Allow daemonset to handle termination signals #171
Conversation
images/entrypoint.sh
Outdated
| trap - SIGINT SIGTERM # clear trap | ||
| exit 0 | ||
| } | ||
| trap exitonsigterm SIGTERM SIGINT |
There was a problem hiding this comment.
I noticed we use exec (https://github.com/k8snetworkplumbingwg/sriov-network-device-plugin/blob/master/images/entrypoint.sh#L70) in sriov device plugin to capture sigterm signal, what is the advantage of using a trap function?
There was a problem hiding this comment.
@zshi-redhat Do you mean calling exec sleep 1234 so that sleep will be pid 1 and receive SIGTERM? I never thought of that! I tested it and it doesn't work as expected.
I think sleep does not handle sigterm but I cannot confirm from POSIX spec.
So, trap allows us to capture a SIGTERM and force exit. If sleep handled a SIGTERM, exec would be fine. I pushed a commit to simplify what I had previously. There was no need to clear the trap.
Without this patch deleting the pod takes 1m30s.
With this patch delete the pod takes 30s.
There was a problem hiding this comment.
@zshi-redhat Do you mean calling
exec sleep 1234so that sleep will be pid 1 and receive SIGTERM?
Yes
I never thought of that! I tested it and it doesn't work as expected.
Thanks for testing it!
I think sleep does not handle sigterm but I cannot confirm from POSIX spec.So, trap allows us to capture a SIGTERM and force exit. If sleep handled a SIGTERM, exec would be fine. I pushed a commit to simplify what I had previously. There was no need to clear the trap.
Without this patch deleting the pod takes 1m30s.
With this patch delete the pod takes 30s.
Big improvement!
Thanks for the fix!
martinkennelly
force-pushed
the
fix_sigintterm
branch
from
087fc06 to
5cabf49
Compare
Signed-off-by: Martin Kennelly <martin.kennelly@intel.com>
martinkennelly
force-pushed
the
fix_sigintterm
branch
from
5cabf49 to
6b8a32a
Compare
Signed-off-by: Martin Kennelly martin.kennelly@intel.com