httpd-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "King Holger (CI/AFP2)" <Holger.K...@de.bosch.com>
Subject [users@httpd] Apache2 2.2.22 | mod_proxy_ajp | Workers in ERROR state on single long running request | Bug-fix
Date Wed, 04 Jul 2012 12:24:26 GMT
Dear Apache2 community,

with an important bug fix in "mod_proxy_ajp" of Apache 2.2.22 (see http://www.apache.org/dist/httpd/CHANGES_2.2.22),
a problem marking an AJP-worker (used for communication between Apache and Tomcat) in error
state when a single long running request occurred, should have gone.

Important: the documentation and change log tells nothing about the condition and pre-requisites
that have to be setup, to allow that bug fix to work. Is it right, that within a "BalancerMember"
configuration directive, the "ping" value for cping/cping communication (details can be found
here: http://httpd.apache.org/docs/2.2/mod/mod_proxy.html#balancermember) should be set to
a value different compared to the default value (=0)?

This conclusion might be right, when taking a look on the commit comment and the implementation
of the bug fix in Apache 2.2.22 (http://mail-archives.apache.org/mod_mbox/httpd-cvs/201201.mbox/%3C20120104201529.54EA823888D2@eris.apache.org%3E):
...
/* If we had a successful cping/cpong and then a timeout
+         * we assume it is a request that cause a back-end timeout,
+         * but doesn't affect the whole worker.
+         */
+        if (APR_STATUS_IS_TIMEUP(status) && conn->worker->ping_timeout_set)
{
+            return HTTP_GATEWAY_TIME_OUT;
+        }

Do you agree?

Our experience is: when having not specified a "ping" configuration property in a BalancerMember
directive like:
BalancerMember ajp://rb-wcmstc1.xx.xxxxx.com:8009 loadfactor=100 retry=10 route=rb-wcmstc1
connectiontimeout=5 timeout=30

the bug fix does not work and Apache 2.2.22 will set the Tomcat AJP worker in ERROR state
when a requests takes longer than the configured timeouts, e.g. the reply took longer than
30 seconds here.

My question is:
- is this ping property a must have to let this bug fix work properly?
- which conditions lead to an ERROR-state switch?

Mit freundlichen Grüßen / Best regards,
Holger King
CI/AFP2

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org


Mime
View raw message