Make sure scala pods can be gracefully terminated within 15 seconds
Previously, we introduced graceful termination logic for all our core repo pods. See #3380
For scala pods, it's done via Akka Coordinated Shutdown. Our implementation includes a 5s waiting period in case of any chunked response and 15shardTerminationDeadline.
However, on GKE, the actual GracePeriod might be much shorter (down to 15s). See this comment.
We should adjust the settings of Coordinated Shutdown i.e.:
Make sure scala pods can be gracefully terminated within 15 seconds
Previously, we introduced graceful termination logic for all our core repo pods. See #3380
For scala pods, it's done via Akka Coordinated Shutdown. Our implementation includes a
5s
waiting period in case of any chunked response and15s
hardTerminationDeadline
.However, on GKE, the actual GracePeriod might be much shorter (down to 15s). See this comment.
We should adjust the settings of
Coordinated Shutdown
i.e.:http.waitBeforeTermination
http.hardTerminationDeadline
akka.coordinated-shutdown.phases.service-unbind.timeout
akka.coordinated-shutdown.phases.service-requests-done.timeout
to make sure the termination can be completed within 15s.
Those parameters should also be configurable via Helm charts