[Ocfs2-users] OCFS2 over VPN

Sunil Mushran sunil.mushran at oracle.com
Wed Nov 5 11:01:45 PST 2008


Increase idle timeout to 30 secs atleast, if not 60 secs.

Lorenzo Milesi wrote:
> Hi
>
> I've set up a DRBD8 Active/Active configuration over two remote places,
> connected via 2 bonded (Open)VPNs on 2 ADSL connection, OCFS2 fs. 
> The two systems are both Debian Etch, kernel 2.6.22-4-686, DRBD v8.0.11,
> OCFS2 1.3.3.
>
> I know the solution is quite hazardous, but so far has been working
> better than I expected. The big issue is that it's very unstable,
> because it's suffering network leaks. Now that I've bonded two VPNs this
> problem should be less critical.
> I have also enabled QoS to give high priority to OCFS2 and DRBD ports.
> Sometimes it can go straight without problems for weeks, sometimes it
> fails every other hour. 
>
> I realized OCFS2 is very sensible to network leaks. I tried raising
> values to the maximum but still quite often I get:
>
> server-2 kernel: o2net: connection to node server-1 (num 1) at
> 192.168.0.4:7777 has been idle for 20.0 seconds, shutting it down.
> server-2 kernel: (0,0):o2net_idle_timer:1419 here are some times that
> might help debug the situation: (tmr 1225887463.671970 now 1225887
> 483.668529 dr 1225887463.671962 adv 1225887463.671974:1225887463.671975
> func (57b88650:500) 1225887463.671971:1225887463.671972)
> server-2 kernel: o2net: no longer connected to node server-1 (num 1) at
> 192.168.0.4:7777
>
> My conf for o2cb is:
> O2CB_HEARTBEAT_THRESHOLD=50
> O2CB_IDLE_TIMEOUT_MS=20000
>
> Is there any other possible tuning I can do?
> Is it possible to also decide which node gets panic'ed?
>
> Thanks...
>   




More information about the Ocfs2-users mailing list