The supervisor configuration currently has a restart policy of true which always restarts the child process if it stops. Using the graceful leave API as mentioned in the documentation on performing updates causes the agent to leave and then shutdown, but supervisor currently restarts it immediately. This causes the agent to re-join the cluster, thus making a graceful shutdown impossible without access to stop supervisor itself.
The supervisor configuration currently has a restart policy of
truewhich always restarts the child process if it stops. Using the graceful leave API as mentioned in the documentation on performing updates causes the agent to leave and then shutdown, but supervisor currently restarts it immediately. This causes the agent to re-join the cluster, thus making a graceful shutdown impossible without access to stop supervisor itself.