cyanide 1.0.0

Introduction

This stress test suite will attempt to break the Celery worker in different ways, and can also be used to write new stress test suites for projects depending on Celery infrastructure.

The worker must currently be started separately, and it’s encouraged to repeat the suite using workers started with different configuration values.

Ideas include:

1. Default, single process:

   $ celery -A cyanide worker -c 1

2. Default, multiple processes:

   $ celery -A cyanide worker -c 8

3. Frequent maxtasksperchild recycling, single child process:

   $ celery -A cyanide worker -c 1 --maxtasksperchild=1

4. Frequent autoscale scale down & maxtasksperchild, single child process:

   $ AUTOSCALE_KEEPALIVE=0.01 celery -A cyanide worker \
   >   --autoscale=1,0 --maxtasksperchild=1

5. Frequent maxtasksperchild, multiple child processes:

   $ celery -A cyanide worker -c 8 --maxtasksperchild=1

6. Processes terminated by time limits:

   $ celery -A cyanide worker --time-limit=1

7. Frequent maxtasksperchild, single child process with late ack:

   $ celery -A cyanide worker -c1 --maxtasksperchild=1 -Z acks_late

8. Worker using the eventlet pool:

   Start the worker, here having a thousand green-threads:

   $ celery -A cyanide worker -c1000 -P eventlet

   You must activate the green test group when starting the test suite:

   $ celery cyanide -g green

9. Worker using the gevent pool:

   Start the worker, here having a thousand green-threads:

   $ celery -A cyanide worker -c1000 -P gevent

   You must activate the green test group when starting the test suite:

   $ celery cyanide -g green

Tips

It’s a good idea to include the --purge <celery worker --purge> argument to clear out tasks from previous runs.

Note that the stress client will probably hang if the test fails, so this test suite is currently not suited for automatic runs.

Configuration Templates

You can select a configuration template using the -Z command-line argument to any celery -A cyanide command or the celery cyanide command used to execute the test suite.

The templates available are:

• default

  Using AMQP as a broker, RPC as a result backend, and using JSON serialization for task and result messages.

  Both broker and result store is expected to run at localhost.

• vagrant1

  Use the VM started by celery vagrant up as the broker and result backend (RabbitMQ).

• vagrant1_redis

  Use the VM started by celery vagrant up as the broker and result backend (Redis).

• redis

  Using Redis as a broker and result backend.

• redistore

  Using Redis as a result backend only.

• acks_late

  Enables late ack globally.

• pickle

  Using pickle as the serializer for tasks and results (also allowing the worker to receive and process pickled messages)

• confirms

  Enables RabbitMQ publisher confirmations.

• events

  Configure workers to send task events.

• proto1

  Use version 1 of the task message protocol (pre 4.0)

You can see the resulting configuration from any template by running the command:

$ celery -A cyanide report -Z redis

$ cyanide -Z redis

$ celery -A cyanide worker -Z redis

$ celery -A cyanide worker -Z redis,acks_late

Test Suite Options

After one or more worker instances are running, you can start executing the tests.

By default the complete test suite will be executed:

$ celery cyanide

You can also specify what test cases to run by providing one or more names as arguments:

$ celery cyanide revoketermfast revoketermslow

A full list of test case names can be retrieved with the -l <celery cyanide -l> switch:

$ celery cyanide -l
.> 1) chain,
.> 2) chaincomplex,
.> 3) parentids_chain,
.> 4) parentids_group,
.> 5) manyshort,
.> 6) unicodetask,
.> 7) always_timeout,
.> 8) termbysig,
.> 9) timelimits,
.> 10) timelimits_soft,
.> 11) alwayskilled,
.> 12) alwaysexits,
.> 13) bigtasksbigvalue,
.> 14) bigtasks,
.> 15) smalltasks,
.> 16) revoketermfast,
.> 17) revoketermslow

You can also start from an offset within this list, e.g. to skip the first two tests use --offset=2 <celery cyanide --offset>:

$ celery cyanide --offset=2

See celery cyanide --help for a list of all available command-line options.

Vagrant

$ celery vagrant up

$ celery -A worker -Z vagrant1
$ celery cyanide -Z vagrant1

$ ssh $(celery vagrant sshargs)

$ celery vagrant halt

$ celery vagrant destroy

Environment Variables

$ CYANIDE_BROKER='amqp://' celery -A cyanide worker # ...
$ CYANIDE_BROKER='amqp://' celery cyanide

$ CYANIDE_BACKEND='amqp://' celery -A cyanide worker # ...
$ CYANIDE_BACKEND='amqp://' celery cyanide

$ CYANIDE_QUEUE='cyanide' celery -A cyanide worker # ...
$ CYANIDE_QUEUE='cyanide' celery cyanide

Custom Suites

You can define custom suites (look at source code of cyanide.suites.default for inspiration), and tell cyanide to use that suite by specifying the celery cyanide -S option:

$ celery cyanide -S proj.funtests:MySuite

Installation

You can install cyanide either via the Python Package Index (PyPI) or from source.

To install using pip:

$ pip install -U cyanide

$ tar xvfz cyanide-0.0.0.tar.gz
$ cd cyanide-0.0.0
$ python setup.py build
# python setup.py install

$ pip install https://github.com/celery/cyanide/zipball/master#egg=cyanide