For those of you running postgresql with some kind of multi-server setup, how do you do load balancing and replication right now?
I'm running pgpool-ii but just tonight noticed some pretty serious discrepancies between the two backend postgresql instances so I am getting rid of pgpool for the time being.
The last place I worked used F5s for load balancing and Bucardo for replication. The F5s worked well enough, and I've become a huge fan of Bucardo. It has capabilities that I think blow Slony away. (Custom code, and pre-warmed slaves are two of the biggest differentiators.)
The new job is doing something that could best be described as very primitive sharding (completely unnecessary, given the demonstrated loads compared to the levels of concurrency PostgreSQL can support with minimal tuning), without replication. I'm eliminating the former, and will eventually be adding some HA capabilities, and probably also bringing Bucardo in.
I don't think SR is going to obsolete a connection pool any time soon -- you still need to direct traffic appropriately between master and slave(s). I've never used it, but I've consistently heard good things about pgBoucner as a connection pooler.
We use londiste (from SkyTools). It's excellent. From what I've read, londiste already plans to start offering streaming replication (vs. trigger-based replication) for the transport mechanism.
This is definitely big news for PostgreSQL. When I had to set up replication for one of the projects I was working on I was highly dssapointed with all the solutions out there. I'm really excited that this will be built in and it looks like it will be really simple to get set up and working.
16 comments
[ 2.8 ms ] story [ 46.4 ms ] threadWell, exactly this is planed for a future release: http://wiki.postgresql.org/wiki/Streaming_Replication#Future...
They running wordpress on mysql or something?
I'm running pgpool-ii but just tonight noticed some pretty serious discrepancies between the two backend postgresql instances so I am getting rid of pgpool for the time being.
Will this change make pgpool-ii obsolete?
https://developer.skype.com/SkypeGarage/DbProjects/PgBouncer
http://www.last.fm/user/Russ/journal/2008/02/21/zd_postgres_...
The new job is doing something that could best be described as very primitive sharding (completely unnecessary, given the demonstrated loads compared to the levels of concurrency PostgreSQL can support with minimal tuning), without replication. I'm eliminating the former, and will eventually be adding some HA capabilities, and probably also bringing Bucardo in.
I don't think SR is going to obsolete a connection pool any time soon -- you still need to direct traffic appropriately between master and slave(s). I've never used it, but I've consistently heard good things about pgBoucner as a connection pooler.
I have never been satisfied with a piece of software as I am with Postgres.
Fact: 11 year old version of postgres are still actively maintained.
I agree, yay!