Questions about Queue Time
We recently added the X-Request-Start header to track queue time, and we're seeing very high queue time values. I've got a few questions to debug this issue:
1) On server clocks - There are a few allusions to server clock synchronization, but it isn't clear which servers must be synchronized. We have two app servers, each running Apache and Passenger, behind a separate load balancer. Since the header is added by each app server, I don't think the clocks need to be precisely synchronized? Or must they be in sync?
2) Does the queue time include the time waiting for Passenger to
spin up a new instance? That would explain some of our issues,
since we have many low-traffic applications.
Comments are currently closed for this discussion. You can start a new one.
Support Staff 2 Posted by Bill Kayser on 17 May, 2010 05:46 PM
1) That is correct. As long as the machine running the rpm agent and the machine adding the header are the same, it's not an issue. You should be fine.
2) Yes that is true since the header is added before handing the request off to Passenger. I don't know for 100% because we haven't done this kind of analysis but I would assume that you are running into a warmup delay. It would be interesting to see what happens if you set a minimum pool size that's comfortably high in one of these applications.
Bill Kayser resolved this discussion on 17 May, 2010 05:46 PM.