Cluster Manager

A nice script written by Marc Hulsman to submit and monitor jobs in Life Science Grid.

PREREQUISITES

1. Add LFC_HOME and LFC_HOST to your environment. e.g. Add to you .bashrc something like:

   export LFC_HOME='/grid/lsgrid/marc/'
   export LFC_HOST=lfc.grid.sara.nl

2. Run startGridSession lsgrid. This script will:

• Generate a proxy certificate,
• Store it in the Myproxy server
• Delegate it to the WMS with your user name as the delegation ID (DID).

INSTALL

1. install 'cluster_manager' using Enhance (see github.com/mhulsman).

Uploading Enhance environment to grid

If grid is going to be used, use <enhance_dir>/upload file to upload environment to grid (note: this should be done everytime you update enhance environment):

./upload grid:/<choose_a_name>.dist (e.g. ipengine.dist)

RUN just python

1. Adapt load_env_worker.sh

2. If you need extra files, put them in the work directory (<enhance_dir>/sys_enhance/work) and update environment (last paragraph)

3. If you use ipengine_loadenv.jdl, change XXNJOBXX to the number of jobs you want to run

4. To submit jobs use:

   glite-wms-job-submit -d <delegation_id> -o <jobid_file>  ipengine_loadenv.jdl

RUN ipython parallel

1. First, make sure ports are open on that were given during install.

2. Then, go to your python directory from which you run your project:

   • cd <your_project>
   • <environment_dir>/cluster.py
   • start controllers and engines from interface

3. ipython

   • start tasks e.g.:

       from IPython.kernel import client
       
       #METHOD 1
       mec = client.MultiEngineClient()
       mec['b'] = 2  #sets b in all clients
       mec.execute("a = b * 2")
       print mec['a']  #prints for each client: 4
       
       #METHOD 2
       tc = client.TaskClient()
       strtask = """a = b * 2"""
       b = 1 
       task = client.StringTask(strtask, ('a',), {'b':2})
       task_id = tc.run(task)
       tres = tc.get_task_result(task_id,block=True)
       print tres['a']  #prints 2
     

CLUSTER STORAGE

For sending large files to/from engines, make use of submit / receive functions in cluster_storage.py:

• Somewhere on grid, do:

   id = submit(your_object)
   #or, with fixed id:
   id = submit(your_object,"your_id")

• Somewhere else on grid, do:

   object = receive(id)    
   
   #clean up
   destroy(id)