Regression: podman exec --preserve-fds: hangs

[edsantiago]

Similar to #4020 but opening as new issue because I have no idea where the root cause is. CI is failing with hangs in the --preserve-fds test. When I tried to reproduce, I got:

tool	works	hangs
podman	1.7.0-3.fc30	1.8.0-4.fc30
conmon	2.0.2-1.fc30	2.0.13-1.fc30
runc	1.0.0-95.rc9.gitc1485a1.fc30	1.0.0-102.dev.gitdc9208a.fc30

Bizarrely, I dnf-upgraded the 'works' system to match all the versions in 'hangs' --- and everything still works. I know it's not just my systems because this is causing CI failures

Symptom:

# podman run --name test1 -d docker.io/library/alpine:latest top
39c9fd570fe7634acfd66140de740472ff611f7d6be6f302927fe38bdfc816ca
# podman exec --preserve-fds 1 test1 date
Sat Mar 28 14:55:36 UTC 2020
[hangs]

If I try a ^C - !! loop, sometimes it exits cleanly, sometimes it hangs with no output at all, but usually it just does the same thing: echo one line, hang.

Originally posted by @edsantiago in #4020 (comment)

[edsantiago]

Another flake: PR 5667, ubuntu-18 remote

[giuseppe]

have you opened the fd 3 before passing it down to Podman and podman exec --preserve-fds 1?

Otherwise it will pass some random fd down to conmon, the runtime and in the container. Depending on what random fd is passed down, it can end up in a hang.

Could you try with: podman exec --preserve-fds 1 test1 date 3<> /dev/null ?

[edsantiago]

[feeling embarrassed today]. I thought I had written all that in #4020, but nope, doesn't look like it.

Yes, I tried a huge number of variants of 3<> and 3< and 3> (and/or 4, and/or 2, and/or 0). I no longer have them in my terminal history, but the gist is: nothing helps.

[giuseppe]

thanks.

The issue seems to be in conmon, the good news is that it is fixed in conmon upstream, probably as part of: containers/conmon#139

[github-actions]

A friendly reminder that this issue had no activity for 30 days.

[rhatdan]

Since this is fixed in master, I am going to close.