Re: How Robust is eDirectory?
joe.doupnik <joe.doupnik <at> oucs.ox.ac.uk>
2011-11-16 09:56:54 GMT
Your concern is not paranoid at all, just the opposite.
I don't have good answers to your questions, but I do have an
observation or two. They amount to this case: master replica holder
suffered a problem and stopped communicating with other servers in the
ring. Time passes while I travel, and eventually I discovered the
situation. I rebooted the master and life resumed with no complaints.
Were I in your shoes I would thoroughly test the delayed power up
situation in the lab. I would not ignore DSBK completely, but I would
get the backups, move them to the down servers on a regular basis and
restore them when a server is started (finesse with lan connectivity,
ifdown eth0 style, would be required to avoid exposing the old replica).
The deep down problem, as I see it, is an old replica has pending
updates which upon a server restart it tries to complete and they
conflict with current conditions. To stop that I would remove a replica
holder from the ring, shutdown the server, and later re-add it when
ready. This should eliminate the pending transaction problem.
On 16/11/2011 03:25, Scott Campbell wrote:
> Just how robust is the 8.8.x incarnation of eDir?
> We are creating a formal DR plan and as part of that we are replicating VMware Guests to a server in another
city. These replicated guests will be cold and bought up only to test or if we have a major disaster.
> To minimise the impact on the email part of the services, we will have a couple of servers running on the
remote host which will have a r/w replica of the root of our tree. In the event of a disaster the DR sites r/w
replica will be newer than the replicated guests which we would want to power up. Infact any of the servers
that we want to bring up will have inconsistent views of the state of the tree.
> So, is eDirectory smart enough to identify the server that has the most current state of the replica and