Can several nodes access mounted docker containers

问题:

I have a VM running an app that uses about 8 Docker containers.

If I move /var/lib/docker to /mnt/containers/, where /mnt/containers is mounted via glusterfs to a larger system, I start getting errors like this:

kernel@192.168.68.14: Jun 17 16:05:10 stackato-ft9y kernel: [ 2174.535122] aufs au_xino_set:1176:docker[7572]: I/O Error, failed creating xino(-27).

kernel@192.168.68.14: Jun 17 16:05:10 stackato-ft9y kernel: [ 2174.538613] aufs au_xino_set:1176:docker[7572]: I/O Error, failed creating xino(-27).

dockerd@192.168.68.14: [error] mount.go:11 [warning]: couldn't run auplink before unmount: exit status 22^M

dockerd@192.168.68.14: file too large

dockerd@192.168.68.14: [d954f89b] -job create(fence_app_staging_fibo_1a992a98_id-3dd68) = ERR (1)^M

dockerd@192.168.68.14: [error] server.go:1025 Error: file too large^M

dockerd@192.168.68.14: [error] server.go:90 HTTP Error: statusCode=500 file too large

I don't see these errors when running out of /var/lib/docker, or even if I move the contents of /var/lib/docker to a different local directory.

Two of us have independently stumbled on http://osdir.com/ml/linux.file-systems.aufs.user/2008-08/msg00016.html , but that doesn't quite look right. So I'm here hoping to get the attention of the resident docker/aufs/glusterfs experts.

回答1:

Looks like conflict between aufs and glusterfs. Have you tried to use btrfs or devicemapper instead? (start the daemon with -s btrfs or -s devicemapper)

回答2:

I've just been told that networked volumes are allowed only if they look like regular ext volumes to the host, which Gluster doesn't support.