Where are your servers located geographically? If you don't see an hourly correlation, do you see a strong time-of-day correlation or day-of-wee correlation?
Can you run a ping from your servers during a period of instability and send the summary at the end of the run? On Sat, May 8, 2010 at 3:00 PM, Naveen Ayyagari <nav...@getsocialscope.com> wrote: > I dont have tcp dumps available.. Ill try to get some captures later > this weekend.. > > Ill also run through the logs to see if they have any correlation with > top of the hour, but my gut instinct is they do not. > > Anywhere specific I should send tcp dumps, as they may contain > username and passwords of our users (we are still in the process of > switching over to OAuth)? Should I just XXX those out? > > > On May 8, 1:50 pm, Ryan Sarver <rsar...@twitter.com> wrote: >> Raj, Naveen, @tjaap, >> >> Do any of you still have tcp dumps of the calls you were making that were >> getting long timeouts? >> >> On Sat, May 8, 2010 at 12:08 AM, Naveen Ayyagari >> <nav...@getsocialscope.com>wrote: >> >> >> >> > We see the same huge latency and timeouts as well (our timeouts are >> > also at 30 seconds). >> >> > We running out of a US data center on multiple machines, we see this >> > issue on all if our servers. >> >> > I agree with @tjaap, would like to hear twitters reaction as well. >> >> > On May 7, 6:02 pm, Tjaap <jdmeij...@gmail.com> wrote: >> > > I'm seeing exactly the same: big latency and a lot of timeouts. When >> > > posting a tweet, Twitter sometimes does not return a response within >> > > 30 seconds. After my curl times out, the app has no way of knowing if >> > > the post made it or not. Sometimes it will appear on Twitter and >> > > sometimes it won't. This effectively breaks our app. Replicated the >> > > error on two different servers (in the Netherlands) and with different >> > > accounts. >> >> > > I would very much appreciate a reaction from Twitter on this. We are >> > > in the process of switching to OAuth, but I would like to know if I >> > > have to implement a workaround while we're on the old system. >> > > Thanks, >> >> > > @tjaap >