docker event for exec_die

[ndeloof]

Description

As a container terminate by itself (main PID 1 process complete) an container die event is reported with exit code.
As an exec process completes, docker inspect do report updated ExecIDs set but no event is logged. As a result, one only can monitor exec completion by polling. An container exec_die event would allow an asynchronous implementation.

My actual use case is to run build processes in an existing containers from a Jenkins plugin. I'd like to avoid polling this container for exec completion.

Steps to reproduce the issue:

1. run docker event
2. start a container with sleep 100
3. run docker exec ID echo hello

Describe the results you received:

logged events:

• container exec_create
• container exec_start
• container die

Describe the results you expected:

logged events:

• container exec_create
• container exec_start
• container exec_die
• container die

[thaJeztah]

Discussing this in the maintainers meeting, and we think adding these events make sense.

There is some discussion around order of events for container die and container exec_die (weren't sure if they can guaranteed to arrive in the correct order); that may need some research.

Are you interested in implementing this?

[ndeloof]

Yes, thanks !

[ndeloof]

I'm not worried about the arrival order, could just be documented not to be guaranteed if there's a risk for some race condition.
For my own use-case I just need some way to know an exec completed without polling.

[ndeloof]

#35744