Description of problem: After awhile of running docker commands, particularly as part of a "watch" statement, I state to get too many pipe2 errors. Version-Release number of selected component (if applicable): How reproducible: Unknown. This has happened to me 3 times so far. Steps to Reproduce: 1. Run a docker command in a watch loop, such as: watch -n 30 docker run -t -i --rm fedora date 2. 3. Actual results: Eventually one starts to see output like: time="2015-04-09T15:25:25-04:00" level="fatal" msg="Post http:///var/run/docker.sock/v1.16/containers/3543b03f71195ca4522e794e7e0d83d8642b59ceb140d30a9d7fcef760424893/start: EOF. Are you trying to connect to a TLS-enabled daemon without TLS?" If I control-C, run "docker rm $(docker ps -q -a)" there are many instances to clean-up. Afterwards, if I try running a new docker command without the watch part I get the error: FATA[0000] Error response from daemon: Cannot start container 6252aa7de7f1217f654d1551c56e9d544eb9a588446996e0ae473458b2bf1251: pipe2: too many open files Expected results: Should just keep working. Additional info: The command I'm actually using is an sqlplus script, but I thought an example with date might be much easier to use to reproduce with. Doing a "service docker restart" resolves the issue, but it also aborts all my running docker containers...
The following identifies the problem without the need to wait a long time: 1. run: ps -ef 2. note the process number for docker, say for example 31215 3. run: ls /proc/31215/fd | wc -l 4. Run a docker command such as: docker run -t -i --rm fedora date 5. run: ls /proc/31215/fd | wc -l Note: Each time you run a docker command the number of open file descriptors is increasing by 2. So eventually the limit on file descriptors will be reached. The docker package I'm using is: docker-io-1.4.1-3.el6.x86_64
The following looks like the same issue in github: https://github.com/docker/swarm/issues/342 There is a fix for it, so presumably the docker-io package just needs to be updated to add it.
*** This bug has been marked as a duplicate of bug 1189028 ***