Eric Dalquist schrieb:
We are seeing a similar problem.
We have Apache 2.2.6 using prefork connecting to Tomcat 5.5.23 via mod_jk
1.2.25
We see infrequent issues with requests getting swapped from request to
Tomcat. After the first time it was reported we added Tomcat access logging
to try and help figure out the problem.
Here are logs from both Apache and Tomcat where UserB ends up seeing the
results of UserA's request for '/portal/content' for their request to
'/portal/Login'. You can also see that UserA doesn't actually get a full
response for their request to /portal/content.
Tomcat Logs:
0.0.0.0 - userA [22/May/2008:21:57:01 -0500] "GET /portal/redirect.jsp
HTTP/1.1" 302 - "-" "Mozilla/4.0" 4
0.0.0.0 - userA [22/May/2008:21:57:01 -0500] "GET /portal/content HTTP/1.1"
200 1154 "-" "Mozilla/4.0" 28
1.1.1.1 - userB [22/May/2008:21:57:03 -0500] "GET /portal/Login HTTP/1.1"
302 - "-" "Mozilla/5.0" 99
Apache Logs:
0.0.0.0 - userA [22/May/2008:21:57:01 -0500] "GET /portal/redirect.jsp
HTTP/1.1" 302 - "-" "Mozilla/4.0" 14895
0.0.0.0 - userA [22/May/2008:21:57:01 -0500] "GET /portal/content HTTP/1.1"
200 550 "-" "Mozilla/4.0" 9009
1.1.1.1 - userB [22/May/2008:21:57:03 -0500] "GET /portal/Login HTTP/1.1"
200 1154 "-" "Mozilla/5.0" 8946
We do not have the native connector deployed with Tomcat (we're using the
distribution from the TC website).
Do you get any info or above log messages in the JK log file close to
this events?
What is the last column in the log formats? Is it both %D? Then there is
also something else strange, namely execution times of 4, 28 and 99
milliseconds in the Tomcat log, and 14.8, 9 and 8.9 milliseconds on the
httpd side. So the first request takes much longer on the httpd side,
the next two seem to be quicker for httpd than for the backend, which is
impossible unless httpd/mod_jk terminate the request/response cycle
prematurely. In this case I would expect a message either in the JK log
or the httpd error log.
Do you have some efficient way to detect those errors (scripts comparing
the log files), or do you only detect them on user feedback? If you've
got scripts, you could also add %P and %{tid}P to the access log format,
which will provide process id and thread id, so we get an idea, if the
switch happens on a single thread.
Regards,
Rainer
---------------------------------------------------------------------
To start a new topic, e-mail: users@tomcat.apache.org
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
