Обсуждение: Clustering and replication options

I am looking for some information about clustering and replication
options for postgresql.

I am aware of pgcluster, but have been unable to find anyone willing
to share details about actually using it in a production environment.
That's a little disconcerting. Is pgcluster not really ready for full
production use right now?

I really would like to setup a multi-master cluster of databases, and
pgcluster seems just perfect for the task. I will be testing it for
myself in a few weeks.

I was hoping I could solicit some advice, or really any information,
about other possible clustering or replication methods. I know slony-
I is available, and that it is used b the .org registry, I believe,
as well as other places. It seems to be pretty mature as well.
However, I'm concerned about the master to multi-slave model. For
databases with mostly reads and fewer writes, that model seems good.
But, it seems like replication lag, not to mention the added
complexity of splitting read and write operations between two
database connections, could cause problems for applications with a
large number of both reads and writes. Any lag would potentially
allow stale data to be read from a slave after it had been updated on
the master.

Can someone point me at some good references for either slony-I or
pgcluster, other than their actual home pages? I've spent more time
reading and analyzing the pgcluster configuration than slony-I, as my
needs include support for frequent, multiple writes.

Thanks,
Greg

On Wednesday June 22 2005 2:16 am, Gregory Youngblood wrote:
> I am looking for some information about clustering and
> replication options for postgresql.

Gregory,

FWIW, I've used Slony 1.0.5 for 10-15 db cluster migrations,
usually from 7.3.4 clusters on one box to 7.4.6 clusters on
another (slony is asyncronous master-slave).  That has typically
involved running slony for anywhere from a few hours to a few
days.  Where a dump and reload has required several hours of
downtime, slony has allowed us to cutover in a matter of a
minute or three.  We've also successfully used a perl version of
dbmirror, also asyncronous master-slave.

I'd be very interested in hearing what syncronous multi-master
replication solutions you find for pgsql.  I haven't kept up,
but I'm not aware of any that are still alive.  The one using
Spread and Ensemble ("Postgres R"?) appears dormant as of 7.2.

Ed