Graceful shutdown endlessly blocks after in-flight tasks have completed

[davidquarles]

Confirmed this is happening on both 0.14 and 0.16 and I can't for the life of me understand why. We have some very long-running workflows so our terminationGracePeriodSeconds (running in Kubernetes) is necessarily quite high. When these workflows are not running, shutdown should be nearly instantaneous, but we are blocked for eons despite the fact that atlantis has already logged server: All in-progress operations complete, shutting down, waiting fot the eventual, subsequent SIGKILL to actually shut down the process.

Am I missing something? Is there some hidden flag we need to be setting or some feature we're using which prevents these two versions from fulfilling the documented behavior? It feels like we're doing something wrong, since there is no evidence of anyone else encountering this issue, but maybe we are simply the first to report it. Any help is appreciated.

Thanks!

[nishkrishnan]

Hmmm, this doesn't happen for us. It happens almost immediately. We have Atlantis deployed as a statefulset. Can you pin point where exactly it's getting stuck in the code?

[jamengual]

is this still relevant and tested with v0.19.8?