Reference

Introduction

A signac project manages a data space which is divided into segments, where each segment is strongly associated with a unique set of parameters: a state point. The signac-flow extension provides means to implement a workflow via the flow.FlowProject which inherits from signac.Project. This workflow is based on two core concepts: job classification and data space operations.

Classfication

We classify the state of a Job using text labels. These labels can be determined by a simple generator function, e.g.:

def classify(job):
   if job.isfile('init.txt'):
       yield 'initialized'

Operations

A data space operation is any action that will modify the data space.

This is an example for an operation implemented in python:

def initialize(job):
    with job:
        with open('init.txt', 'w') as file:
            file.write('Hello world!')

The initialize operation will create a file called init.txt within a job‘s workspace.

The default workflow

Combining the concepts of classification and operations we can define the workflow logic of a flow.FlowProject by implementing the classify() and the next_operation() method:

from flow import FlowProject
from flow import JobOperation

class MyProject(FlowProject):

    def classify(self, job):
       if job.isfile('init.txt'):
           yield 'initialized'
       if job.isfile('dump.txt'):
           yield 'processed'

    def next_operation(self, job):
       labels = set(self.classify(job))

       def op(name):
           return JobOperation(name, job, 'python scripts/run.py {} {}'.format(name, job))

       if 'initialized' not in labels:
           return op('initialize')
       if 'processed' not in labels:
           return op('process')

The next_operation() returns the default operation to execute next for a job in the identified state. This operation is a command, which can be executed on the command line. In the template, all operations are defined in the scripts/operations.py module and are executed by the scripts/run.py script.

We can get a quick overview of our project’s status via the print_status() method:

>>> project = MyProject()
>>> project.print_status(detailed=True, params=('a',))
Status project 'MyProject':
Total # of jobs: 10
label        progress
-----------  -------------------------------------------------
initialized  |########--------------------------------| 20.00%
processed    |####------------------------------------| 10.00%

Detailed view:
job_id                            S    next_op       a  labels
--------------------------------  ---  ----------  ---  ----------------------
108ef78ec381244447a108f931fe80db  U !  sample      1 1  processed, initialized
be01a9fd6b3044cf12c4a83ee9612f84  U !  process     3 2  initialized
32764c28ef130baefebeba76a158ac4e  U !  initialize  2.3
# ...

Tip

You can print the project’s status from the command line by executing $ python -m my_project.status.

Running operations

All python-based operations are implemented in the scripts/operations.py module. We can use the scripts/run.py script to execute them directly, e.g.:

$ python scripts/run.py initialize 108e

This command will execute the initialize operation for the job identified by the 108e... id.

Scheduling

To take full advantage of the workflow management, it is advantagous to use a Scheduler which schedules the execution of job-operations for us. The project template attempts to detect available schedulers through the environment module, but might require some tweaking based off your particular computing environment.

To submit job-operations to a scheduler, call the submit() method.

Tip

You can submit job operations to a scheduler from the command line, by executing $ python my_project.submit.

The submit() method will schedule the execution of operations for specified jobs by generating and submitting a jobscript to the scheduler.

Every job submission script has the same basic structure:

1. environment dependent header (e.g. scheduler options)
2. operation-agnostic header (e.g. switching into the project root directory)
3. commands to execute operations

The scheduler header will vary across different scheduler implementations and should be configured via the environment module.

In summary, if we only execute operations defined in the operations module, we can run them either directly or submit them to a scheduler:

python scripts/run.py OPERATION [JOBID] ...
python -m my_project.submit [-j OPERATION] [JOBID] ...