On Thu, 2007-11-08 at 10:13 +0530, Asankha C. Perera wrote: > Hi Will > > 2007-11-07 17:54:35,065 [128.8.10.71-z.glue.umd.edu] > > [HttpServerWorker-15] ERROR Axis2Sender Unexpected error during > > sending message out > > java.lang.IllegalStateException: I/O reactor has been shut down > Whats interesting is what caused the IO reactor to shut down, and any > log messages that indicate that error is what I would need to figure > things out.
Asankha, You can prevent the I/O reactor from shutting itself down in case of an unchecked exception by using a custom IOReactorExceptionHandler http://jakarta.apache.org/httpcomponents/httpcomponents-core/httpcore-nio/apidocs/org/apache/http/nio/reactor/IOReactorExceptionHandler.html I know this will not fix the cause of the problem but can at least help ensure the I/O reactor keeps running if the exception is deemed non-fatal. Cheers Oleg > Also please confirm that you are using a Unix/Linux system > for load testing, and let me know the OS version, exact JDK version, any > OS tuning parameters you may have used (e.g. whats listed at the end of > http://wso2.org/library/1721) and most importantly information about > your client environment - i.e. if HTTP 1.0 or 1.1 (chunked encoding) is > used and if keepalives are enabled. I believe you are testing this with > the latest Synapse 1.1 codebase and not the 1.0 release. We are about to > release the 1.1 code early next week and thus thats what you should use > for any testing right now. > > The best I've tested so far is 1000 requests with concurrency level of > > 50. I tried tweaking the thread settings in synapse.properties file > > without any luck. Has synapse been tested against handling large > > number of concurrent requests? Any thoughts why I'm getting this error? > The default burned in settings for the NIO transport can manage well > over many hundreds of thousands of messages and even reach thousands of > concurrent connections - but provided you have properly tuned your OS! I > would recommend a Linux System with JDK 1.5.x with adequate OS tuning as > specified in the above link and a client that uses HTTP 1.1 with > keepalives to reach the peak levels. > > I would be happy to explain the reasons behind not using HTTP 1.0 or > keepalives for a performance test if you are interested. If you could > provide me with complete logs and the above details, I could try to > reproduce your scenario if I can to investigate further. However, for > now you do not need to think about overriding the default settings of > the NIO transport. > > asankha > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
