Open ablotim opened 5 years ago
It seems the process running in the container (the php process) did get killed but the container itself is not cleaned up correctly. The only workaround we found until now is to restart the dockerd on the host which cleans up the zombie containers and returns the host to a somewhat stable state.
Expected behavior
When we deploy a docker stack on our swarm all service instances should update to the new image version
Actual behavior
Quite often containers keep running an older version of the image
Steps to reproduce the behavior
Output of
docker version
:Output of
docker info
:Additional environment details (AWS, VirtualBox, physical, etc.)
example of a service that often exhibits this behaviour:
example outputs of "docker service ps":
also
We noticed the issues most often are on the same handful of nodes, notably docker3036 and docker3037, but the only this 'special' about these is that they're also running redis containers. When we restart dockerd things always get back to normal, though we occasionally need to restart it twice.