From: Jan Zuchhold (Jan_at_Zuchhold.com)
Date: Sat 03 May 2003 - 17:40:09 BST
> > > > I use the vserver patch together with a Raid1+DRBD/Heartbeat failover
> > > > config. If the primary server fails all vservers are started on the
> > > > secondary within a few seconds. Besides the recently fixed bug in
> > > > ctx15/ctx16 this setup is 100% stable.
> > >
> Very interesting. Is there any write-up available for this solution? I am
> eager to learn more about the specifics.
I don't know one which describes the whole setup. There are howtos/docs on
DRBD+Heartbeat (try Google).
Basically my setup works like this:
BoxA and BoxB are monitoring eachother over ethernet (udp/ethX) and serial
A pair of servers (lets call them BoxA and BoxB) share two DRBD devices
(drbd0, drbd1), which map to a local lower level block device.
If both servers are up, one DRBD device is primary (=write access) on each
- drbd0: primary -> (maps to /dev/hdXX or /dev/mdX or whatever)
- drbd1: secondary
- drbd0: secondary
- drbd1: primary
All changes to a primary device are replicated to the other node.
(On most servers I use a local software raid1 device, which means that the
data exists on four physical disks).
/dev/drbd0 is mounted as /vservers-BoxA (if it is primary)
/dev/drbd1 is mounted as /vservers-BoxB (if it is primary)
in /vservers there is a symlink to /vservers-Box? for each vserver
1. Heartbeat on BoxB notices BoxA isn't there any longer and starts taking
2. DRBD (actually the datadisk script as specified in the Heartbeat resources
file) takes over drbd0 and mounts it:
- drbd0: primary
- drbd1: primary
Now both /vservers-BoxA and /vservers-BoxB are mounted on BoxB.
3. Heartbeat starts BoxA's Vservers on BoxB.
> I gather that one important point
> to consider is the fact that the vserver kernel doesn't support journaling
> file systems.
Are you talking about the precompiled kernel? I never used it.
I use both reiserfs and ext3 on drbd on raid1 with the vserver patch without