2.3. Part 3: Remote Job Submission

Next, we take the previous example and modify it, so that our job is executed on a remote machine instead of localhost. This examples shows one of the most important capabilities of SAGA: abstracting system heterogeneity. We can use the same code we have used to run a job via ‘fork’ with minimal modifications to run a job on a different resource, e.g., via ‘ssh’ on another remote system or via ‘pbs’ or ‘sge’ on a remote cluster.

2.3.1. Prerequisites

This example assumes that you have SSH access to a remote resource, either a single host or an HPC cluster.

The example also assumes that you have a working public/private SSH key-pair and that you can log-in to your remote resource of choice using those keys, i.e., your public key is in the ~/.ssh/authorized_hosts file on the remote machine. If you are not sure how this works, you might want to read about SSH and GSISSH first.

2.3.2. Hands-On: Remote Job Submission

Copy the code from the previous example to a new file saga_example_remote.py. Add a saga.Context and saga.Session right before the job.Service object initialization. Sessions and Contexts describe your SSH identity on the remote machine:

ctx = saga.Context("ssh")
ctx.user_id = "oweidner"

session = saga.Session()
session.add_context(ctx)

To change the execution host for the job, change the URL in the job.Service constructor. If you want to use a remote SSH host, use an ssh:// URL. Note that the session is passed as an additional parameter to the Service constructor:

js = saga.job.Service("ssh://remote.host.net", session=session)

Alternatively, if you have access to a PBS cluster, use a pbs+ssh://... URL:

js = saga.job.Service("pbs+ssh://remote.hpchost.net", session=session)

There are more URL options. Have a look at the chapter_adaptors section for a complete list. If you submitting your job to a PBS cluster (pbs+ssh://), you will probably also have to make some modifications to your job.Description. Depending on the configuration of your cluster, you might have to put in the name of the queue you want to use or the allocation or project name that should be credited:

jd = saga.job.Description()

jd.environment     = {'MYOUTPUT':'"Hello from SAGA"'}
jd.executable      = '/bin/echo'
jd.arguments       = ['$MYOUTPUT']
jd.output          = "mysagajob.stdout"
jd.error           = "mysagajob.stderr"

jd.queue           = "short" # Using a specific queue
jd.project         = "TG-XYZABCX" # Example for an XSEDE/TeraGrid allocation

2.3.2.1. Run the Code

Save the file and execute it (make sure your virtualenv is activated):

python saga_example_remote.py

The output should look something like this:

Job ID    : None
Job State : New

...starting job...

Job ID    : [ssh://gw68.quarry.iu.teragrid.org]-[18533]
Job State : Done

...waiting for job...

Job State : Done
Exitcode  : 0

Values marked as ‘None’ could not be fetched from the backend, at that point.

2.3.2.2. Check the Output

As opposed to the previous “local” example, you won’t find a mysagajob.stdout file in your working directory. This is because the file has been created on the remote host were your job was executed. In order to check the content, you would have to log-in to the remote machine. We will address this issue in the next example.

2.3.3. Discussion

Besides changing the job.Service URL to trigger a different middleware plug-in, we have introduced another new aspect in this tutorial example: Contexts. Contexts are used to define security / log-in contexts for SAGA objects and are passed to the executing plug-in (e.g., the SSH plug-in).

A context always has a type that matches the executing plug-in. The two most commonly used contexts in SAGA are ssh and gsissh:

# Your ssh identity on the remote machine
ctx = saga.Context("ssh")
ctx.user_id = "oweidner"

A Context can’t be used by itself, but rather has to be added to a saga.Session object. A session can have one or more Contexts. At runtime, RADICAL-SAGA will iterate over all Contexts of a Session to see if any of them can be used to establish a connection.

session = saga.Session()
session.add_context(ctx)

Finally, Sessions are passed as an extra parameter during object creation, otherwise they won’t get considered:

js = saga.job.Service("ssh://remote.host.net", session=ses)

The complete API documentation for Session and Context classes can be found in the Library Reference section of this manual.