Only recreate containers whose configuration needs updating

[aanand]

Currently, docker-compose up deletes and recreates every container, so as to make sure any changes to docker-compose.yml have been picked up. Instead, it should inspect container configuration to see if they need recreating, and do the minimum necessary.

[dnephin]

Would this mean that if a container had changed state in some way (ex: wrote any files to a non-volume directory), the old container would be started instead? For testing/CI this wouldn't be ideal, so a way to preserve the existing behaviour might be nice. Best practice in those situations would be to docker-compose rm before running the test suite anyway, but maybe still worth calling out that this isn't backwards compatible in some situations.

[viranch]

This would be a problem if I am binding a config directory in the container and I just want to recreate the container due to config updates. There will be essentially no changes in container config but the bound volume will have changed content.

[aanand]

@viranch In that case - where you don't actually need a new container, you just want to restart the process - docker-compose restart would be the appropriate command.

[viranch]

Right - fair enough.

[TomasTomecek]

How do you inspect that configuration for particular container has changed?

[aanand]

@TomasTomecek It's implemented in #1399 - we add a label to each container with a hash of the image and service config and compare hashes when checking for staleness.