When you say that 10% fail immediately, do you mean that 10% fail in less than a second? What is the failure indication? This sounds a lot like a problem on your server, perhaps a shared resource exhaustion -- connections, memory, file descriptors, etc.
I see the similar latency from here in San Francisco, on a weekend--- say 80% < 0 seconds, 5% < 1 second, 4% < 2 seconds, 1% < 3 seconds, and none beyond 3 seconds. During a weekday, I could see how latency might increase somewhat beyond this for very short periods. I don't see any test tweets failing, with curl or with a Scala program, after hundreds of POSTs. Are you seeing requests fail that take longer than 8 seconds? Also, when the latency increases to 8 seconds, can you run a ping to www.twitter.com and post the summary information at the end of the run? Where, geographically, are your servers located? On Sun, May 9, 2010 at 4:23 AM, Tjaap <jdmeij...@gmail.com> wrote: > > Hi, sorry I didn't get back to you earlier. I was only monitoring this > thread: > http://code.google.com/p/twitter-api/issues/detail?id=1543 > > Right now about 10% of our post fails immediately. Latency is between > 0.7s and 3s, with peaks to 7 or 8 seconds. > We are not logging tcp dumps at the moment, but I will see what I can > do. I also will try to get some more data from our logs to see when > this is happening the most. > > Thanks Ryan and John for looking into it! >