Postgres 6.5 Is Fantastic! - Mailing list pgsql-sql
From | Wayne Piekarski |
---|---|
Subject | Postgres 6.5 Is Fantastic! |
Date | |
Msg-id | 199907240352.NAA13113@helpdesk.senet.com.au Whole thread Raw |
List | pgsql-sql |
Hi everyone! You may remember me as the one who was having lots of trouble using Postgres 6.4.2 with spinlock, BTP_CHAIN, and other miscellaneous problems. I tried patches and other tricks but everything was running really badly and we constantly had to intervene to keep the database running. Well, last Sunday I made the decision to upgrade to Postgres 6.5 - we had done some extensive thrash testing and we were very impressed with how it handled the load. The machine was a lot more idle, the disks were being used less, and so on, when compared to similar testing performed on a 6.4.2 database. Also, we managed to reproduce our spinlock problem under 6.4.2 but it never appeared on 6.5, probably due to the really cool MVCC concurrency code. So we did a pg_dump and reload into 6.5, which ran very smoothly - especially considering we have lots of functions, plpgsql procedures, triggers, rules, views, along with a few million rows in our tables. We restarted our software, and everything worked flawlessly. (Well almost..) It even worked using the libpq from 6.4, although we recompiled with the 6.5 version just in case. Everything worked right out of the box, which is good because we were worried about compatiblity problems with our programs running under MVCC. We had one problem with the reload, and that was the problem where the backend would get SIGSEGV when we tried to load in our plpgsql functions, which I've written another email about... The only thing we had to change were a few GROUP BY clauses in our queries, the parser in 6.4.2 was a bit slack in checking the arguments and making sure you did a group by on all columns not in an aggregate, and 6.5 didn't like some of the queries - we fixed them up and that wasn't a problem any more. The next day, everything ran like a complete dream. The machine was idle almost all day (compared to very busy on the CPU and disks) even during times when Postgres was doing lots of queries. In the past, when we ran really large queries during times when the machine was loaded down, we would always get either BTP_CHAIN or a spinlock problem - this is no more! We can thrash the machine to our hearts content and it always performs brilliantly without any one else even noticing! The postmaster has been running for a week now without requiring any form of human intervention, compared to once per day under 6.4.2. As you can imagine this is also good :) As of now, we've only found one problem with the plpgsql functions, but apart from that, I am *really* happy with this version of Postgres. With 6.4.2 I was having serious doubts about the reliability of the code, it just wasn't working for us - we were almost ready to give up on it and go to Oracle or something which I can imagine has its own set of problems and issues as well. 6.5 on the other hand, especially with the MVCC support, is absolutely awesome and my faith in Postgres has been restored. This is definitely the first commercial quality release that I could recommend to others. (By commercial quality, I mean reliable enough that you can thrash it real hard and it won't ever fall over). 6.4 and predecessors used to work fine but only under lightly loaded conditions and would sometimes break with wierd error messages. So for anyone who is using anything less than 6.5, you should not walk, but *RUN* to your local mirror site and grab a copy of 6.5 and install immediately! It is absolutely awesome .... Probably the most exciting part about it is the MVCC stuff, it puts Postgres up there in the big leagues and you don't have to lock whole tables when doing writes, this makes a huge difference. Also the snapshot backups are really cool too - my project runs 24 hours per day and it can't be shut down to perform backups so this is a real plus. I'd just like to offer my congratulations and thanks to all the developers of Postgres for your efforts and work - it is truly a great program and I hope to be able to help more with its development and support. [I hope this information is useful to the developers, I think someone said they wanted some feedback on 6.5] MVCC! Regards, Wayne ------------------------------------------------------------------------------ Wayne Piekarski Tel: (08) 8221 5221 Research & Development Manager Fax: (08) 8221 5220 SE Network Access Pty Ltd Mob: 0407 395 889 222 Grote Street Email: wayne@senet.com.au Adelaide SA 5000 WWW: http://www.senet.com.au