Open darkl0rd opened 5 years ago
We have also seen this issue.
It happens because the docker plugin has a bug in the reference counter feature, so that when a container is stopped it thinks there are still containers referencing the volume and does not detach it from the vm host.
Setup: VSphere cluster, consisting of 4 ESX hosts. Several Docker Hosts running on one of the ESX hosts, many of it's containers are using persistent volumes.
When the Docker VM is migrated to another ESX host, some, but not all of the volumes no longer correctly attach. After restarting the Docker Engine (sometimes several times), the volumes can be attached correctly. Strangely enough, in some cases the containers do start, but when inspecting the container, the volume isn't actually mounted; instead a "local" volume is used (even though docker volume ls does not show the volume as being local). This can be verified by docker exec mount
Summary: