On Fri, Jul 30, 2010 at 02:57:15PM -0400, Phil Dupont wrote: > Thanks, I'm certainly on the same page with you... I've certainly run into > this kind of stuff when there's a run away process and the like. > > That said, I've run the production log of ruby through their log analysis > tool (request-log-analyzer). That tool will chew on the logs and give you > the min, max, and average load times for any transaction that occurs in > Ruby... thing is... I don't see any query with a taking over a few seconds. > > > Essentially, from what I can tell, apache is responding within seconds... or > if it's not, it's not tossing any errors to the access or error logs.
OK, but if you see haproxy report 504, you must check haproxy's logs to find what requests experienced those errors, when, after how much time. Maybe you'll find that you never see them in apache logs simply because the apache process dies after running a loop and cannot log. It's common not to find errors on the component which causes them. That's the reason why you must first isolate those errors from haproxy's logs, and then search them on the other components. If you see that haproxy says that apache took 40 seconds while apache says it took 2 seconds on the same requests, then you know it's somewhere between the two. But as long as you don't have anything to compare, you're in the total darkness. Willy
