Oneclient docker not exposing NFS

[valerioformato]

After starting a container, passing host and token as env variables, the NFS access point is not exposed to the host system.

$ docker run -it --privileged -e ONECLIENT_PROVIDER_HOST -e ONECLIENT_ACCESS_TOKEN -v ~/.oneclient_local:/root/.local/share/oneclient onedata/oneclient:17.06.0-beta6
Connecting to provider 'oneprovider.cloud.cnaf.infn.it:5555' using session ID: '7800789129593989533'...
Getting configuration...
Oneclient has been successfully mounted in '/mnt/oneclient'.

but logging in the container one can easily see that NFS is not running, and doesn't start

$ docker ps
CONTAINER ID        IMAGE                             COMMAND                  CREATED             STATUS              PORTS               NAMES
40057641ed30        onedata/oneclient:17.06.0-beta6   "/root/run.sh /mnt..."   2 minutes ago       Up 2 minutes                            eloquent_morse
$ docker exec -it 40057641ed30 bash
root@40057641ed30:/tmp# service nfs-kernel-server status\
nfsd not running
root@40057641ed30:/tmp# service nfs-kernel-server start
 * Exporting directories for NFS kernel daemon...                                                              [ OK ]
 * Starting NFS kernel daemon                                                                                  [ OK ]
root@40057641ed30:/tmp# service nfs-kernel-server status
nfsd not running

as confirmed by trying to mount the remote volume on the host

$ docker inspect --format "{{ .NetworkSettings.IPAddress }}" $(docker ps -ql)
172.17.0.2

Tested on 17.06.0-beta6, OSX version 10.12.5, Docker version 17.06.0-ce-mac19 (18663)

[groundnuty]

oneclient: 17.06.0-beta6
Docker CE Version: 17.06.0-ce-mac19 (18663), Channel: stable, c98c1c25e0
macos10.12.6 (16G29)

@valerioformato yes, unfortunately, you are right. Moreover, it's not even possible to ping a container. The reasons for that are explained here.

The only viable solution proposed by Docker is to expose container ports to host:

• nfs: -p 111:111 -p 111:111/udp -p 2049:2049 -p 2049:2049/udp
• smb: -p 137:137/udp -p 138:138/udp -p 139:139 -p 445:445

The full command is as follows:

docker run -it -p 137:138/udp -p 137:138/udp -p 139:139 -p 445:445 -p 111:111 -p 111:111/udp -p 2049:2049 -p 2049:2049/udp --privileged -e ONECLIENT_PROVIDER_HOST -e ONECLIENT_ACCESS_TOKEN=-v /tmp/o:/root/.local/share/oneclient onedata/oneclient:17.06.0-beta6

Unfortunately, that command refused to work with 137:138/udp -p 137:138/udp:

docker: Error response from daemon: driver failed programming external connectivity on endpoint stupefied_mayer (c6c3530b390fb0aac6abefb43fac9c6d7c339cbc0fb130a62ce1fc0bd8edbb1b): Error starting userland proxy: Bind for 0.0.0.0:137: unexpected error (Failure EADDRINUSE).

without those ports command works as expected.

I performed few tests from the command line:

SMB

smbutil view -g //localhost
Share                                           Type    Comments
-------------------------------
onedata                                         Disk
IPC$                                            Pipe    IPC Service (Samba 4.3.11-Ubuntu)

mount_smbfs //Guest@localhost/onedata /tmp/onedata

One can list the content of the mounted directory but its quite slow, much slower than from within the container.

NFS

The command used to list shares, shows nothing:

showmount -e localhost

From within the container we can see that nfs is working:

root@7eb5eced30b4:/tmp# showmount -e 172.17.0.2
Export list for 172.17.0.2:
/mnt/oneclient *

We will continue to investigate, but atm. I'm out of ideas. This feature was developed before native docker for macos was released. With docker running on virtualbox it works well.

@bkryza @luman75 please be aware of this issue.