Vagrant

Relying on Vagrant for deploying the HPCCloud infrastructure’s is great but we can’t assuming that it will always work and additional interaction with the infrastructure may be required. Therefore, that documentation aims to provide various how to with Vagrant and HPCCloud deployment.

Setting-up HPCCloud for development

The very first time you may want to run the following commands.

$ mkdir HPCCloud
$ cd HPCCloud

$ git clone https://github.com/Kitware/HPCCloud.git hpccloud
$ git clone https://github.com/Kitware/HPCCloud-deploy.git
$ git clone https://github.com/Kitware/cumulus.git
$ git clone https://github.com/Kitware/girder.git

$ virtualenv python
$ source ./python/bin/activate
$ pip install -r hpccloud/requirements.txt
$ pip install -r HPCCloud-deploy/requirements.txt

$ cd HPCCloud-deploy
$ DEVELOPMENT=1 vagrant up

Then you can simply start again the Virtual Machine with the following commands.

$ cd HPCCloud

$ source ./python/bin/activate

$ cd HPCCloud-deploy
$ DEVELOPMENT=1 vagrant up

In order to fully stop the Virtual Machine, you can run the following commands.

$ cd HPCCloud

$ source ./python/bin/activate

$ cd HPCCloud-deploy
$ vagrant halt

SSH to Virtual Machine

$ cd HPCCloud

$ source ./python/bin/activate

$ cd HPCCloud-deploy
$ vagrant ssh
$ sudo -iu hpccloud
$ cd /opt/hpccloud

Once logged to the VM, services can be stoped and restarted with the following commands

sudo service girder [stop/start/restart]
sudo service celeryd [stop/start/restart]

Fixing invalid Hostname

When HPCCloud run jobs on a remote cluster, that cluster needs to contact back the HPCCloud server and therefore the proper information needs to be given to those cluster.

This can be adjusted in the following file (replace: HOSTNAME_OR_VALID_IP by the proper value):

/opt/hpccloud/cumulus/cumulus/conf/config.json

{
  "girder": {
    "baseUrl": "http://HOSTNAME_OR_VALID_IP:8080/api/v1",
    "user": "cumulus",
    "group": "cumulus"
  },
  "moabReader": {
    "pluginPath": "/opt/hpccloud/moabreader/build/libvtkCmbMoabReaderPlugin.so"
  },
    "ssh": {
    "keyStore": "/opt/hpccloud/keys"
  },

  "amis": {
  },
  "taskflow": {
    "path": ["/opt/hpccloud/hpccloud/server/taskflows"]
  }
}

Same for girder so http://localhost:8080 properly reach girder

/opt/hpccloud/girder/girder/conf/girder.local.cfg

[global]
server.socket_host = "0.0.0.0"
server.socket_port = 8080
server.thread_pool = 100

[...]

Running Celery by hand

This can be usefull to follow any runtime error live or if the taskflow did properly load.

$ /usr/bin/python -m celery worker --app=cumulus --loglevel=INFO --config=cumulus.celery.commandconfig --concurrency=6 --logfile=/var/log/celery/command.log --pidfile=/var/run/celery/command.pid --hostname=command@vagrant-ubuntu-trusty-64