On Sat, 2018-11-17 at 10:47 -0700, Gary Gregory wrote: > Hi Oleg, > > I was able to reproduce the problem, the results are in the branch > ggregory-async-reverse-proxy-problem, folder > httpcore5/AsyncReverseProxyExample-run1. > > I edited the stock example AsyncReverseProxyExample and ran it from > Eclipse > and redirected output to a file which I then copied > to httpcore5/AsyncReverseProxyExample- > run1/AsyncReverseProxyExample.log. > > The edit is: > > final IOReactorConfig config = IOReactorConfig.custom() > .setSoTimeout(10, TimeUnit.SECONDS) > .setIoThreadCount(1) > .setSelectInterval(TimeValue.ofMilliseconds(10)) > .build(); > > I ran the test using JMeter like this: > > rmdir /q/s c:\test\jmeter-OneByteHttpCore & del /F > c:\test\OneByteHttpCore.csv & jmeter -n -t > C:\Users\ggregory\Downloads\OneByteHttpCore.jmx -l > c:\test\OneByteHttpCore.csv -e -o c:\test\jmeter-OneByteHttpCore > > The JMX test plan is in > httpcore5/AsyncReverseProxyExample-run1\OneByteHttpCore.jmx. >
Hi Gary I updated your branch, ran 1 mio requests against the proxy example without logging, ran 1 mio requests against the proxy example with logging, updated IOReactorConfig as described above, even though based on my understanding of how HttpCore works it should not matter, ran ran 1 mio requests against the proxy example without logging, grudgingly downloaded JMeter, installed JMeter, ran the test plan for 30 min, stopped, ran the same test plan for another 10 minutes. I am sorry but I am not able to reproduce the problem. From the description of the problem I somehow cannot shake off the feeling that you are testing against the old code without the recent fix. I am going to review the AbstractHttp1StreamDuplexer class again and will try to think of any remaining race conditions but I am pretty much out of ideas by now. Oleg --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
