Autofs and volume_mount

Nomad
1

Hi,

I’m hoping to use AutoFS to manage the mounts in /smb. On the host it works fine, but I’m having an issue from inside a running container.

In my client config, I have:

"host_volume": {
      "autofs_smb": {
        "path": "/smb",
        "read_only": false
      }

In my group config I have:

    volume "autofs_smb" {
      type      = "host"
      read_only = false
      source    = "autofs_smb"
    }

and in the task I have:

      volume_mount {
        volume      = "autofs_smb"
        destination = "/smb"
      }

I’m using the docker driver.

The symptoms are:

  1. ls /smb/servername in the container, it mounts and I can see the share folders.
  2. ls /smb/servername/sharename in the container returns “No such file or directory”.
  3. ls /smb/servername/sharename on the host returns files and folders
  4. Restart the Nomad task and ls /smb/servername/sharename in the container returns files and folders

It looks like AutoFS lazy-mounts the shares, and the container isn’t picking up that change. i.e. ls /smb/servername returns a list of local files that autofs just created. ls /smb/servername/sharename then mounts that share to that location, and that change isn’t “seen” by the container.

I’ve seen references to switching to Docker bind mounts with rshared - but if I can, I’d like to stick to the Nomad volumes for the increased security they give me.

Anyone else come across this? any hints?

Many thanks,

Geoff

EDIT: I gave up (this post was at the end of several hours of trying) and switched to bind mounts in the Job->Group->task->config section:

        mount {
          type = "bind"
          target = "/net"
          source = "/net"
          readonly = false
          bind_options {
            propagation = "rshared"
          }
        }

and set driver.docker.volumes.enabled to true on the clients.

I would have preferred to keep the Nomad volumes, as they provide better control. But this works, and I’m able to browse the autofs mounts with no issues.

2

Me replying to myself because I had a weird situation that I wanted to share, in case it helps someone else.

If I restarted autofs, the containers would lose the connections and would not re-establish them.

Essentially, with the bind mounts set for the autofs root folders, when autofs restarted, it recreated those locations, breaking the link.

The fix is easy and obvious - bind mount one level higher.

So I created a /network folder, configured autofs to mount inside that folder, and bind-mounted /network instead. As a bonus, I have one, not two, mounts in the config now - and it survives a restart of autofs :slight_smile: