Hot restart Python wrapper

Typically, Envoy will be hot restarted for config changes and binary updates. However, in many cases, users will wish to use a standard process manager such as monit, runit, etc. We provide /restarter/ to make this straightforward.

The restarter is invoked like so: might be defined like so (using salt/jinja like syntax):


ulimit -n {{ pillar.get('envoy_max_open_files', '102400') }}
sysctl fs.inotify.max_user_watches={{ pillar.get('envoy_max_inotify_watches', '524288') }}

exec /usr/sbin/envoy -c /etc/envoy/envoy.cfg --restart-epoch $RESTART_EPOCH --service-cluster {{ grains['cluster_name'] }} --service-node {{ grains['service_node'] }} --service-zone {{ grains.get('ec2_availability-zone', 'unknown') }}

Note on inotify.max_user_watches: If Envoy is being configured to watch many files for configuration in a directory on a Linux machine, increase this value as Linux enforces limits on the maximum number of files that can be watched.

The RESTART_EPOCH environment variable is set by the restarter on each restart and must be passed to the --restart-epoch option.


Special care must be taken if you wish to use the --use-dynamic-base-id option. That flag may only be set when the RESTART_EPOCH is 0 and your must obtain the chosen base ID (via --base-id-path), store it, and use it as the --base-id value on subsequent invocations (when RESTART_EPOCH is greater than 0).

The restarter handles the following signals:

  • SIGTERM or SIGINT (Ctrl-C): Will cleanly terminate all child processes and exit.

  • SIGHUP: Will hot restart by re-invoking whatever is passed as the first argument to the hot restart script.

  • SIGCHLD: If any of the child processes shut down unexpectedly, the restart script will shut everything down and exit to avoid being in an unexpected state. The controlling process manager should then restart the restarter script to start Envoy again.

  • SIGUSR1: Will be forwarded to Envoy as a signal to reopen all access logs. This is used for atomic move and reopen log rotation.