This is the container that will be deployed as a side task (storagetask) in a
Nomad job.
Right now we are using rclone to mount a remote file system
(therefore privileged permissions, or similar, are needed in the Nomad task).
In the future this same container could be adapted to use
Alluxio or any other storage solution we decide to support.
The Docker image is available in the project's Harbor registry:
registry.services.ai4os.eu/ai4os/docker-storage:latest
When launching this Dockerfile you should pass appropriate RCLONE configuration as ENV variables as well as mounting a volume in the provided LOCAL_PATH
⚠️ If several containers are using that volume, this container has to be ran first.
Mounting a remote filesystem requires either privileged flag or set of little more
restrictive permissions [cap-add|device|security-opt]
(ref).
docker run \
-e RCLONE_CONFIG='/srv/.rclone/rclone.conf' \
-e RCLONE_CONFIG_RSHARE_TYPE='webdav' \
-e RCLONE_CONFIG_RSHARE_URL='https://data-deep.a.incd.pt/remote.php/webdav/' \
-e RCLONE_CONFIG_RSHARE_VENDOR='nextcloud' \
-e RCLONE_CONFIG_RSHARE_USER='**my-nextcloud-token-user**' \
-e RCLONE_CONFIG_RSHARE_PASS='**my-nextcloud-token-pass**' \
-e REMOTE_PATH='rshare:/' \
-e LOCAL_PATH='/storage' \
-v /home/iheredia/demo-data:/storage:shared \
--device /dev/fuse \
--cap-add SYS_ADMIN \
--security-opt apparmor:unconfined \
-ti ignacioheredia/ai4-docker-storage /bin/bashIn the example above the RCLONE mounted dir will be accessed in the host machine at
/home/iheredia/demo-data (root permissions needed to access).
After exiting the container, you will need to unmount the folder:
umount /home/iheredia/demo-data
rmdir /home/iheredia/demo-dataRCLONE version
Latest RCLONE (1.63.1) did not work due to bug.
We settled with (1.62.2).
Base image
We do not start from official RCLONE images because we also need [sh|bash|curl] (not
provided)
Nomad task
-
If the user inputs the wrong USER/PASS, the job still runs fine. Only when trying to access the storage folder from his task they will see:
`ls: reading directory '/storage/': Input/output error`When the job finishes, you still need to unmount the folder before deleting.
-
the
rclone mountcommand never finishes so the job can still be run in principle as typeservice. But the docker start will fail the second time (at restart) because the volume folder has already been created.Error while creating mount source path '/etc/nomad.d/storage/some-random-uuid5': mkdir /etc/nomad.d/storage/some-random-uuid5: file exists
So we need to either:
- use
allow_otherin RCLONE (ref), still withservicemode. - run it in
batchmode (this means suboptimal placement).
We choose the first option.
- use