[ZODB-Dev] ZODB + RelStorage FailOver

[Shane Hathaway]

Maxence DUNNEWIND wrote:
> I'm working on some zope based cluster. The network schema can be found
> at http://people.dunnewind.net/maxence/cluster3.png .
> All the database part (the red/pink parts) are managed with postgresql
> (with replication).

The relationships between the databases in the diagram are unclear.  Let 
me take a guess at what it means.  There is one logical database but 
three database servers.  The solid arrows represent synchronous 
replication and the dashed arrows represent asynchronous replication. 
All commits eventually get reflected throughout the cluster.

If that's right, I don't see how it can actually work correctly with 
most ZODB applications (especially Plone and anything that uses ZCatalog 
or BTrees).  It will produce massive data corruption.  OTOH, maybe I 
didn't understand the diagram correctly.

Incidentally, I don't think it's wise to push RelStorage beyond simple 
slave replication (one master with multiple slaves) at this time.  I 
think multi-master replication is possible, but would require careful 
thought, trade-offs, and site-specific configuration.

> My problem is : I would that each zope server uses mainly one pgsql
> server (like on the picture). Anyway, if one of the database server
> fails, then I would than zope/zodb/relstorage switch to another database
> server, and go back to his server when it will be back.
> 
> Any idea to make that with zodb/relstorage ? 
> At the moment, the only idea I have is to use an external check/watchdog
> service which change the config file and reload zope in case of database
> problem.

You could achieve this with iptables or a load balancer like Linux 
Virtual Server, but again, be sure your replication strategy makes sense 
before going down this road.

Shane