I am trying to use Docker to containerize a MongoDB instance using a drive mounted on host.
Using the mongo:latest image:
[user@dcos-master ~]$ docker run -d --name mongo -v /local/cluster/drive:/data/db mongo:latest
But constantly fails with:
exception in initAndListen: 98 Unable to lock file: /data/db/mongod.lock Function not implemented. Is a mongod instance already running?, terminating
Considered it a permission problem so I have attempted opening up the permissions to 777 on the host directory and also just running the mongod instance as root and I get the same problem.
There definitely is not another mongod instance running and pointing at the same data directory.
Lastly, something worth pointing out: The only common denominator seems to be the filesystem. Mongod consistently fails on Lustre drives but not on any other filesystem. If I the source directory is on an NFS drive or anything else, it runs great.
https://anilmaurya.wordpress.com/2015/05/06/abaqus-with-lustre-file-system/
By default lustre doesn't have the the flock feature [1], you have to add the "flock" option when mounting [2] the filesystem to make it work [3]
[1]
$ touch temp
$ strace -e flock perl -e 'open(FH, "+< temp"); flock(FH, 2) or exit 1; sleep 1; close(FH);'
flock(3, LOCK_EX) = -1 ENOSYS (Function not implemented)
+++ exited with 1 +++
[2]
You have to umount and mount the filesystem as lustre doesn't support the -o remount option:
$ mount | grep /data
mgs@tcp:/data on /data type lustre (rw,lazystatfs)
$ mount -o remount,flock /data
$ mount | grep /data
mgs@tcp:/data on /data type lustre (rw,lazystatfs)
# umount /data
# mount -t lustre -o rw,flock mgs@tcp:/data /data
$ mount | grep /data
mgs@tcp:/data on /data type lustre (rw,flock,lazystatfs)
[3]
$ strace -e flock perl -e 'open(FH, "+< temp"); flock(FH, 2) or exit 1; sleep 1; close(FH);'
flock(3, LOCK_EX) = 0
+++ exited with 0 +++