Fail fast on errors during mount

[Vad1mo]

We have a hyper-converged system that runs Gluster and Swarm on the same node. so we are mounting the volumes like this "voluri": "localhost:gv-portainer"

When there is an error in gluster cluster. eg. Gluster servie isn't stated, then it is still possible to spinup a container that mounts a gluster volume. It is impossible to know if a volume was really mounted or not because in both cases the application runs. One only see in the application itself that the data is missing.

It would be a good idea to fail fast if a volume can't be mounted, so the container never comes up and can be addressed accordingly.

I am not sure if that is related to #18.

[sapk]

How do you start container? Are you sure that it is a gluster volume that is started ? One problem that I encounter myself (still in integration tests), the script fail to create the volume but after docker create automatically a local volume (that is empty) with the name of the volume.

Now that I use glusterfs cli directly, I could detect when background process failed but after volume is mount docker drivers have no possibility to inform docker host that the mount point is unavailable.

[Vad1mo]

Its a docker stack ( with compose) and the plugin is also an container.

When I inspect the container I looks like it mounted the gluster volume. However when I inspect the volume I see that the mountpoint doesn't exist.

docker volume inspect portainer_portainer-data 
[
    {
        "CreatedAt": "0001-01-01T00:00:00Z",
        "Driver": "glusterfs:latest",
        "Labels": {
            "com.docker.stack.namespace": "portainer"
        },
        "Mountpoint": "/var/lib/docker-volumes/gluster/portainer_portainer-data",
        "Name": "portainer_portainer-data",
        "Options": {
            "voluri": "localhost:gv-portainer"
        },
        "Scope": "local"
    }
]

Docker Container:

                "Type": "volume",
                "Name": "portainer_portainer-data",
                "Source": "/var/lib/docker/plugins/eb54b605aac01e577ebad9e73439f0a228a1bbcc2789a7f8dc4a69149fbce971/rootfs/var/lib/docker-volumes/gluster/portainer_portainer-data",
                "Destination": "/data",
                "Driver": "glusterfs:latest",
                "Mode": "",
                "RW": true,
                "Propagation": ""

[sapk]

As for #18 it is mostly to try to resolve hostname of server (and bricks) before creation and various little pre-check to limit error related to configuration later at mount.

[sapk]

Otherwise, it seems to be using the plugin (I suppose you use a custom alias glusterfs). I will look at it near https://github.com/sapk/docker-volume-gluster/blob/82079baff00622cdfd637ddb7e8a7a344497900e/gluster/driver/driver.go#L241 to catch more errors. (Maybe keep process in foreground and attaching it to mount)

[Vad1mo]

Do you need more info to track down the issue. currently we can't use the plugin due to this issue

Config.json from cat /var/lib/docker/plugins/eb54b605aac01e577ebad9e73439f0a228a1bbcc2789a7f8dc4a69149fbce971/config.json

{
  "plugin": {
    "Config": {
      "Args": {
        "Description": "Arguments to be passed to the plugin",
        "Name": "args",
        "Settable": ["value"],
        "Value": []
      },
      "Description": "GlusterFS plugin for Docker",
      "DockerVersion": "17.10.0-ce",
      "Documentation": "https://docs.docker.com/engine/extend/plugins/",
      "Entrypoint": ["/usr/bin/docker-volume-gluster", "daemon"],
      "Env": [{ "Description": "", "Name": "DEBUG", "Settable": ["value"], "Value": "0" }],
      "Interface": { "Socket": "gluster.sock", "Types": ["docker.volumedriver/1.0"] },
      "IpcHost": false,
      "Linux": {
        "AllowAllDevices": false,
        "Capabilities": ["CAP_SYS_ADMIN"],
        "Devices": [{ "Description": "", "Name": "", "Path": "/dev/fuse", "Settable": null }]
      },
      "Mounts": null,
      "Network": { "Type": "host" },
      "PidHost": false,
      "PropagatedMount": "/var/lib/docker-volumes/gluster",
      "User": {},
      "WorkDir": "",
      "rootfs": {
        "diff_ids": ["sha256:8a21bfe4b75043c94d7f9a9201ea73932c3e489bd7fd2881021ceedcec3b19d5"],
        "type": "layers"
      }
    },
    "Enabled": true,
    "Id": "eb54b605aac01e577ebad9e73439f0a228a1bbcc2789a7f8dc4a69149fbce971",
    "Name": "glusterfs:latest",
    "PluginReference": "docker.io/sapk/plugin-gluster:latest",
    "Settings": {
      "Args": [],
      "Devices": [{ "Description": "", "Name": "", "Path": "/dev/fuse", "Settable": null }],
      "Env": ["DEBUG=0"],
      "Mounts": []
    }
  },
  "PropagatedMount":
    "/var/lib/docker/plugins/eb54b605aac01e577ebad9e73439f0a228a1bbcc2789a7f8dc4a69149fbce971/rootfs/var/lib/docker-volumes/gluster",
  "Rootfs": "/var/lib/docker/plugins/eb54b605aac01e577ebad9e73439f0a228a1bbcc2789a7f8dc4a69149fbce971/rootfs",
  "Config": "sha256:a1fbb7a93f194f604d1532d75913e4b89aa8c75c8dd4896baace96a41363ec6b",
  "Blobsums": ["sha256:c367876793b34af008c4e1f6d19ed7a584352092021e9ef03259c97bf94566f0"],
  "SwarmServiceID": ""
}

ls -la /var/lib/docker-volumes/gluster/portainer_portainer-data/ fails as this directory doesn't exist

[sapk]

I try to setup tests in travis to deliver new image based on glusterfs cli. That could maybe fix the error since command could failed directly but I will try to improve handling cli by keep process monitored by the driver and logging output directly via the plugin.

[sapk]

A new version should be released soon. It use gluster cli (plugin wasn't updated without a tag) that may return more error case at start-up when mount command would have detach it-self. I will definitively improve process handling but that could already fix your problem.

[Vad1mo]

Alright, I try that out later today and let you know about the outcome.