pendant

Python 3.6+ library for submitting to AWS Batch interactively.

❯ pip install pendant

Features:

• Submit Batch jobs

Read the documentation at: pendant.readthedocs.io

End-to-end Example

The principle object for deploying jobs to AWS Batch is the Batch job definition. Every Batch job definition has a name, parameters, and some form of optional parameter validation.

>>> from pendant.aws.batch import BatchJob, JobDefinition
>>> from pendant.aws.s3 import S3Uri
>>> from pendant.aws.exception import S3ObjectNotFoundError

>>> class DemoJobDefinition(JobDefinition):
...     def __init__(self, input_object: S3Uri) -> None:
...         self.input_object = input_object
... 
...     @property
...     def name(self) -> str:
...         return 'demo-job'
... 
...     def validate(self) -> None:
...         if not self.input_object.object_exists():
...             raise S3ObjectNotFoundError(f'S3 object does not exist: {self.input_object}')

Let's instantiate the definition at a specific revision and validate it.

>>> definition = DemoJobDefinition(input_object=S3Uri('s3://bucket/object')).at_revision('6')
>>> definition.validate()
None

Validation is also performed when a job definition is wrapped by a BatchJob so the call to .validate() above was redundant. Wrapping a job definition into a Batch job is achieved with the following, but no useful work will happen until the job is submitted.

>>> job = BatchJob(definition)

Now we are ready to submit this job to AWS Batch! Submitting this Batch job is easy, and introspection can be performed immediately:

>>> response = job.submit(queue='prod')
>>> job.is_submitted()
True

When the job is in a RUNNING state we can access the job's Cloudwatch logs. The log events are returned as objects which have useful properties such as timestamp and message.

>>> for log_event in job.log_stream_events():
...     print(log_event)
LogEvent(timestamp="1543809952329", message="You have started up this demo job", ingestion_time="1543809957080")
LogEvent(timestamp="1543809955437", message="Configuration, we are loading from...", ingestion_time="1543809957080")
LogEvent(timestamp="1543809955437", message="Defaulting to approximate values", ingestion_time="1543809957080")
LogEvent(timestamp="1543809955437", message="Setting up logger, nothing to see here", ingestion_time="1543809957080")

And if we must, we can cancel the job as long as we provide a reason:

>>> response = job.terminate(reason='I was just testing!')