Lustre的Docker Swarm Volumes？

Question

I have a docker swarm setup for HDFS data nodes as follows (truncated):

  datanode:
    image: bde2020/hadoop-datanode:2.0.0-hadoop2.7.4-java8
    volumes:
      - datanode:/hadoop/dfs/data
    deploy:
      mode: global
      placement:
        constraints: [ node.role == worker ]
  volumes:
    datanode:

Each datanode container has an associated datanode volume that are currently being stored in /var/lib/docker/volumes (local to each node). I have access to a much larger (and faster) lustre file system and am seeking to understand how one could store the volumes on the Lustre file system.

One potential is to symlink /var/lib/docker/volumes to /lustre/nodeN (where N is the physical node number), but it seems that the namenode would not be able to handle a stack restart because the data would (potentially) be moved to another node.

Does a docker drive exist to support docker volumes on lustre file systems in a swarm? If not, how are teams managing docker volume data persistence for systems like HDFS?

Answer 1

I'm afraid not if I haven't misunderstood you, because driver: lustre or similar doesn't exist for docker configuration in the same way it exist for nfs , for example:

volumes:
   db-data:
      driver: local
      driver_opts:
        type: <here is where you could try to put lustre, but it doesn't exist>
        o: addr=$IPADDR,rw
        device: ":/hadoop/dfs/data"

So, if you regard the following schema, try with lustre-graph-driver

https://github.com/bacaldwell/lustre-graph-driver/blob/master/lustre-graph-driver.jpg

More information in lustre-graph-driver HOWTO