Skip to main content

Azkaban CLI

Project description

A lightweight Azkaban client providing:

  • A command line interface to run jobs, upload projects, and more.

    $ azkaban upload
    Project my_project successfully uploaded (id: 1, size: 205kB, version: 1).
    Details at https://azkaban.server.url/manager?project=my_project
  • A simple syntax to define workflows from a single python file.

    from azkaban import Job, Project
    project = Project('my_project')
    project.add_file('/path/to/bar.txt', 'bar.txt')
    project.add_job('bar', Job({'type': 'command', 'command': 'cat bar.txt'}))


Using pip:

$ pip install azkaban

Command line interface

Once installed, the azkaban executable provides the following commands:

azkaban (create | delete) [options]
azkaban run [options] FLOW [JOB ...]
azkaban upload [options] ZIP

Running azkaban --help shows the full list of options.

URLs and aliases

The previous commands all take a --url, or -u, option used to specify where to find the Azkaban server (and which user to connect as).

$ azkaban create -u

In order to avoid having to input the entire URL every time, it is possible to defines aliases in ~/.azkabanrc:

default.alias = foo
foo =
bar = baruser@

We can now interact directly with each of these URLs using the --alias, or -a option followed by their corresponding alias. Since we also specified a default alias, it is also possible to omit the option altogether. As a result, the commands below are all equivalent:

$ azkaban create -u
$ azkaban create -a foo
$ azkaban create

Finally, our session ID for a given URL is cached on each successful login, so that we don’t have to authenticate on every remote interaction.


  • Creating and deleting projects:

    $ azkaban create
    Project name: my_project
    Description [my_project]: Some interesting description.
    Project my_project successfully created.
    Details at https://azkaban.server.url/manager?project=my_project
    $ azkaban delete -a bar
    Project name: my_project
    Project my_project successfully deleted.
  • Uploading an already built archive to an Azkaban server:

    $ azkaban upload -p my_project
  • Run entire workflows, or individual jobs:

    $ azkaban run my_workflow
    Flow my_workflow successfully submitted (execution id: 1).
    Details at https://azkaban.server.url/executor?execid=1


For medium to large sized projects, it quickly becomes tricky to manage the multitude of files required for each workflow. .properties files are helpful but still do not provide the flexibility to generate jobs programmatically (i.e. using for loops, etc.). This approach also requires us to manually bundle and upload our project to the gateway every time.

We provide here a convenient framework to define jobs from a single python file. This framework is entirely compatible with the command line interface above, and even provides additional functionality (e.g. building and uploading projects in a single command).


We start by creating a configuration file for our project. Let’s call it, the default file name the command line tool will look for. Here’s a simple example of how we could define a project with a single job and static file:

from azkaban import Job, Project

project = Project('foo')
project.add_file('/path/to/bar.txt', 'bar.txt')
project.add_job('bar', Job({'type': 'command', 'command': 'cat bar.txt'}))

The add_file method adds a file to the project archive (the second optional argument specifies the destination path inside the zip file). The add_job method will trigger the creation of a .job file. The first argument will be the file’s name, the second is a Job instance (cf. Job options).

Once we’ve saved our jobs file, the following additional commands are available to us:

  • azkaban list, see the list of all jobs in the current project.
  • azkaban view, view the contents of the .job file for a given job.
  • azkaban build, build the project archive and store it locally.

Job options

The Job class is a light wrapper which allows the creation of .job files using python dictionaries.

It also provides a convenient way to handle options shared across multiple jobs: the constructor can take in multiple options dictionaries and the last definition of an option (i.e. later in the arguments) will take precedence over earlier ones.

We can use this to efficiently share default options among jobs, for example:

defaults = {'': 'boo', 'retries': 0}

jobs = [
  Job({'type': 'noop'}),
  Job(defaults, {'type': 'noop'}),
  Job(defaults, {'type': 'command', 'command': 'ls'}),
  Job(defaults, {'type': 'command', 'command': 'ls -l', 'retries': 1}),

All jobs except the first one will have their property set. Note also that the last job overrides the retries property.

Alternatively, if we really don’t want to pass the defaults dictionary around, we can create a new Job subclass to do it for us:

class BooJob(Job):

  def __init__(self, *options):
    super(BooJob, self).__init__(defaults, *options)


Nested options

Nested dictionaries can be used to group options concisely:

# e.g. this job
  'proxy.user': 'boo',
  'proxy.keytab.location': '/path',
  'param.input': 'foo',
  'param.output': 'bar',
# is equivalent to this one
  'proxy': {'user': 'boo', 'keytab.location': '/path'},
  'param': {'input': 'foo', 'output': 'bar'},

Merging projects

If you have multiple projects, you can merge them together to create a single project. The merge is done in place on the project the method is called on. The first project will retain its original name.

from azkaban import Job, Project

project1 = Project('foo')
project1.add_file('/path/to/bar.txt', 'bar.txt')
project1.add_job('bar', Job({'type': 'command', 'command': 'cat bar.txt'}))

project2 = Project('qux')
project2.add_file('/path/to/baz.txt', 'baz.txt')
project2.add_job('baz', Job({'type': 'command', 'command': 'cat baz.txt'}))

# project1 will now contain baz.txt and the baz job from project2

Job details

The info command becomes quite powerful when combined with other Unix tools. Here are a few examples:

$ # To count the number of jobs per type
$ azkaban info -o type | cut -f 2 | sort | uniq -c
$ # To only view the list of jobs of a certain type with their dependencies
$ azkaban info -o type,dependencies | awk -F '\t' '($2 == "job_type")'
$ # To view the size of each file in the project
$ azkaban info -f | xargs -n 1 du -h

Next steps

Any valid python code can go inside the jobs configuration file. This includes using loops to add jobs, subclassing the base Job class to better suit a project’s needs (e.g. by implementing the on_add and on_build handlers), …



Because pig jobs are so common, a PigJob class is provided which accepts a file path (to the pig script) as first constructor argument, optionally followed by job options. It then automatically sets the job type and adds the corresponding script file to the project.

from azkaban import PigJob

project.add_job('baz', PigJob('/.../baz.pig', {'dependencies': 'bar'}))

Using a custom pig type is as simple as changing the PigJob.type class variable.

This extension also comes with the azkabanpig executable to run pig scripts directly. azkabanpig --help will display the list of available options (using UDFs, substituting parameters, running several scripts in order, etc.).

Project details

Release history Release notifications

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size & hash SHA256 hash help File type Python version Upload date
azkaban-0.3.1.tar.gz (22.7 kB) Copy SHA256 hash SHA256 Source None

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page