[ZODB-Dev] Clustering ZEO Server with drbd

[Jim Fulton]

On May 29, 2006, at 12:02 PM, Christian Theune wrote:

> garthpl at periweb.com wrote:
>> If you setup drbd and heartbeat , the spare will take over the  
>> floating ip from the zeo server and restart the service. drbd is  
>> simply doing a block level replication of the zeo server file  
>> system where the zeo data is stored. The question is what files  
>> needs to be replicated so that when the the service is restarted  
>> on the spare, all the files needed by the zeo server to operate  
>> are there? I'm trying to ensure from a system architecture  
>> standpoint I'm not missing anything.
>
> When using FileStorage, you'd need the Data.fs.* files, except  
> Data.fs.lock.
>
> In the end, this might not be exactly HA because you'll have a  
> period of time that isn't covered and may stretch for quite a  
> while. (But I don't know about the HA terminology anyway...)

I think this is relative.  It's still higher availability than having  
to resort to a
a backup (if you have one).


> (Re)starting the ZEO server when it is needed might take a while on  
> large databases if the index file happens to be broken.

Note that with ZRS, you have to manually restart a secondary to have  
it become a primary.  With
ZRS, you also have to arrange for your clients to switch servers. An  
advantage of the drdb approach
is that the new server has the same IP as the old.

Someday, I hope we'll implement a quorum-based replication strategy  
for ZRS
so that we don't have to manually restart servers.

Another advantage of ZRS is that it supports geographic replication.
Your secondaries can be offsite.  We will also be adding an option for
remote backups using the ZRS recovery protocol without actually  
running a
full-bore secondary.  We are starting to use this for some of our  
customers
now.

Jim

--
Jim Fulton			mailto:jim at zope.com		Python Powered!
CTO 				(540) 361-1714			http://www.python.org
Zope Corporation	http://www.zope.com		http://www.zope.org