Logging in High Performance Systems: Design Hacking

With Christmas approaching, the regulars at the BWPUG only think of one thing...how in the world are the retail related servers we manage going to survive all the on-line ordering. One of the challenges of a busy database server is getting the amount of logging detail right. Logging is a good way to track down many types of performance issues, but when the system is seriously overloaded the overhead of logging can be too expensive.

Recently OmniTI's Theo Schossnagle submitted a PostgreSQL patch (http://archives.postgresql.org/pgsql-hackers/2011-11/msg01364.php) to suggest some alternate ways to handle logging in systems where writing every line to disk just isn't feasible. Since several of the other people who responded are regulars at the BWPUG, this month we're going to get together to talk about what it would take to make this specific part of the database better. What people in this situation want, how the code might be implemented, how it might be used...that's all up for grabs at this point. If you're interested in this topic and want to watch or get involved in how features are added to PostgreSQL, here's your chance.