Yesterday was not just the first day for my new web site. It was also the day I rolled out a major project at Autoweb. The project was, surprise surprise, deployment of our new postgres replication system using you guessed it Slony-I. sure we had done lots of testing so one might say why would you think there could be any problems. Unfortunately no matter how much testing we do, its nothing like between the hours of seven and eight and the morning when the server was going to get 3000 users slamming it. Good old postgres ran like a champ! I sat there watching all day expecting something to go wrong but know nothing of any excitement the server just kept running and running queries and the slave servers stayed fully replicated. So I think I’m in a go-ahead and give this project a cautious thumbs up will get another couple weeks to make sure nothing breaks but I think we got one right.
I guess that means tomorrow I can start on the next phase of the project true hot failover. Using something like heartbeat I need to detect when the primary postgres server goes down and rolling everything over to the secondary without the users ever knowing. Things get a lot crazier here and I certainly don’t want my script to accidentally roll over to the slave when the masters still is the master.
All this reminds me with my recent work on postgres and Slony-I replication I have some scripts are really need to try to get put into the CVS tree. One in particular is really slick at finding objects not being replicated thank you Brian for the help with that one.