ros2 daemon hangs on arch with docker

[Sonic803]

Bug report

Required Info:

• Operating System:
  • Arch Linux 6.1.38-1-lts
• Installation type:
  • Docker
• Version or commit hash:
  • humble

Steps to reproduce issue

Just run

docker run --rm -it ros:humble /bin/bash -c "ros2 topic list"

Expected behavior

Output something like

/parameter_events
/rosout

Actual behavior

Gets stuck

Additional information

I tried the same thing on Ubuntu and there it works.

Adding --no-daemon makes it work on arch

docker run --rm -it ros:humble /bin/bash -c "ros2 topic list --no-daemon"

On arch I have
docker 1:25.0.3-1

[tfoote]

I can confirm that it works on my debianish machine with docker Docker version 20.10.21, build baeda1f

I see that docker 25 is much newer, and potentially there's changes in docker's default behavior or how Arch's networking is setup. Arch isn't a standard development platform so we're going to need to rely on more info and debugging from you or other Arch users.

[akssri-sony]

Can confirm,

• It doesn't work humble-onwards on Archlinux (tested with iron/rolling); but foxy is fine.

• Using cyclonedds doesn't fix the issue.

• It works fine on Ubuntu-22.04

• On a working older version of osrf/ros:humble-desktop; upgrading ros-humble-ros2cli ( (0.18.7-1jammy.20231117.193530 -> 0.18.9-1jammy.20240217.070501) causes the behavior to reappear.

[akssri-sony]

• Breaks at make handles not inheritable to prevent from blocking durning tab-completion ros2cli#852 (61f98ce6103e762c248e1a08fba4f5a2033eb06c)

[akssri-sony]

Temporary fix; call,

ulimit -n 1024

in the container before running ros2 cli stuff.

[gavanderhoorn]

Reminds me of Odd Docker roscore/roslaunch delays with Linux kernel 6.1 on ROS Answers.

[clalancette]

Reminds me of Odd Docker roscore/roslaunch delays with Linux kernel 6.1 on ROS Answers.

Yep, very good point. That also leads to https://stackoverflow.com/questions/899038/getting-the-highest-allocated-file-descriptor/918469#918469 and https://github.com/ruby/ruby/blob/v3_0_2/io.c#L6557-L6597 , which are interesting.

We could change this to try a (Linux-specific) hint, like that Ruby code (but obviously in Python). Basically see if we can get the FDSize from /proc, and if so, use that as the hint on where to stop. Otherwise, we would still fall back to the soft-limit. This won't work for non-Linux, and won't work for systems where /proc isn't mounted, but those should be the minority of systems.

@akssri-sony If you'd like to give that a shot, I'd be happy to review it.

[akssri-sony]

Thanks for the pointers @clalancette ! I've created a PR here,
ros2/ros2cli#888