Bug: in-memory v.connections value lost when it crashes and restarted

[ghostplant]

When multi-containers share volume of same name, crash of plugin driver will be unable to correctly handle their umount once it is restarted.

[vieux]

@ghostplant I'll look into it, any hints on why the plugin crashed ?

[vieux]

@ghostplant I believe it's fixed by fe8b54f

[ghostplant]

@vieux
No, Volume.Umount is solved but it causes another bug from Volume.Remove:
docker volume rm sshvolume
Error response from daemon: Error while removing volume sshvolume: remove sshvolume: VolumeDriver.Remove: volume sshvolume is currently used by a container

Actually, no containers exist.

[ghostplant]

I think API should contain req.{Name, ID, nameIsShared:bool}. Thus it will simplify implementation for most of 3rd-party plugin driver and no longer has any kinds of counter bugs.

As you can see, over 30% code is extended to focus on how to persist stateful data while it still not fixes this issue. All mount-based plugin driver implementations also need such complex handling.

[ghostplant]

Case 1:
$ docker volume plugin install vieux/sshfs
$ docker volume create -d vieux/sshfs -o sshcmd=a@0.0.0.0:/tmp -o password=123 sshvolume
$ docker run -it --name n1 -d -v sshvolume:/ssh alpine:3.4 sleep 1h
$ killall docker-volume-sshfs
$ docker volume rm sshvolume
Error response from daemon: Unable to remove volume, volume still in use: remove sshvolume: volume is in use - [2f56f17a3bba9a2f1ed72ae3db495e2417b1219ec17a4b324445e3c481551209]
$ docker kill n1 && docker rm n1
$ docker volume rm sshvolume
Error response from daemon: Error while removing volume sshvolume: remove sshvolume: VolumeDriver.Remove: remove /mnt/volumes/59faf0aa8522fecd92fd6a1af9c91230: device or resource busy

[ghostplant]

Case 2:
$ docker volume plugin install vieux/sshfs
$ docker volume create -d vieux/sshfs -o sshcmd=a@0.0.0.0:/tmp -o password=123 sshvolume
$ docker run -it --name n1 -d -v sshvolume:/ssh alpine:3.4 sleep 1h
$ docker run -it --name n2 -d -v sshvolume:/ssh alpine:3.4 sleep 1h
$ killall docker-volume-sshfs
$ docker kill n1 && docker rm n1
$ docker kill n2 && docker rm n2
$ docker volume rm sshvolume
Error response from daemon: Error while removing volume sshvolume: remove sshvolume: VolumeDriver.Remove: volume sshvolume is currently used by a container

[vieux]

@ghostplant the old usecase where you have docker-volume-sshfs isn't supported anymore.
Can you try with 1.13 and the new style plugins ?

[ghostplant]

@vieux
I certainly did the test on Docker 1.13. See both Case 1 & Case 2 contain docker volume plugin xxx which is unsupported before 1.13.

Note that there are many ways to crash a plugin driver which runs inside a docker container.
Externally executing kill docker-volume-sshfs is just one approach to break docker-volume-sshfs inside a container down.

[vieux]

I believe this issues was fixed in 1.0 the state is now saved on disk.