[Moosefs-users] have mfsmount 'use' a specified network interface?

[Travis <tra...@tr...>]

Im running moosefs on my notebook, the latest in the git repository

MFS version 1.6.11
FUSE library version: 2.7.4



I have set up a 'tun' interface and given that an ip address of 
10.0.0.1, and have the msfmaster and chunkserver attach to this. The 
motivation being I have both wireless and wired network interfaces on 
the notebook and don't want the service to have any kind of dependency 
on on which ever one is currently active, but also, because I move 
between different places, different IP addresses on the LAN too.

the mfsmaster and one mfschunkserver process run on this 10.0.0.1 address

and then the mfsmount command in use is

mfsmount -H 10.0.0.1 /mnt/mfs


But it appears the mfsmount is choosing one of the other interfaces to 
connect 'from', such as eth0, or wlan0, and connecting from (what ever 
IP address is on that LAN at the time).

I noticed this as I had to enter the 192.168.x.x address into the 
mfsexports.cfg in addition to the 10.0.0.x one, which I added to allow 
it to mount for now.

The main issue is usually after suspending the notebook and coming back 
online (in a different network) the mfs mount seems to get lost

ls /mnt/mfs
ls: cannot access /mnt/mfs: Input/output error

So is there a way to make the mfsmount work on a specified interface 
that it attempts to use when connecting to its master and chunk servers?

Not that most people would run a complete self contained stack on the 
notebook right. My motivation is to have the services on the host and to 
have the mfsmount client inside one or more virtual machines, that also 
work on this tun interface, so as to create a shared file system for all 
the virtual machines. and in my road warrior setup the everything on one 
host works well.

But I wonder if it could come up in complex clusters or other 
multi-homed machines where a high performance network and a low 
performance management network were both available and the networks 
could some how route between each other, that could cause the data 
traffic from the mount to end up going over the lower performance 
management network or something .