6.12.0-rc7 docker autostart not working

There is some kind of race condition here.

 

If I add a sleep 10 into the rc.docker script before start_containers then when I disable and re-enable docker then it works.

 

It is failing is_docker_running due to lack of /var/run/docker.sock

This appears <2s after start_containers complains that docker isn't running from my testing.

I setup an inotify on /var/run and ran a tail of /var/log/syslog

 

Jun 11 23:06:47 unraid emhttpd: shcmd (362): /etc/rc.d/rc.docker start
Jun 11 23:06:47 unraid root: starting dockerd ...
/var/run/ CREATE dockerd.pid
Jun 11 23:06:47 unraid avahi-daemon[14531]: Server startup complete. Host name is unraid.local. Local service cookie is 39991689.
Jun 11 23:06:48 unraid avahi-daemon[14531]: Service "unraid" (/services/ssh.service) successfully established.
Jun 11 23:06:48 unraid avahi-daemon[14531]: Service "unraid" (/services/smb.service) successfully established.
Jun 11 23:06:48 unraid avahi-daemon[14531]: Service "unraid" (/services/sftp-ssh.service) successfully established.
Jun 11 23:07:02 unraid root: Docker not running, can't start containers
/var/run/ CREATE docker.sock

 

You can see that the CREATE of docker.sock is after start_containers complains docker isn't running.

 

Looking through rc.docker this puzzles me as wait_daemon is called in start_network so I'm at a bit of a loss

So wait_daemon() currently waits up to 15 seconds. If you remove your "sleep 10" and bump wait_daemon() up to 30 does that solve the problem?

17 hours ago, ljm42 said:

So wait_daemon() currently waits up to 15 seconds. If you remove your "sleep 10" and bump wait_daemon() up to 30 does that solve the problem?

 

Yes that does indeed fix the issue, so I got digging.

 

I have DOCKER_OPTS set in /boot/config/docker.cfg as per

 

 

Removing this and problem is solved, and as I no longer need remote access I'm good, however this will bite others potentially.

 

It's the -H tcp://0.0.0.0:2375 that causes the delay for me, somewhere between 15 and 30s.

 

Thanks

 

Thanks, we modified wait_daemon() to wait up to 30 seconds in the next release.

 

I'm glad you were able to figure out the source of the delay. Putting an actual IP address rather than 0.0.0.0 may allow this customization to work.