> There is no extra AV,firewall, just the Windows firewall. Even if it's one
> of these things, it shouldn't freeze up the whole process.
Can you get a backtrace?
I don't know what you mean by backtrace. I tried traceroute back to client
and it completed without problem.This problem is specific to IE. Firefox has no
problem. I tried to put other images in that Javascript code and come out with
the same result. I tried to remove QoS and other Microsoft network
service/client thing but still the same.
> I already tried
> Win32DisableAcceptEX. It helps but Apache become bloated and unstable. If I
> hit F5 (refresh) fast enough, I can make its memory go up 10-20MB/s. By the
> way, that MaxMemFree thing doesn't work.
Once or for every request?
MaxMemFree only applies to pool memory, so maybe the allocation isn't
in a pool. Are you using any additional modules?
I meant I hit F5 continuously like 5 times/sec and the memory will go up at
that rate. If I hit it slower like once per second, the memory will go up too
but at slower rate. That page I tried was from Tomcat though. So the leak might
be in mod_jk. The leak is higher with Win32DisableAcceptEX.
> I still ask the ASF people, if possible,
> please give us more worker process in mpm_winnt.
Why would that help here?
Another thing I just notice is the delay image loading also occur in direct
access to Tomcat but it never lock up Tomcat. For Apache, the delay is a little
longer (than Tomcat) and the server is not processing new request. I tried
telnet in. Apache accept my connection. I typed in "get /" and it stuck there
until image loading is finish.
So I think if we have extra worker process, it might pick up and serve the
request while the first one blocked. It's just my imagination though. I don't
know the way requests are distributed to workers and 20 seconds (blocked time)
is a long time too.
Anan
