[devpro]

Is it possible to update the Helm chart to be able to override the default spec.strategy.type (RollingUpdate -> Recreate)?