[CI]: docker now fails (a lot) TestLogsWithStartContainer

[apostasie]

Description

docker rm -f foo; docker run -ti --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- "while true; do echo lol; sleep 1; done"

kill -s INT DOCKER_PID

Will work fine in docker 26.

No longer works in Docker 28.

Maybe this is related to #5302

[Shubhranshu153]

i do notice it as well. Will try to fix it.

[apostasie]

Update.

This is possibly a docker bug:

dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ sudo docker rm -f foo; sudo docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- "sleep inf" &
Error response from daemon: No such container: foo
[1] 38915
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ ps aux | grep docker
root        2609  0.0  1.9 2412376 78320 ?       Ssl  May05   2:01 /usr/bin/dockerd -H fd:// --containerd=/run/containerd/containerd.sock
root       15424  0.0  0.1 1671296 4224 ?        Sl   May05   0:01 /usr/bin/docker-proxy -proto tcp -host-ip 0.0.0.0 -host-port 443 -container-ip 172.17.0.3 -container-port 443 -use-listen-fd
root       15430  0.0  0.1 1745344 4096 ?        Sl   May05   0:01 /usr/bin/docker-proxy -proto tcp -host-ip :: -host-port 443 -container-ip 172.17.0.3 -container-port 443 -use-listen-fd
root       38915  0.0  0.1  17692  5632 pts/0    S    10:32   0:00 sudo docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- sleep inf
root       38916  0.0  0.0  17692  2280 pts/2    Ss+  10:32   0:00 sudo docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- sleep inf
root       38917  0.0  0.7 1773548 28400 pts/2   Sl   10:32   0:00 docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- sleep inf
dmp        39362  0.0  0.0   6672  1792 pts/0    S+   10:44   0:00 grep --color=auto docker
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ jobs
[1]+  Running                 sudo docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- "sleep inf" &
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ kill -s INT 38915
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ jobs
[1]+  Running                 sudo docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- "sleep inf" &
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ jobs
[1]+  Running                 sudo docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- "sleep inf" &
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ kill -s INT 38915
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ jobs
[1]+  Running                 sudo docker run --sig-proxy=false --name foo ghcr.io/stargz-containers/alpine:3.13-org sh -c -- "sleep inf" &
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$ kill -s INT 38915
dmp@lima-docker-rootful:/Users/dmp/Projects/go/nerd/nerdctl$
got 3 SIGTERM/SIGINTs, forcefully exiting

[apostasie]

Tagging @thaJeztah if you have free cycle for a quick consulting on this ^
It apparently took three tries for SIGINT to get through.

[apostasie]

@Shubhranshu153 suggesting we disable the test for docker for now.
Do you want to PR this ^?

[apostasie]

@AkihiroSuda + third-party + bug + area CI?

I wish regular contributors (eg: Subhransu, myself, maybe some others?) could set labels on tickets so that we can stop bothering you to slap labels on things.
Is there a GitHub setting that would allow that or is this something you prefer to keep strict control over for now?

[Shubhranshu153]

SIGINTs is a graceful call so a single call may be sleep is handling differently?

[apostasie]

SIGINTs is a graceful call so a single call may be sleep is handling differently?

In the test, what we actually do is:

	while true; do
		printf "waiting...\n"
		sleep 0.5
	done

Not sure indeed how this works internally...

[Shubhranshu153]

--sig-proxy false i would think it should detach the container without stopping it may be the test is wrong and the way nerdctl handles it is not correct so it passes in nerdctl.

[apostasie]

Maybe.
It used to work in docker as well though.

sig-proxy false indeed means the bash trap should never see the signal. But then shouldn't the container be stopped anyhow?

[Shubhranshu153]

container can be stopped using docker container stop or something like that.
This test should only be done in STDIN attached mode i guess where we have a distinction between sig-proxy

[Shubhranshu153]

for a background process i dont know SIGPROXY=true or SIGPROXY=false has any difference

[apostasie]

Both background and foregrounded + interactive do work for nerdctl it seems (<- not an endorsement that this is right).