[ZODB-Dev] DBTab and ZEO Failover

sean.upton at uniontrib.com sean.upton at uniontrib.com
Thu Apr 17 11:21:14 EDT 2003


There are two ways to make this work with stock-ZEO (DBTab aside):

1 - Assume smart client; ZEO 2 introduced the ability to have a client
choose from a list of multiple ZSS servers to connect to. If the first ZSS
dies, the ZEO client tries the second, etc.  If I understand correctly, this
is necessary to support ZC's ZRS product, but will support failover across
any two ZEO storage servers (replication is a whole other matter entirely).
You have the challenge of making sure you have two concurrently running,
identical ZSS nodes.

2 - Assume dumb clients, and use smart servers via IP+resource takeover.
The secondary ZSS takes over the IP of the first and fires up the backup ZSS
process.  After a timeout counter, the ZEO client retries/reconnects to the
IP address of the ZSS (thinking there is just one), but gets a connection on
the secondary.

Replication options:
- Low-level (file copy, network raid, LVM snapshots, etc)
- Storage-level (DirectoryStorage replica tool, for example) - suitable for
hot-standby, but ZSS on secondary must be offline while replication is
happening, so you need IP takeover on the ZSS cluster (i.e. program like
heartbeat on secondary takes over the IP and fires up the ZSS).
- Transaction-level (ZRS) - multiple live servers

Sean

-----Original Message-----
From: Lou Scalpati [mailto:lou at hi-privacy.net]
Sent: Thursday, April 17, 2003 9:28 AM
To: zodb-dev at zope.org
Subject: [ZODB-Dev] DBTab and ZEO Failover


I have zope 2.6.1 and Zeo 2.02  and DBTab 1.1

My question is about zeo fallback connections.  I noticed in the dbtab.conf
file it shows fallbacks for a mount for ZRS.  Does this also apply to other
zeo servers without ZRS?
 I have 2 database servers that I am running zeo server on.  If for some
reason the primary dies will the client time out and go to the secondary if
it is defined like it shows for ZRS?  or is that a function of additional
code that ZRS introduces?  If that will work is there a maximum number of
fallback servers it can address?  Is this method reliable or am I better off
using another failover strategy such as ipvs in front of the zeo servers.


_______________________________________________
For more information about ZODB, see the ZODB Wiki:
http://www.zope.org/Wikis/ZODB/

ZODB-Dev mailing list  -  ZODB-Dev at zope.org
http://mail.zope.org/mailman/listinfo/zodb-dev



More information about the ZODB-Dev mailing list