Re: Apache creates Semaphore

[Hemant Chaudhary <hemantdude.chaudhary@xxxxxxxxx>]

Hi,

I tried with the patch of code you have asked for prefork.c and worker.c. But If I am closing apache gracefully then semaphore is not in shared location but if I am killing process then still semaphore is there in shared location.

Is there any way I get rid of semaphore because I want to run single process only.

Thanks

hemant

On Fri, Nov 3, 2017 at 2:56 AM, Yann Ylavic <ylavic.dev@xxxxxxxxx> wrote:

Hi Hemant,

On Thu, Nov 2, 2017 at 12:08 PM, Hemant Chaudhary
<hemantdude.chaudhary@gmail.com> wrote:
>
> For my product I need to run apache as single process. As httpd -X works for
> me. But the issue is to stop httpd -X, we need to kill process. While
> killing the process, semaphore exists in kernel directory.

As Luca said, the default semaphore mechanism (i.e. if not changed by
./configure) used by httpd is "sysvsem", and it indeed leaks a
descriptor when not exiting properly (or from debug mode).

The easier way to use another mechanism is the Mutex directive, for instance:
    Mutex pthread default
defined globally will use a mechanism which does not have this issue.

If you really want to stay with "sysvsem" mechanism, the following
patch will do it for the debug mode too:

Index: server/mpm/event/event.c
===================================================================
--- server/mpm/event/event.c    (revision 1814101)
+++ server/mpm/event/event.c    (working copy)
@@ -691,6 +691,7 @@ static void clean_child_exit(int code)

     if (one_process) {
         event_note_child_killed(/* slot */ 0, 0, 0);
+        atexit(apr_terminate);
     }

     exit(code);
Index: server/mpm/prefork/prefork.c
===================================================================
--- server/mpm/prefork/prefork.c        (revision 1814101)
+++ server/mpm/prefork/prefork.c        (working copy)
@@ -229,6 +229,7 @@ static void clean_child_exit(int code)

     if (one_process) {
         prefork_note_child_killed(/* slot */ 0, 0, 0);
+        atexit(apr_terminate);
     }

     ap_mpm_pod_close(my_bucket->pod);
Index: server/mpm/worker/worker.c
===================================================================
--- server/mpm/worker/worker.c  (revision 1814101)
+++ server/mpm/worker/worker.c  (working copy)
@@ -435,6 +435,7 @@ static void clean_child_exit(int code)

     if (one_process) {
         worker_note_child_killed(/* slot */ 0, 0, 0);
+        atexit(apr_terminate);
     }

     exit(code);
_

>
> Therefore I commented following lines in worker.c
>
> if ((rv = SAFE_ACCEPT((apr_snprintf(id, sizeof id, "%i",
> i),ap_proc_mutex_create(&all_buckets[i].mutex,NULL, AP_ACCEPT_MUTEX_TYPE,
> id, s, pconf, 0))))) {
>            ap_log_error(APLOG_MARK, APLOG_CRIT | level_flags, rv, (startup ?
> NULL : s),  "could not create accept mutex");
>                  return !OK;
>
>  After commenting above line, when I am starting httpd -X, it is giving
> signal 31 error.

This can't work, the mutex will be used later on so it will crash.
Thus this change is incomplete, and the complete one is probably more
complex than the changes proposed above.

>
> What should I change if I want to start apache in debug mode and not to
> create semaphore ?

I don't think suppressing this mutex (or any other process/global
mutex which your configuration may need, soon or later...) is wise.
You should go with avoiding the leak (as proposed above).

Regards,
Yann.

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