skip to navigation
skip to content

jobspy 0.26.1

Use Redis as a job input/output coordinator.

Package Documentation

Job resource input/output control using Redis as a locking layer

Copyright 2016 Josiah Carlson

This library licensed under the GNU LGPL v2.1

The initial library requirements and implementation were done for OpenMail LLC. (this library) was more or less intended to offer input and output control like Luigi and/or Airflow (both Python packages), with fewer hard integration requirements. In fact, has been used successfully as part of jobs running in a cron schedule via Jenkins, in build chains in Jenkins, inside individual rpqueue tasks, and even inside individual Flask web requests for some high-value data ( is backed by Redis, so job locking overhead can be low, even when you need to keep data safe).

Source: PyPI: Docs:


Input/output locking on multiple named keys, called “inputs” and “outputs”.

  • All keys are case-sensitive
  • Multiple readers on input keys
  • Exclusive single writer on output keys (no readers or other writers)
  • All inputs must have been an output previously
  • Optional global and per-job history of sanitized input/output edges (enabled by default)
  • Lock multiple inputs and outputs simultaneously, e.g. to produce outputs Y and Z, I need to consume inputs A, B, C.

How to use

  • Install

    $ sudo pip install jobspy
  • Import and configure the Redis connection required (maybe put this in some configuration file):

    # in or
    import jobs
    jobs.CONN = redis.Redis(...)
  • Use as a decorator on a function (must explicitly .start() the job, .stop() performed automatically if left uncalled):

    # in
    @jobs.resource_manager(['input1', 'input2'], ['output1', 'output2'], duration=300, wait=900)
    def some_job(job):
        job.add_inputs('input6', 'input7')
        # At this point, all inputs and outputs are locked according to the
        # locking semantics specified in the documentation.
        # If you call job.stop(failed=True), then the outputs will not be
        # "written"
        # If you call job.stop(), then the outputs will be "written"
        # Alternating job.stop() with job.start() is okay! You will drop the
        # locks in the .stop(), but will (try to) get them again with the
        # .start()
        # But if you need the lock for longer than the requested duration, you
        # can also periodically refresh the lock. The lock is only actually
        # refreshed once per second at most, and you can only refresh an already
        # started lock.
        # If an exception is raised and not caught before the decorator catches
        # it, the job will be stopped by the decorator, as though failed=True:
        raise Exception("Oops!")
        # will stop the job the same as
        #   job.stop(failed=True)
        # ... where the exception will bubble up out of the decorator.
  • Or use as a context manager for automatic start/stop calling, with the same exception handling semantics as the decorator:

    def multi_step_job(arg1, arg2, ...):
        with jobs.ResourceManager([arg1], [arg2], duration=30, wait=60, overwrite=True) as job:
            for something in loop:
                # do something
            if bad_condition:
                raise Exception("Something bad happened, don't mark arg2 as available")
            elif other_bad_condition:
                # stop the job, not setting
        # arg2 should exist after it was an output, and we didn't get an
        # exception... though if someone else is writing to it immediately in
        # another call, then this may block...
        with jobs.ResourceManager([arg2], ['output.x'], duration=60, wait=900, overwrite=True):
            # something else
        # output.x should be written if the most recent ResourceManager stopped
        # cleanly.

More examples

  • Scheduled at 1AM UTC (5/6PM Pacific, depending on DST):

    import datetime
    FMT = '%Y-%m-%d'
    def yesterday():
        return (datetime.datetime.utcnow().date() - datetime.timedelta(days=1)).strftime(FMT)
    @jobs.resource_manager([], (), 300, 900)
    def aggregate_daily_events(job):
        yf = yesterday()
        # outputs 'reporting.events_by_partner.YYYY-MM-DD'
        # we can add job inputs and outputs inside a decorated function before
        # we call .start()
        # actually aggregate events
  • Scheduled the next day around the time when we expect upstream reporting to be available:

    @jobs.resource_manager((), (), 300, 900)
    def fetch_daily_revenue(job):
        yf = yesterday()
        # actually fetch daily revenue
  • Executed downstream of fetch_daily_revenue():

    @jobs.resource_manager((), (), 300, 900)
    def send_reports(job):
        yf = yesterday()
        # having jobs inputs here ensures that both of the *expected* upstream
        # flows were *actual*
        # inputs are available, go ahead and generate the reports!
  • And in other contexts…:

    def make_recommendations(partners):
        yf = yesterday()
        for partner in partners:
            with jobs.ResourceManager([jobs.NG.reporting.report_by_partner[yf]],
                    [jobs.NG.reporting.recommendations_by_partner[yf][partner]], 300, 900):
                # job is already started
                # generate the recommendations for the partner

Configuration options

All configuration options are available as options on the module itself, though you can override the connection explicitly on a per-job basis. See the ‘Connection configuration’ section below for more details.:

# The Redis connection, REQUIRED!
jobs.CONN = redis.Redis()

# Sets a prefix to be used on all keys stored in Redis (optional)

# Keep a sanitized ZSET of inputs and outputs, available for traversal
# later. Note: sanitization runs the following on all edges before storage:
#   edge = re.sub('[0-9][0-9-]*', '*', edge)
# ... which allows you to get a compact flow graph even in cases where you
# have day-parameterized builds.

# To use a logger that doesn't print to standard output, set the logging
# object at the module level (see below). By default, the built-in "default
# logger" prints to standard output.
jobs.DEFAULT_LOGGER = logging.getLogger(...)

Using with a custom Redis configuration

If you would like to use as a script (for the convenient command-line options), you need to create a wrapper module, which can also act as your general configuration updates for (hack because I needed to release this as open-source before the end of summer):

import jobs
jobs.CONN = ...
jobs.GLOBAL_PREFIX = ...
jobs.GRAPH_HISTORY = ...

from jobs import *

if __name__ == '__main__':

Then you can use this as:

$ python --help

And you can use everywhere, which will have all of your configuration handled.:

import myjobs

def daily_reporting(job, ...):
    # exactly the same as before.
File Type Py Version Uploaded on Size
jobspy-0.26.1.tar.gz (md5) Source 2017-01-25 16KB