Occasionally, maybe 1/2500 sessions, the browser pod associated with the session will get stuck and not terminate indefinitely (we've waited a few days). There's nothing in the logs we've found to differentiate a "clean" session and "stuck" session.
We've worked around this with a cronjob that deletes pods older than a given age, but it would be nice if that wasn't necessary.
The K8s pod spec supports an activeDeadlineSeconds field that instructs K8s to kill pods over a certain age. Could Moon take advantage of this to help ensure pods are going away? It could be possibly be configurable via a browserset.
Occasionally, maybe 1/2500 sessions, the browser pod associated with the session will get stuck and not terminate indefinitely (we've waited a few days). There's nothing in the logs we've found to differentiate a "clean" session and "stuck" session.
We've worked around this with a cronjob that deletes pods older than a given age, but it would be nice if that wasn't necessary.
The K8s pod spec supports an
activeDeadlineSecondsfield that instructs K8s to kill pods over a certain age. Could Moon take advantage of this to help ensure pods are going away? It could be possibly be configurable via a browserset.