Coclobas is a scheduler for HPC-like jobs accessible through HTTP.

It can be setup with three kinds of configurations:

• Using Kubernetes and the Google Container Engine, i.e. using a Kubernetes “eleastic” cluster setup with gcloud and submitting jobs as Kubernetes “pods”.
• Using the server's machine as a one-node cluster and submitting jobs as docker containers given a maximal number of jobs.
• (Experimental) using an AWS-Batch job queue with Docker containers.

Coclobas provides logging facilities (e.g. maintaining logs long after Kubernetes disposes of them).

Finally, it makes it easy to submit arbitrary scripts to be run in any Docker container, which makes it easier than using raw Kubernetes or Docker to submit arbitrarily complex jobs.

If the ketrew package is present, Coclobas comes with a Ketrew plugin (loadable dynamically) and a build of the Ketrew binary with the plugin already loaded in (for deployment convenience).

Coclobas 0.0.1 is in opam (supports GKE/Kubernetes and local-docker modes).

You can just use Opam to get things going quickly:

opam pin add coclobas https://github.com/hammerlab/coclobas.git

Or you may clone this repo and:

make

In GKE/Kubernetes mode, gcloud and kubectl must be installed (and authenticated) with the Coclobas server.

In Local/Docker mode, docker must be present (and accessible to the Coclobas server's user).

In AWS-Batch mode, the aws command line application must be present (and it must be recent enough to have aws batch enabled).

You first need to create a “root” directory, see:

coclobas configure --help

Example 1: GKE/Kubernetes mode:

coclobas config --root $root \
         --cluster-kind gke \
         --gke-cluster-name "my-coclotest-cluster" \
         --gcloud-zone "us-east1-c" \
         --max-nodes 5

Example 2: Local/Docker mode:

coclobas config --root $root \
         --cluster-kind local-doker \
         --max-nodes 5

Example 3: AWS-Batch mode:

coclobas config --root $root \
    --cluster-kind aws-batch \
    --aws-queue awsuser-jq01 \
    --min-sleep 4 \
    --max-update-errors 4 \
    --max-nodes 4

Now that Coclobas has a configuration (this information is stored in the --root, so you can have many possible Coclobas servers configured, all in different directories), you'll want to start it:

coclobas start-server --root $root --port 8999

You can use this curl http://127.0.0.1:8999/status to see if Coclobas is ready to go (if it says Initializing usually it means that it is setting up a GKE-cluster which takes a few minutes).

The most common way of submitting jobs is through the Ketrew plugin, see examples in src/test/workflow_test.ml.

In GKE/Kubernetes mode:

workflow_node without_product ~name:"Coclobas that uses the GKE/Kubernetes"
  ~make:(
    Coclobas_ketrew_backend.Plugin.kubernetes_program
      ~base_url:"http://127.0.0.1:8999/"
      ~image:"ubuntu"
      ~volume_mounts:[
        `Nfs (
          Coclobas.Kube_job.Specification.Nfs_mount.make
            ~host:"nfs-server.example.com"
            ~path:"/path/inside/nfs-server"
            ~point:"/mount/point/" ())
      ]
      Program.(
        chain [
          shf "hostname";
          shf "du -sh /mount/point";
          shf "sleep 60";
        ]
      )
  )

In Local/Docker mode:

workflow_node without_product ~name:"Coclobas test of local-docker jobs"
  ~make:(
    Coclobas_ketrew_backend.Plugin.local_docker_program
      ~base_url:"http://127.0.0.1:8999/"
      ~image:"ubuntu"
      ~volume_mounts:[
        `Local ("/usr/bin", "/hostusrbin")
      ]
      Program.(
        exec ["find"; "/hostusrbin"]
      )
  )

All the “job inspection” features are available through the Ketrew UIs as “backend queries.”

One can also directly query the server through the command line, see coclobas client --help, for example:

coclobas client --server http://127.0.0.1:8999 list

If you have any questions, you may submit an issue, or join the authors on the public “Slack” channel of the Hammer Lab:

It's Apache 2.0.