Docker networking broken after docker restart (xhyve restart required to fix)

[quinncomendant]

I have an issue that appears to be related to CoreOS. I'm raising the issue here because corectl is the only method I'm using CoreOS, and I'm sure you can point me where to move the issue if it's not appropriate here.

I recently updated to the latest beta CoreOS version and have been able to reproduce a broken state of docker networking (of all the images I have saved, the issue occurs with beta 1353.4.0 and beta 1381.1.0):

1. Start corectl: corectl start -H --offline --channel beta --volume ~/var/coreos/var_lib_docker.img.qcow2 --cloud_config ~/var/coreos/docker-only-with-persistent-storage.txt --cpus 1 --memory 1024 --name containerland --uuid 11111111-1111-1111-1111-111111111111
2. Bring up some docker containers with docker-compose up -d (see my (redacted) docker-compose.yml).
3. Stop, then restart the docker containers with docker-compose down … up or docker-compose restart
4. At this point, the website is not accessible at the local IP from the macOS host (http://192.168.64.2). The apache+php container hosting the site still function (docker-exec php bash then curl localhost works) but the php container no longer can communicate with the db container (the aforementioned curl command load the app but with "db connect failed").

The only way to get it working again is to restart the xhyve instance (corectl kill containerland && corectl start …).

The problem does not happen when using beta 1192.2.0 or alpha 1395.0.0. Restarting containers, bringing down && up, doesn't hurt networking—just works.

Since the problem works in the latest alpha version, the problem might have already been fixed. Should I report this somewhere? Is there any other info I can provide for diagnostics?