---
layout: default
navsection: api
navmenu: Concepts
title: Crunch scripts

...

h2. Crunch scripts

A crunch script is responsible for completing a single JobTask. In doing so, it will:

* (optionally) read some input from Keep
* (optionally) store some output in Keep
* (optionally) create some new JobTasks and add them to the current Job
* (optionally) update the current JobTask record with the "output" attribute set to a Keep locator or a fragment of a manifest
* update the current JobTask record with the "success" attribute set to True

A task's context is provided in environment variables.

table(table table-bordered table-condensed).
|Environment variable|Description|
|@JOB_UUID@|UUID of the current "Job":methods/jobs.html|
|@TASK_UUID@|UUID of the current "JobTask":methods/job_tasks.html|
|@ARVADOS_API_HOST@|Hostname and port number of API server|
|@ARVADOS_API_TOKEN@|Authentication token to use with API calls made by the current task|

The crunch script typically uses the Python SDK (or another suitable client library / SDK) to connect to the Arvados service and retrieve the rest of the details about the current job and task.

The Python SDK has some shortcuts for common operations.

In general, a crunch script can access information about the current job and task like this:

<pre>
import arvados
import os

job = arvados.api().jobs().get(uuid=os.environ['JOB_UUID']).execute()
$sys.stderr.write("script_parameters['foo'] == %s"
                  % job['script_parameters']['foo'])

task = arvados.api().job_tasks().get(uuid=os.environ['TASK_UUID']).execute()
$sys.stderr.write("current task sequence number is %d"
                  % task['sequence'])
</pre>