fix(resource): properly handle TERM signal after catch

[feiyudev]

TERM signal was catched but not handled properly, which causing waiting so long to delete a pod

Fixes #10658

Motivation

pod can't be stopped via argo terminate, works after I changed the context.

Modifications

changed the parent context in argoexec resource subcommand.

Verification

1. stop pod via kubectl delete pod, no more 30s needed.
2. stop via argo terminate works.

[juliev0]

Thanks for fixing. Can you clarify where "argoexec resource" is invoked?

[feiyudev]

Thanks for fixing. Can you clarify where "argoexec resource" is invoked?

When resource defined in one step, then the pod to run this step, generated by workflow controller, runs commands as follows:

/usr/bin/containerd-shim-runc-v2 -namespace k8s.io -id <id> -address /run/containerd/containerd.sock
 \_ /pause
 \_ /var/run/argo/argoexec emissary --loglevel info --log-format text -- argoexec resource apply --loglevel debug
     \_ /bin/argoexec resource apply --loglevel debug

[juliev0]

Thanks for fixing. Can you clarify where "argoexec resource" is invoked?

When resource defined in one step, then the pod to run this step, generated by workflow controller, runs commands as follows:

/usr/bin/containerd-shim-runc-v2 -namespace k8s.io -id <id> -address /run/containerd/containerd.sock
 \_ /pause
 \_ /var/run/argo/argoexec emissary --loglevel info --log-format text -- argoexec resource apply --loglevel debug
     \_ /bin/argoexec resource apply --loglevel debug

Thank you! I’m OOO right now but I’ll review next week when back if this hasn’t been merged by then.

[feiyudev]

All right, thanks anyway.

[alexec]

LGTM

[agilgur5]

This seems to have effectively been a follow-up to #10523 (comment), if I'm understanding the history correctly