On 13.03.2009 17:50, SQ wrote:
Rainer Jung-3 wrote:
Just to make sure, we are talking about the same kind of observation:
could you please describe independently, how the observed problem looks
like in your case?
In development, the developers are getting other people pages. So user1
requests pageA and gets user2's pageB.
In production, we don't get user
input, but the probe on the load balancer is not getting the response it is
looking for, so it thinks the machine its checking is down. The probe is
called serverlive.jsp. Here is the accesslog entry during the problem (13
being the primary LB, 14 the backup):
xxx.xxx.xxx.14 - - [12/Mar/2009:23:09:53 -0500] "GET /serverlive.jsp " 200
13
xxx.xxx.xxx.13 - - [12/Mar/2009:23:09:54 -0500] "GET /serverlive.jsp " 200
13
xxx.xxx.xxx.14 - - [12/Mar/2009:23:09:59 -0500] "GET /serverlive.jsp " 200
13
xxx.xxx.xxx.13 - - [12/Mar/2009:23:09:59 -0500] "GET /serverlive.jsp " 200
13
xxx.xxx.xxx.14 - - [12/Mar/2009:23:10:04 -0500] "GET /serverlive.jsp " 503
1070
xxx.xxx.xxx.13 - - [12/Mar/2009:23:10:05 -0500] "GET /serverlive.jsp " 503
1070
xxx.xxx.xxx.14 - - [12/Mar/2009:23:10:10 -0500] "GET /serverlive.jsp " 503
1070
xxx.xxx.xxx.13 - - [12/Mar/2009:23:10:10 -0500] "GET /serverlive.jsp " 503
1070
xxx.xxx.xxx.14 - - [12/Mar/2009:23:11:00 -0500] "GET /serverlive.jsp " 503
1070
xxx.xxx.xxx.14 - - [12/Mar/2009:23:12:34 -0500] "GET /serverlive.jsp " 503
997
xxx.xxx.xxx.13 - - [12/Mar/2009:23:11:46 -0500] "GET /serverlive.jsp " 503
997
xxx.xxx.xxx.13 - - [12/Mar/2009:23:12:31 -0500] "GET /serverlive.jsp " 503
997
xxx.xxx.xxx.13 - - [12/Mar/2009:23:11:01 -0500] "GET /serverlive.jsp " 503
997
I guess you mean the lines with the 503 are the bad responses? But those
do not indicate, that the probe gets back the page requested by someone
else, it shows that the web server or Tomcat throw an HTTP error, namely
503. In this case I would guess, that mod_jk detected an error and put
th enode into error status. You should check your mod_jk log file. It
might also be good to temporarily activate the access log of Tomcat too,
in order to check, whether the 503 already came from there or not.
I would expect the develop observation and this one are two different
things.
I'll try and get some log entries from development.
Rainer Jung-3 wrote:
Since you see the problem with mod_jk2 and with mod_jk I somehow doubt,
that it comes form mod_jk (but hey, I'm involved in mod_jk development,
so that might simply be defense.
This is the main reason I posted here. If I'm indeed seeing the same
problem as the others here, then my case may disprove the mod_jk theory. Or
perhaps the issue resides in both my versions?
Do both (mod_jk and mod_jk2 show the problems A=develop and B=probes?
Rainer Jung-3 wrote:
What is obvious, your Tomcat is *very* outdated. You are using a no
longer supported major version (5.0) and with 5.0 you are using a very
old minor version.
If you have any chance, upgrade your Tomcat.
Yes, I know. I'll see what I can do.
Rainer Jung-3 wrote:
Apart from that: what else can you tell about the problem? Are there log
entries either from mod_jk, Apache httpd or Tomcat associated with these
events? Would you be able to snoop traffic between httpd and Tomcat and
between httpd and the clients?
As indicated above: if the system using mod_jk logs status code 503 in
the access log (and the 503 is not in the Tomcat access log), it is
*very* likely, that mod_jk writes something to its JkLogFile. Set
JkLogLevel to info (but info message alone are not relevant; when you
get a 503 it should log some error and interesting info messages at the
same time).
We haven't been seen any errors, in any logs. I can go through the logs and
compare them, and then compare those findings between the enviroments. Not
sure whats involved in snooping traffic. I can look into that as well.
Rainer Jung-3 wrote:
Where did you get your mod_jk from? How was it build?
Not sure the answer to that. Both were installed by other people, who
either don't recall their orgins, or are no longer employed here. I'm
working on building the 1.2.27 from source right now. We're x86, not sparc,
by the way.
OK. For Solaris x86 we never provided bins (I think), so someone might
have built them. Under Solaris you might run into some build troubles,
in case you are using a Sun provided httpd. Sun often compiles it with
the Sun compiler and there is a slight chance, that a gcc compiled
mod_jk will crash with a Sun compiled httpd. The Sun compiler is free
though. I'm just mentioning this, so you know that it would be best if
the compiler used for httpd and used for mod_jk are the same or at least
close to each other.
Thanks for your help. Please let me know of anything else I can provide. I
will make updates as new information comes up.
Regards,
Rainer
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
