Chat update
Alright, that was pretty exciting. After posting this story on Ajaxian we got a lot of traffic.
I have to first set the record straight, while I have helped in passing on chat infection until today I hadn’t done any work on it directly. Eddie wrote the whole thing following railsconf at the bar and on the plane ride back to the office. Amazingly, the app handled most of the ajaxian traffic on a single mongrel process!
Today after catching wind of the story on ajaxian, I jumped in and got monit, mongrel_cluster, apache and pound setup this evening to load balance with help from Eddie and Warren on apache. Warren had good experience with configuring and monitoring apache after getting Statisfy to scale. The “the easy way” wasn’t an option because we already had apache 2.0.x running instead of apache 2.2.
The site was under fairly heavy load so, at first we decided to just get mongrel back up and apache backup. Looking at the apache configs, we first disabled apaches keepalive, and that seemed to help, but mongrel continued to crash. Turns out we had an older version of fastthread on the box, upgrading to 1.0 helped a little. Next we got mod_status setup (thanks to Warren) and that helped us figure out that we should have some keep alive.
The app is crapping out once in awhile, which could be an app problem or could be a ruby/mongrel problem (need to do some more digging). For now, the solution is to run a few mongrels and keep them all running using monit.
When I get a little more energy and run things by Eddie and Warren, I may blog more about the details of how we have things setup…
Recent Comments