Hello, I've messing about with Plack::Handler::Mongrel2 plugged into a Dancer application and obviously Mongrel2.
Under JMeter load testing hitting the same URL repeatedly I am getting into a situation where the front end requests start receiving a "java.net.SocketTimeoutException: Read timed out" error. The tests are doing a repeated request "GET http://localhost:6767/app/polamaolaolaolaoaloalaoalaoalaoalaoalaoalaaoalaoaalaoalaoalaoa?j=8&t=1 " Mongrel will always produce the following error for a failed request: [INFO] (src/mongrel2.c:343) Starting Mongrel2/1.7.5. Copyright (C) Zed A. Shaw. Licensed BSD. [INFO] (src/server.c:272) Starting server on port 6767 [INFO] (src/task/fd.c:151) MAX limits.fdtask_stack=102400 [INFO] (src/superpoll.c:102) Allowing for 256 hot and 768 idle file descriptors (dividend was 4) [INFO] (src/handler.c:209) MAX allowing limits.handler_targets=128 [INFO] (src/handler.c:285) Binding handler PUSH socket tcp://127.0.0.1:9998with identity: E80576A8-AC0B-11DF-A841-3D4975AD5E34 [INFO] (src/handler.c:311) Binding listener SUB socket tcp://127.0.0.1:9999subscribed to: D807E984-AC0B-11DF-979C-3C4975AD5E34 [INFO] (src/control.c:401) Setting up control socket in at ipc://run/control *[ERROR] (src/register.c:214: errno: Resource temporarily unavailable) Nothing registered under id 0.* *[ERROR] (src/register.c:199: errno: None) Invalid FD given for exists check * * * Mongrel finish's the tcp connection to the http client after the zmq message is sent to the Plank back end: http client -> Mongrel 6767 Mongrel -> plank 9998 *This is when the problem occurs and client http connection is finished* Plank -> Mongrel 9999 Mongrel -> Client 'high port' sudo tcpdump -n -i lo -s 158 -p tcp port 6767 or tcp port 9999 or tcp port 9998 18:11:11.991287 IP 127.0.0.1.52686 > 127.0.0.1.6767: Flags [P.], seq 30808:31037, ack 79381, win 384, options [nop,nop,TS val 41369964 ecr 41369963], length 229 18:11:11.991470 IP 127.0.0.1.9998 > 127.0.0.1.55744: Flags [P.], seq 17675:18211, ack 3, win 256, options [nop,nop,TS val 41369964 ecr 41369955], length 536 18:11:11.991480 IP 127.0.0.1.55744 > 127.0.0.1.9998: Flags [.], ack 18211, win 384, options [nop,nop,TS val 41369964 ecr 41369959], length 0 18:11:12.001689 IP 127.0.0.1.52686 > 127.0.0.1.6767: Flags *[F.]*, seq 31037, ack 79381, win 384, options [nop,nop,TS val 41369966 ecr 41369963], length 0 18:11:12.001759 IP 127.0.0.1.6767 > 127.0.0.1.52686: Flags *[F.]*, seq 79381, ack 31038, win 384, options [nop,nop,TS val 41369966 ecr 41369964], length 0 18:11:12.001772 IP 127.0.0.1.52686 > 127.0.0.1.6767: Flags [.], ack 79382, win 384, options [nop,nop,TS val 41369966 ecr 41369966], length 0 18:11:12.010615 IP 127.0.0.1.50171 > 127.0.0.1.9999: Flags [P.], seq 21123:21763, ack 3, win 257, options [nop,nop,TS val 41369968 ecr 41369956], length 640 18:11:12.010655 IP 127.0.0.1.9999 > 127.0.0.1.50171: Flags [.], ack 21763, win 384, options [nop,nop,TS val 41369968 ecr 41369960], length 0 Whereas it normally looks like this: 18:11:11.968949 IP 127.0.0.1.52686 > 127.0.0.1.6767: Flags [P.], seq 29663:29892, ack 76441, win 384, options [nop,nop,TS val 41369958 ecr 41369958], length 229 18:11:11.969095 IP 127.0.0.1.9998 > 127.0.0.1.55742: Flags [P.], seq 17137:17673, ack 3, win 256, options [nop,nop,TS val 41369958 ecr 41369954], length 536 18:11:11.971890 IP 127.0.0.1.50169 > 127.0.0.1.9999: Flags [P.], seq 20483:21123, ack 3, win 257, options [nop,nop,TS val 41369959 ecr 41369955], length 640 18:11:11.972006 IP 127.0.0.1.6767 > 127.0.0.1.52686: Flags [P.], seq 76441:77029, ack 29892, win 384, options [nop,nop,TS val 41369959 ecr 41369958], length 588 18:11:11.973493 IP 127.0.0.1.52686 > 127.0.0.1.6767: Flags [P.], seq 29892:30121, ack 77029, win 384, options [nop,nop,TS val 41369959 ecr 41369959], length 229 The only thing of note I found in Plack::Handler::Mongrel2 at the same time as the errors were undefined variable errors, but they were all related to the METHOD=JSON disconnect message being sent out after this issue has occured. I can increase the the frequency of the errors by increasing the debug logging on the Perl/Plack side of things so I believe it's something in the back end but I'm at a loss at the moment as to what could be causing it. It happens with or without multiple threads or a single thread at each http client and plack levels. Any ideas on where to look next would be appreciated. Cheers, Matt
