podman container wait --interval - what does the parameter do?

[pspacek]

Version: podman 5.4.0

I have trouble understanding the podman container wait man page. It currently says:

podman container wait [options] container [...]

   --interval, -i=duration
       Time interval to wait before polling for completion. A duration string is a  sequence  of  decimal  numbers,
       each  with  optional  fraction  and a unit suffix, such as "300ms", "-1.5h" or "2h45m". Valid time units are
       "ns", "us" (or "µs"), "ms", "s", "m", "h". Time unit defaults to "ms".

At the same time, the podman container wait --interval 10s definitely does not check only once per 10 seconds, or waits 10 seconds before checking. Example:

# podman run --detach --rm --name wait archlinux:latest sleep 3; time podman wait --interval 10s wait
2a1f13d27392ba9e6a080df65ba18b98bfe12ae6532d334123c14aedcf4253bf
0

real	0m3.316s
user	0m0.021s
sys	0m0.016s

If it were checking status only once per 10 seconds, or after first 10 seconds elapsed, it could not possibly detect the container existed after 3 seconds.

What am I missing?

Thank you!

[Luap99]

It is the polling interval, that is generally used for most container state checks (i.e. how often we sync and compare the container state).
However the default wait waits for container exit. And that is special coded since not to long ago (8a94331) as I converted it to use a pidfd, with that we have no need to poll actively, the kernel wake us if the pid exits with that so there is no extra delay.

[pspacek]

Thank you very much, this is very informative.

I need to verify how "lightweight" podman container wait is because in my use-case it will be running on a machine which executes a sensitive benchmark.

I peeked into strace -f podman container wait output and noticed it spawns 12 threads and calls roughly 120 nanosleep() syscalls per second, despide using ppoll to watch for the process exit. Is that expected?

Thank you again for your time!

[pspacek]

FTR here's strace.log generated using podman run --detach --rm --name wait archlinux:latest sleep 3; strace -o strace.log -yy -tt -T -f podman wait --interval 10s wait --latest.

[Luap99]

Podman is written in go and thus threads are managed by the go runtime. We have little control over it and the go runtime itself heavily use epoll and nanosleep I think so this is not something we call directly. Of course our code influences what the runtime does in one way or another.

Just looking at strace of wait

pidfd_open(23967, 0) = 7<pid:23967> <0.000013>
ppoll([{fd=7<pid:23967>, events=POLLIN}], 1, {tv_sec=10, tv_nsec=0}, NULL, 0 <unfinished ...>
....
<... ppoll resumed>) = 1 ([{fd=7, revents=POLLIN}], left {tv_sec=7, tv_nsec=10705245}) <2.989310>

That seems right the 10s interval is used for the ppoll call and it stopped after about 3s which is expected.

So unless someone has insides into the go runtime and how it uses epoll and nanosleep I am not sure there anything on our side to change.