Open stuart-c-moore opened 6 years ago
justyntemme
commented
6 years ago Testing on Debian 9 via digital ocean I was not able to get docker installed with the given scripts. Please let me know if you're scripts are still successful I will test again.
mogthesprog
commented
6 years ago I'm also seeing this with exactly the same installation steps.
I think it may be a race condition in between docker.socket and docker.service but i'm not getting enough logs to confirm this.
See these logs:
Aug 29 21:53:18 dev-jira-84dj systemd[1]: Starting Docker Application Container Engine...
Aug 29 21:53:18 dev-jira-84dj dockerd[29285]: Failed to load listeners: no sockets found via socket activation: make sure the service was started by systemd
Aug 29 21:53:18 dev-jira-84dj systemd[1]: docker.service: Main process exited, code=exited, status=1/FAILURE
Aug 29 21:53:18 dev-jira-84dj systemd[1]: Failed to start Docker Application Container Engine.
Aug 29 21:53:18 dev-jira-84dj systemd[1]: docker.service: Unit entered failed state.I can't seem to get any more verbosity out of it despite googling.
lexelby
commented
6 years ago It's definitely a race condition, but I'm seeing it consistently. @mogthesprog I think you're on the right track there. Check out this snippet from /var/lib/dpkg/info/docker-ce.postinst:
# Automatically added by dh_systemd_enable
# This will only remove masks created by d-s-h on package removal.
deb-systemd-helper unmask docker.service >/dev/null || true
# was-enabled defaults to true, so new installations run enable.
if deb-systemd-helper --quiet was-enabled docker.service; then
# Enables the unit on first installation, creates new
# symlinks on upgrades if the unit file has changed.
deb-systemd-helper enable docker.service >/dev/null || true
else
# Update the statefile to add new symlinks (if any), which need to be
# cleaned up on purge. Also remove old symlinks.
deb-systemd-helper update-state docker.service >/dev/null || true
fi
# End automatically added section
# Automatically added by dh_systemd_enable
# This will only remove masks created by d-s-h on package removal.
deb-systemd-helper unmask docker.socket >/dev/null || true
# was-enabled defaults to true, so new installations run enable.
if deb-systemd-helper --quiet was-enabled docker.socket; then
# Enables the unit on first installation, creates new
# symlinks on upgrades if the unit file has changed.
deb-systemd-helper enable docker.socket >/dev/null || true
else
# Update the statefile to add new symlinks (if any), which need to be
# cleaned up on purge. Also remove old symlinks.
deb-systemd-helper update-state docker.socket >/dev/null || true
fi
# End automatically added sectiondocker.service declares a dependency on docker.socket, but apparently that's ignored if docker.socket isn't enabled. By running journalctl (to view the entire system log with all services interleaved), I can see that docker.service does indeed try to start up before docker.socket. It fails, and by then docker.socket has started, and then docker.service is able to start successfully on the next try.
As @migs pointed out, the apt-get install fails which means that whatever script is invoking it also fails. In my case, that's Chef, and I get a failed chef run. :(
lexelby
commented
6 years ago It seems almost as if this is a bug in systemd (for allowing docker.service to start even though it Requires something that's not enabled) or dh_systemd_enable (for ignoring dependencies and ordering).
lexelby
commented
6 years ago Workaround: "mask" docker.service before installing docker-ce and unmask it after. Masking is described in the third bullet point here.
In chef, it looks like this:
systemd_unit 'docker.service' do
action :mask
end
docker_installation_package 'default' do
action :create
end
systemd_unit 'docker.service' do
action :unmask
end
docker_service_manager_systemd 'default' do
action [:start]
end
njam
commented
2 years ago I experience the same with Debian 10 on GCP (the ticket is about Debian 9).
njam
commented
2 years ago Thanks @lexelby for the workaround using "mask". Works for me.
To install docker on Debian 10 on GCP I use:
systemctl mask docker.service
apt-get install -qy docker-ce
systemctl unmask docker.service
systemctl enable --now docker.service
Expected behavior
Docker starts immediately after installation
Actual behavior
Docker fails immediately after installation on Debian 9 on Google Cloud Platform. Docker is then restarted by systemd, and then works fine.
Unfortunately, this means that if the installation of the
docker-cepackage is run with a script containingset -e, that script will fail.Steps to reproduce the behavior
The virtual machine is destroyed and rebuilt to provide clean results.
After the startup-script is run (contents shown below), the following is seen in
sudo journalctl -u docker.service:This is the output of
/var/log/syslog:Output of
docker version:Output of
docker info:The full contents of the startup script used for the VM: