tcp4 / tcp port not being exposed/mapped to running container after it's been in use before

[omartin2010]

Expected behavior

Port exposed after relaunching a countainer

Actual behavior

It's only exposed on tcp6 (that I can see when issuing a netstat -an | grep -i listen). Also, if I do a curl http://[::1]:8888 for example, that curl returns the expected content. But not if I run http://127.0.0.1:8888 which should also answer.

Information

• Full output of the diagnostics from "docker-diagnose" ran from one of the instance
• A reproducible case if this is a bug, Dockerfiles FTW
• Page URL if this is a docs issue or the name of a man page

Steps to reproduce the behavior

Run following commands:

nvidia-docker run --name jetson_training -d -p:8888:5000 -v /data-drive-1:/data-drive-1:ro -v ~/digits/digits-jobs/:/workspace/jobs nvcr.io/nvidia/digits:1
8.05
docker stop jetson_training
docker start jetson_training

Or I can try to delete the container (docker rm jetson_training) and re-creating it on the same port (8888 in this case) doesn't work. If however I use a new port (one docker wasn't bound to), that worked.
I rebooted the machine, didn't change anything. I also tried with another image. I'm running this on a Azure data science VM (DSVM) running ubuntu.