[KLUG Members] Postgres Cluster
Adam Tauno Williams
adam at cisco-inc.com
Tue Jul 19 10:43:10 EDT 2005
> I wouldn't evan use the word cluster to describe a backup situation like
> this. It's splitting hairs a bit. I would only use the word cluster to
> describe a situation where each machine has live data ( such as the one
> described ) and each machine is accepting querries (unlike the one
> described). The application being used could be responsible for only
> performing update querries against a machine which would be a master.
> Clustering with multiple masters get's really sticky because of record
> locking. Man what a pain. I could not evan imagine trying to lock the
> same record on 1 or many replicants for a multiple master configuration.
Yep, agree. I've looked at MySQL and PostgreSQL's "cluster" support.
And I don't think any of then qualify as a "cluster" in the way most
people intuitively think of the term. They either user enourmous
amounts of RAM to keep data resident or they are simply replicants.
> I use unidata and would definitely not call what I do a cluster.
> Unidata keeps a transaction log. I periodically (every 15 minutes)
> rsync the transaction log over to a backup unidata server. If there is
> a major catastrophe on the main server, a script can be run to play the
> transacitions into the backup from the last night. The maximinum amount
> of transactions I can lose is 15 minutes worth.
Right. Our Informix database keeps transactions logs and automatically
spools a log to tape when it complete. The logs + a full backup can
restore/replay the entire database back to the point when the
transaction log was closed.
> I like talking word semantics. So I ask the question is ldap with
> replicant servers an ldap "cluster". I'd say yes, but no-one I have
> ever heard has called ldap servers with replicants a cluster. Why???
Because they are not a cluster, they are a master (or set of masters)
and a hive of replicants. A cluster is a peer-to-peer construct where
nodes have functional equality or at least the potential for functional
equality.
> Does the dictionary say that a cluster must be on a data repository to
> some xxx standard?? I would guess not. Language can be fun.
> Unfortunately, I have never attempted to cluster postgres servers, so an
> answer to your direct question is absent from my experience.
> > Hey all.. I was looking at the Postgres documentation and their definition
> > of a Database Cluster is different from what I am use to with MS SQL. The
> > definition in the manual basically states a cluster is multiple databases
> > run by a single instance of Postgres. A cluster in the terms of my thinking
> > is a setup of multiple SQL installs/machines that replicate from a Main
> > point incase one machine/install fails the next machine takes over until the
> > primary machine can be repaired and then replicate the data from the backup
> > machine before it becomes live again.. can postgres do this and if so can
> > someone point me to documentation on it.
I think so, take a look at "slony".
http://gborg.postgresql.org/project/slony1/projdisplay.php
http://candle.pha.pa.us/main/writings/pgsql/replication.pdf
Slony is "Slony-I is a "master to multiple slaves" replication system
with cascading and failover."
Still not what I'd call a cluster, but it gives you redundancy.
I think if you want a true cluster your options are: Oracle or DB2
More information about the Members
mailing list