tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kenichi Masuko" <zod...@ss.iij4u.or.jp>
Subject mod_jk2 unix channel leak?
Date Tue, 10 Aug 2004 12:51:35 GMT
Hello.

We are running apache-2.0.50 + mod_jk2 + tomcat-5.0.27
on RedHat AS 3 with j2sdk-1_4_2_04.
We are using mod_jk2 that I have compiled
from jakarta-tomcat-5.0.27-src.tar.gz.
We have a situation that more than 1000 users request
a page at the exact same time
(they want more than 2000 but anyway).

We first used socket channel in mod_jk2 to connect
apache and tomcat which was functioning fine
but we weren't satisfied with the performance.
Then, we tried in-process JNI but we couldn't
make it function properly after a week (ashamed).
So, we are left with unix channel.

With unix channel, some of the clients (IE) start to freeze
after a while.  We have
"[error] ajp13.service() ajpGetReply recoverable error 3"
in the apache error log and nothing relevant in the tomcat log.
With netstat --unix, we observe the number of
unix domain sockets increases as apache accepts new requests,
but we also start to see several unix domain sockets which stay
CONNECTING status for a long time (15 min)
after the number of unix domain socets approaches 200.
We think those lingering unix domain sockets
are causing client freezes.

The client freezes don't happen when
MAX_CLIENTS in httpd.conf is less than 200,
but this decreases the overall performance by 30%.
We tried to increase UNIX_HASH_SIZE
in include/net/af_unix.h but nothing changes
(no sure if this is relevant).

Has anyone come across similar problems?

Kenichi


---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org


Mime
View raw message