Install graceful shutdown on server #18
Conversation
There was a problem hiding this comment.
I don't understand much but it looks a lot like the example implem so it's OK.
Thanks for the PR, sorry I didn't see it sooner, we are in the process of moving to https://github.com/Trow-Registry/trow
| @@ -226,6 +226,10 @@ impl TrowBuilder { | |||
| // Start GRPC Backend task | |||
| tokio::task::spawn(init_trow_server(self.config.clone())?); | |||
|
|
|||
| // Listen for termination signal | |||
| let handle = axum_server::Handle::new(); | |||
There was a problem hiding this comment.
I would prefer handle to be called shutdown_handle, but it's OK
This commit enables gracefull shutdown on the server to handle SIGTERM signals from the cluster and improve service restarts
awoimbee
force-pushed
the
graceful-shutdown
branch
from
4f9fe61
to
b338e4b
Compare
|
Thanks for merging it. I'm glad the transfer worked out. I guess I could have done a better job explaining the need for this change. While I was tuning resource utilization and healthchecks timeout, I've noticed that It would hang for up-to 30s the graceful time expired and get SIGKILL-ed. This caused a larger delay as the scheduler has to wait for the process to stop to start another instance (in my case). I believe this was being caused by the GRPC + Axum spawning tasks that would not receive the termination signal. With this change, calling ctrl-c, for example, immediatelly asks the reactor and axum to terminate connections and tasks, so it can finish faster and be re-scheduled. |
This commit enables gracefull shutdown on the server to handle SIGTERM signals from the cluster and improve service restarts