Need way to limit exponential container restart delay

[macdjord]

When a container with a restart policy fails repeatedly, Docker imposes an exponentially increasing delay between restart attempts. This is, in general, useful, since it prevents a container which cannot start successfully from overloading the system. However, there is currently no way to limit how large this delay can grow.

The particular case I am dealing with is a container which depends on a particular system being in a ready state before it can start up, and exits if it is not. Given it is relatively cheap to start the container and check the state, I would like it to try no less than once every 10s. As-is, if the system is down, the restart delay can grow to an absurd length.

Things Which Will Not Work

• 'Use container dependencies'
  • The system which the container needs to wait on is not itself a container.
• 'Move the check-wait-recheck loop inside the container'
  • This results in the container being 'up' when it is not, in fact, in operation condition
• 'Docker swarm has this option called 'restart_policy'...'
  • I am not using Swarm. This functionality should be in base docker.

Other Things Which Would Be Nice To Be Able To Adjust

• The minimum delay at which the exponential backoff starts
• The amount of time the container must be up in order to count as 'successful' and reset the backoff timer
• Whether a run of the container which lasts longer than the required time but exits with a nonzero code should count as 'successful' at all (or conversely, whether a run which doesn't last that long but exits with code 0 should count as 'successful' automatically)