regression: podman run: exit status 0 after 'rm --force`

[edsantiago]

# (podman run --name foo alpine sleep 30;echo $?)&
[1] 13212
# podman rm --force foo
[pause 10 seconds]
57c5063848af9f40b436d224ea2ef5808c5db591b6cd1aa0addb3be7e3eddf34
# 0   <--- this is the exit status. It should be 137

podman-1.5.0-2.fc30.x86_64

[rhatdan]

Why should it be 137? Didn't podman rm --force foo work?

[mheon]

When trying to debug this, I found that events is acting properly strange.
podman events --since 1h --until 1s on 1.4.4 exits instantly, shows me the events in question. In 1.5.0, it seems to hang.

[edsantiago]

podman rm worked. The exit status pointed to by the arrow is that of the container itself, which was killed, therefore its exit status should be 137. Another way to say this:

In window 1:

# podman run --name foo alpine sleep 30

In window 2:

# podman rm -f foo

Then, after the 10-second wait, go back to window 1:

# echo $?
0

Sorry for not being clear.

[edsantiago]

Side note: podman kill foo and podman stop foo both result in a container exit status of 137, as expected. It's only rm -f that is weird.

[edsantiago]

@mheon I cannot reproduce that behavior. I tried with a running container, without; also with running container and running podman rm -f foo during the 10-second wait. The podman events command in question always dumps a bunch of lines, then exits 0.

[mheon]

It's journald. We're scanning every single journal entry on the system. It's taking 30 seconds to complete on my desktop.

[mheon]

@baude We're probably going to need a revert of #3745 - it nuked performance of events on systems with long journals.

[rhatdan]

This seems like something we should fix before releasing to Fedora.

[mheon]

I think I got the core issue here - problem was that we were testing with file backend and not journald. Might want to look into that later.

The events performance thing is separate, and probably does require a revert.

[edsantiago]

Oh, yuk:

# podman run --rm alpine sh -c /bin/false;echo $?
0    <--- should be 1

Without --rm, works as expected (1). Without sh -c, also works as expected.

This may merit a separate issue but I've got a gut sense that they're related. LMK.

[mheon]

Same issue. Fix in progress.

[mheon]

#3797 to fix both issues