--pid=host: error waiting for container child process

[edsantiago]

The following combination of from and run will occasionally fail:

# buildah from --pid=host alpine
alpine-working-container
# while :;do buildah run alpine-working-container date;done
...  [could be ten seconds, could be two minutes]
Fri Dec 13 02:47:14 UTC 2019
ERRO error waiting for container child process 193984: no child processes
                                                                          Fri Dec 13 02:47:15 UTC 2019

When this happens, buildah completely hangs and cannot be killed from the command-line: ^C, ^Z, ^\ are all ineffectual. killall buildah from another shell works, and the loop continues, and I can then ^C from the terminal.

It only seems to be --pid=host. I've tried --net=host, --ipc, --userns, --utc, and all combined; those all go on for a long time.

buildah-1.12.0-0.86.dev.gitbc8feee.fc32
conmon-2.0.4-0.2.dev.git42bce45.fc32
crun-0.10.6-1.fc32

I was not able to reproduce with buildah-1.11.4-2.fc30 & conmon-2.0.2-1.fc30 & runc-1.0.0-95.rc9.gitc1485a1.fc30, at least within 15 minutes.

[rhatdan]

Buildah does not use conmon.

[rhatdan]

@giuseppe Thoughts?

[giuseppe]

Buildah sets itself as child subreaper but actually it waits only for the container pid. This may leave intermediate processes created by the OCI runtime not reaped.

It is probably easier to fix it in crun, so I'll move it there

[giuseppe]

@edsantiago by the way, great test case!

[giuseppe]

@edsantiago new release available here: https://bodhi.fedoraproject.org/updates/FEDORA-2019-743e6dffa1

[edsantiago]

Cannot reproduce in crun-0.11-1.fc32 over a 10-minute test. I'm calling this fixed, thank you.