start before docker.service

[awyl]

AzureFile volume driver should start before docker.

The reason is when using docker run --restart with the volume driver, auto restart will failed because the driver is not available until docker started.

Changing the .service file, as following works:
After=nfs-utils.service docker.service to
Before=nfs-utils.service docker.service

[ahmetb]

@awyl thanks for reporting this. In my tests with docker 1.10 or above, I noticed no difference between before/after and I kind of remember asking about this on #docker-dev IRC channel. Folks did not seem to have a consensus, because normally Docker retries any plugin communication for 30 seconds (and during that time, containers with --restart=always should be blocked from starting until the plugin is reachable).

Do you have any logs about these containers failed to start? Also what version of docker-engine are you running?

In any case if we were to take the change, we should probably keep After=nfs-utils.service and separate Before=docker.service.

[awyl]

Sorry I am not very familiar with Linux and Docker. Where can I look for the logs?

I am using Docker on Ubuntu image on Azure. Ubuntu 15.10 and Docker is 1.11.1.

Where I saw the "could not start" message was from "systemctl status docker.service". It said the volume cannot be found.

I am sure that after boot the container has not been started after a long time while I was looking for reasons.

[ahmetb]

@awyl can you try sudo journalctl -u docker.service?

[awyl]

I cannot get journalctl to show the full line. So I try again with systemctl:

May 06 00:54:06 docker01 docker[666]: .time="2016-05-06T00:54:06.607288800Z" level=warning msg="Unable to locate plugin: azurefile, retrying in 1s"
May 06 00:54:07 docker01 docker[666]: time="2016-05-06T00:54:07.607475100Z" level=warning msg="Unable to locate plugin: azurefile, retrying in 2s"
May 06 00:54:09 docker01 docker[666]: time="2016-05-06T00:54:09.612964300Z" level=warning msg="Unable to locate plugin: azurefile, retrying in 4s"
May 06 00:54:13 docker01 docker[666]: time="2016-05-06T00:54:13.613335400Z" level=warning msg="Unable to locate plugin: azurefile, retrying in 8s"
May 06 00:54:21 docker01 docker[666]: time="2016-05-06T00:54:21.905119000Z" level=error msg="Failed to start container 21da2ff86d0576708eec1be475d5a467f9357fc228edbc0913084b48f6263f15: get servernginx: Error looking up volume plugin azurefile: plugin not found"
May 06 00:54:21 docker01 docker[666]: time="2016-05-06T00:54:21.905581600Z" level=info msg="Loading containers: done."
May 06 00:54:21 docker01 docker[666]: time="2016-05-06T00:54:21.905792200Z" level=info msg="Daemon has completed initialization"
May 06 00:54:21 docker01 docker[666]: time="2016-05-06T00:54:21.905987900Z" level=info msg="Docker daemon" commit=5604cbe graphdriver=aufs version=1.11.1
May 06 00:54:21 docker01 systemd[1]: Started Docker Application Container Engine.
May 06 00:54:21 docker01 docker[666]: time="2016-05-06T00:54:21.931319900Z" level=info msg="API listen on /var/run/docker.sock"

So it did retry and fail.

[ahmetb]

@awyl oh interesting, it appears like it is not starting docker engine until loading all its previous containers. This behavior might have been changed lately. We should fix the .service unit file. Thanks for reporting!

[ahmetb]

@awyl has doing

 After=nfs-utils.service
 Before=docker.service

fixed the problem for you?

[awyl]

Yes it does. This is what I currently have.

[ahmetb]

@awyl thanks for confirming. I'm submitting a fix.