imclab/cronq

An application cron-like system

CronQ

A cron-like system to run your application tasks across any node, instead of one special snowflake. This is done by keeping your tasks in MySQL and publishing them over AMQP to workers that will run your tasks and eventually save the results back into the DB. This was started as a hackathon project at SeatGeek

Requirements

• Python 2.7
• RabbitMQ 3.x
• MySQL 5.x

Installation

Using PIP via PyPi:

pip install cronq

Using PIP via Github:

pip install git+git://github.com/seatgeek/[email protected]#egg=cronq

Adding to your requirements.txt file (run pip install -r requirements.txt afterwards):

git+ssh://[email protected]/seatgeek/[email protected]#egg=cronq

Usage

There are various workers that are used by Cronq, as well as a web admin.

cronq-runner

The runner executes tasks, and should be run on hosts that will actually perform work. There is no limit to the number of runners that can execute.

The runner requires /var/log/cronq/ to exist and be writable by the user executing the runner.

# setup rabbitmq connection info
export RABBITMQ_HOST=localhost
export RABBITMQ_USER=guest
export RABBITMQ_PASS=guest

# specify the rabbitmq queue to listen to
export CRONQ_QUEUE=cronq_jobs # `cronq_jobs` is the default queue

# run commands
cronq-runner

When run, cronq-runner will:

• Setup a rabbitmq connection
• Listen to the cronq_jobs queue
• Retrieve commands from the queue
• Publish a message saying the command is started
• Run the command in a shell subprocess
• Publish a message on success/failure to the cronq exchange and cronq_results queue. This is not configurable.
• Listen for more messages

cronq-injector

The cronq-injector command will non-destructively create any necessary cronq tables, though it will need a database to perform this action against. Please note that if you do not have tables created, it is helpful to run the injector first.

The injector is used to retrieve jobs from the database and publish them to AMQP. Jobs are published in the following format:

# where job is a database record
{
    'name': job.name,
    'command': unicode(job.command),
    'id': job.id,
}

You can ostensibly run as many injectors as necessary. MySQL isolation levels are used to attain locks on job records.

# setup rabbitmq connection info
export RABBITMQ_HOST=localhost
export RABBITMQ_USER=guest
export RABBITMQ_PASS=guest

# specify the database connection string
export CRONQ_MYSQL=mysql+mysqlconnector://cronq:cronq@localhost/cronq

# run the comand injector
cronq-injector

cronq-injector perform a 1 second sleep between job injections, but may perform an unlimited number of job injections in that time period.

Note that jobs are not queued up at the exact time you specify in the database. Rather, jobs that matches the following heuristic are queued one-at-a-time until no jobs are left to be queued for that injection cycle:

Job.next_run < NOW() OR Job.run_now = 1

cronq-results

The results aggregator listens to the cronq_results queue for the results of cronq-runner executions. You can run as many of these as possible, as they will retrieve results one-at-a-time from rabbitmq.

# setup rabbitmq connection info
export RABBITMQ_HOST=localhost
export RABBITMQ_USER=guest
export RABBITMQ_PASS=guest

# specify the database connection string
export CRONQ_MYSQL=mysql+mysqlconnector://cronq:cronq@localhost/cronq

# run the results-aggregator
cronq-results

These results can be viewed for particular commands within the web-admin, or by inspecting the database.

cronq-web

The web view is a WSGI app run from cronq.web:app and requires only database access. The following is an example for running the web admin using webscale technologies:

# install libevent-dev
sudo apt-get install libevent-dev

# install required python libraries
sudo pip install greenlet gevent gunicorn

# specify the database connection string
export CRONQ_MYSQL=mysql+mysqlconnector://cronq:cronq@localhost/cronq

# run the web admin
gunicorn --access-logfile - -w 2 --worker-class=gevent cronq.web:app

# access the panel on http://127.0.0.1:8000

The web admin will list available commands, their result history, and a button to allow you to immediately schedule a job.

Categories Api

The web admin exposes a category endpoint which allows you to replace a set of jobs with a single API call

curl -v 'localhost:5000/api/category/example' -f -XPUT -H 'content-type: application/json' -d '
{
    "category": "example",
    "jobs": [{
        "name": "Test Job",
        "schedule": "R/2013-05-29T00:00:00/PT1M",
        "command": "sleep 10",
        "routing_key": "slow"
    }]
}'

This adds / updates a job named Test Job in the example category. The time format is ISO 8601. Any jobs no longer defined for the example category will be removed. This allows you to script job additions / removes in your VCS.

License

BSD