| Summary: | crash in ap_log_rerror() from jk_log_to_file() | ||
|---|---|---|---|
| Product: | Tomcat Connectors | Reporter: | Steve Alexander <sca> |
| Component: | Common | Assignee: | Tomcat Developers Mailing List <dev> |
| Status: | CLOSED DUPLICATE | ||
| Severity: | normal | ||
| Priority: | P2 | ||
| Version: | unspecified | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Linux | ||
| Attachments: | apr.h | ||
Generally this works. I assume you are using the worker MPM and a 32Bit Linux? If the lock could not be initialized correctly, we would expect at least one of the following messages with levels: mod_jk: could not init JK log lock in child (ERROR) mod_jk: could not create jk_log_lock (CRITICAL) mod_jk: Could not set permissions on jk_log_lock; check User and Group directives (CRITICAL) Any of these in you mod_jk-Log? Just to make sure, could you also give the exact messages from the failed lock/unlock. Concerning ap_log_rerror: you are right, I will correct to ap_log_error for the upcoming 1.2.16. As always: any chance to test with 2.0.58 and Subversion HEAD of mod_jk? Finally: Could you provide the following things form inside your Apache home: bin/httpd -V include/apr.h include/apr.h Created attachment 18541 [details]
apr.h
What I have realized now is that these messages are only coming out in between "graceful restart requested, doing restart" and "Apache configured -- resuming normal operations" and only if a request is in progress during the graceful restart. A normal 'apachectl graceful' does not produce this. My ignorance of APR is vast since I only got sucked into this yesterday, but since jk_log_lock is a global mutex, perhaps it is being cleaned up during the restart by global_mutex_cleanup()? I do not see any of the errors you list in the log. [Tue Jun 27 13:19:41 2006] [notice] Graceful restart requested, doing restart [Tue Jun 27 13:19:41 2006] [error] (22)Invalid argument: apr_global_mutex_lock(jk_log_lock) failed [Tue Jun 27 13:19:41 2006] [error] (22)Invalid argument: apr_global_mutex_lock(jk_log_lock) failed [Tue Jun 27 13:19:41 2006] [error] (22)Invalid argument: apr_global_mutex_lock(jk_log_lock) failed [Tue Jun 27 13:19:41 2006] [error] (22)Invalid argument: apr_global_mutex_lock(jk_log_lock) failed [Tue Jun 27 13:19:41 2006] [error] (22)Invalid argument: apr_global_mutex_lock(jk_log_lock) failed ... ... [Tue Jun 27 13:19:42 2006] [notice] Apache configured -- resuming normal operations ############ httpd -V ############ [root@zm9 logs]# /usr/sbin/httpd -V Server version: Apache/2.0.52 Server built: Dec 15 2005 04:09:14 Server's Module Magic Number: 20020903:9 Architecture: 32-bit Server compiled with.... -D APACHE_MPM_DIR="server/mpm/prefork" -D APR_HAS_SENDFILE -D APR_HAS_MMAP -D APR_HAVE_IPV6 (IPv4-mapped addresses enabled) -D APR_USE_SYSVSEM_SERIALIZE -D APR_USE_PTHREAD_SERIALIZE -D SINGLE_LISTEN_UNSERIALIZED_ACCEPT -D APR_HAS_OTHER_CHILD -D AP_HAVE_RELIABLE_PIPED_LOGS -D HTTPD_ROOT="/etc/httpd" -D SUEXEC_BIN="/usr/sbin/suexec" -D DEFAULT_PIDLOG="logs/httpd.pid" -D DEFAULT_SCOREBOARD="logs/apache_runtime_status" -D DEFAULT_LOCKFILE="logs/accept.lock" -D DEFAULT_ERRORLOG="logs/error_log" -D AP_TYPES_CONFIG_FILE="conf/mime.types" -D SERVER_CONFIG_FILE="conf/httpd.conf" apr.h added as an attachment. If the problem is only occuring during graceful restart it's most likely BZ 39834. I noticed a mutex error when reproducing 39834 to. Should be fixed by Mladen in Subversion HEAD and in the soon to come 1.2.16. *** This bug has been marked as a duplicate of 39834 *** Move a couple of fixed JK issues from resolved to closed. |
mod_jk-1.2.15, httpd-2.0.52 with RH ES4 patches applied running on RH ES4 update 3. I haven't figured out what the root cause of this is: jk_log_lock doesn't appear to get initialized, so both the lock and the unlock fail with EINVAL (presumably coming from pthread_mutex_{lock,unlock} ()). The unlock error case calls ap_log_rerror() with a NULL request pointer, but this routine dereferences the pointer immediately to pick up the server information, causing a core dump. I changed it to ap_log_error() and that fixes the crash. I still don't know why the mutex is not initialized; I can only assume that the post config hook is not being called or is failing for some reason. Let me know if any further information is needed.