Gracefully restarting processes is a delicate business. rolld helps you restart any group of processes without losing a single request. This is especially useful during no-downtime deployments and in continuous integration implementations.

There are different tactics when restarting a process group. rolld does the following:

• Starts a new set of processes in different ports
• Runs smoke tests (eg. send predefined requests to each port)
• Updates your load balancer configuration and reloads it
• Waits until there is no connection on the old set of processes
• Kills old processes

rolld uses a simple yml file for configuration:

app-servers:
  cmd: '../env/bin/python app.py {numproc}'
  directory: '/Users/aybarsbadur/projects/rolld/example'
  stderr: 'logs/tornado-{numproc}.err'
  stdout: 'logs/tornado-{numproc}.log'

  workset:
    - numprocs: [8001, 8002, 8003, 8004]
    - numprocs: [8501, 8502, 8503, 8504]

  smoke_test:
    - check_all_http_ports: numprocs

  after_start:
    - replace_file: [numprocs, 'example/nginx-tornado.conf.templ', 'example/nginx-tornado.conf']
    - run_proc: ps auxw| grep nginx | grep master | grep -v grep | awk '{print $2}' | xargs kill -HUP

Then you can run rolld like this:

rolld rolld.yml

When you want to restart your processes send a HUP signal to rolld

killall -HUP rolld

or

ps auxw | grep rolld | grep -v grep | awk '{print $2}' | xargs kill -HUP

Always refer to your application server documentation first (uwsgi, gunicorn, passenger etc.) they may have graceful restart built-in already.

If you are not sure if you need rolld, there is an easy way to figure it out. You can easily check if your application server does graceful loading by running curl every second:

while curl --fail -s localhost:80 > /dev/null; do echo "passed"; sleep 1; done;

And then reload/restart your application server. If it doesn't exit, then you are fine, else you need a solution like rolld to handle it for you.

rolld is brought to you by Aybars Badur and the Hipo Team.

rolld is licensed under the terms of the Apache License, version 2.0. Please see the LICENSE file for full details.