Docker container with status "Dead" after consul healthcheck runs - docker

Question

I am using consul's healthcheck feature, and I keep getting these these "dead" containers:
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
20fd397ba638 progrium/consul:latest "\"/bin/bash -c 'cur 15 minutes ago Dead
What is exactly a "Dead" container? When does a stopped container become "Dead"?
For the record, I run progrium/consul + gliderlabs/registrator images + SERVICE_XXXX_CHECK env variables to do health checking. It runs a healthcheck script running an image every X secs, something like docker run --rm my/img healthcheck.sh
I'm interested in general to what "dead" means and how to prevent it from happening. Another peculiar thing is that my dead containers have no name.
this is some info from the container inspection:
"State": {
"Dead": true,
"Error": "",
"ExitCode": 1,
"FinishedAt": "2015-05-30T19:00:01.814291614Z",
"OOMKilled": false,
"Paused": false,
"Pid": 0,
"Restarting": false,
"Running": false,
"StartedAt": "2015-05-30T18:59:51.739464262Z"
},
The strange thing is that only every now and then a container becomes dead and isn't removed.
Thank you
Edit:
Looking at the logs, I found what makes the container stop fail:
Handler for DELETE /containers/{name:.*} returned error: Cannot destroy container 003876e41429013e46187ebcf6acce1486bc5011435c610bd163b159ba550fbc:
Driver aufs failed to remove root filesystem 003876e41429013e46187ebcf6acce1486bc5011435c610bd163b159ba550fbc:
rename /var/lib/docker/aufs/diff/003876e41429013e46187ebcf6acce1486bc5011435c610bd163b159ba550fbc
/var/lib/docker/aufs/ diff/003876e41429013e46187ebcf6acce1486bc5011435c610bd163b159ba550fbc-removing:
device or resource busy
Why does this happen?
edit2:
found this: https://github.com/docker/docker/issues/9665

Answer 1

Update March 2016: issue 9665 has just been closed by PR 21107 (for docker 1.11 possibly)
That should help avoid the "Driver aufs failed to remove root filesystem", "device or resource busy" problem.
Original answer May 2015
Dead is one if the container states, which is tested by Container.Start()
if container.removalInProgress || container.Dead {
return fmt.Errorf("Container is marked for removal and cannot be started.")
}
It is set Dead when stopping fails, in order to prevent that container to be restarting.
Amongst the possible cause of failure, see container.Kill().
It means kill -15 and kill -9 are both failing.
// 1. Send a SIGTERM
if err := container.killPossiblyDeadProcess(15); err != nil {
logrus.Infof("Failed to send SIGTERM to the process, force killing")
if err := container.killPossiblyDeadProcess(9); err != nil {
That usually mean, as the OP mention, a busy device or resource, preventing the process to be killed.

Answer 2

There are a lot of bugs caused by EBUSY, in particular when devicemapper is used.
There is a tracker bug for all of the EBUSY related issues.
see https://github.com/docker/docker/issues/5684#issuecomment-69052334