Re: App performance degradation at scale
Nicklas Börjesson <[email protected]>
| Newsgroups | gmane.comp.python.cherrypy |
|---|---|
| Message-ID | <[email protected]> |
If you have a high consistent load, I presume that you have tests to evaluate performance. It is important that these tests doesn't involve any outside stuff, like memcached, but you have mockups instead. If that is the case, you can start isolating the issues by gradually replacing the code with mockups. You say you use a virtualized kernel, this is common of course, but since you mention it, do you mean that you use some AWS micro instance or other hosted virtualization solution? Because some of their cheaper hosts automatically scale down performance after you have reached a limit, and that limit can be different for network load(requests from outside) and CPU load(command line). //Nicklas On Wednesday, March 30, 2016 at 12:45:14 PM UTC+2, Vishal Shah wrote: > > FYI, > > This is how my config file looks like: > > [global] > server.socket_host = "0.0.0.0" > server.socket_port = 9010 > server.thread_pool = 25 > server.environment = "production" > > i am running 4 cherrypy processes across 2 servers. The peak load expected > is roughly 50 req/sec. > > Regards, > > Vishal. > > On Wednesday, March 30, 2016 at 4:09:35 PM UTC+5:30, Vishal Shah wrote: >> >> Hi Michael, >> >> Thanks for the reply. i have checked the memory on the server, and am not >> seeing much increase over time. Is there a way i can recycle cherrypy >> threads after x number of requests? Any other ideas? >> >> Regards, >> >> Vishal. >> >> On Wednesday, March 30, 2016 at 3:40:58 PM UTC+5:30, Michiel Overtoom >> wrote: >>> >>> >>> > On 2016-03-30, at 07:38, Vishal Shah <[email protected]> wrote: >>> > >>> > The app basically gets some data from memcache and a column store db >>> for every request. What i am noticing though, is that the performance keeps >>> degrading as the server gets more and more requests. It starts off with >>> roughly 50 ms per response, but after a few thousands of requests, the >>> performance becomes as slow as 30 seconds. >>> >>> That's odd. I have a similar setup and I don't see slower response times >>> after a few thousand requests. Could there be a memory leak in your >>> application? Have you monitored the memory usage over time of the Python >>> process which runs CherryPy? >>> >>> Greetings, >>> >>> -- You received this message because you are subscribed to the Google Groups "cherrypy-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to cherrypy-users+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/[email protected] To post to this group, send email to cherrypy-users-/JYPxA39Uh5TLH3MbocFF+G/[email protected] Visit this group at https://groups.google.com/group/cherrypy-users. For more options, visit https://groups.google.com/d/optout.