Restart problems

[Stanley Gambarin <ga...@OpenMarket.com>]

	Greeting, Apache developers.

	I currently maintain the mod_fastcgi (http://www.fastcgi.com) module
for the Apache web server and we have been experiencing certain problems.
Allow me to extrapolate on the cause of the problem:
	- in our init() function, we spawn off another process which performs
certain computation, including starting other processes using fork()/exec()
mechanism.  When the clear_pool() function is invoked upon restart of the
server, it calls the free_proc_chain() function.  The latter function sends
a SIGTERM signal to our main process, sleeps for 3 seconds and then sends
SIGKILL signal.  Even though the main process catches SIGTERM, sometimes
it does not have time to kill all fork()ed/exec()ed applications, which 
results in the parentless apps being run on the machine, just wasting 
resources.  
	- my suggestion is to introduce another kill_condition, which would
make free_proc_chain function just issue a SIGTERM, without issuing SIGKILL
later.  I think the above might also be related to the 3 second delay in the
execution of the certain CGI programs.

	The diffs are attached below:
*** alloc.h     Fri Feb 28 17:41:14 1997
--- alloc.h.old Fri Feb 28 17:40:41 1997
***************
*** 227,234 ****
   * to call in the child, and an argument to be passed to the function.
   */
       
! enum kill_conditions { kill_never, kill_always, kill_after_timeout, 
just_wait,
!                      kill_only_once };
  
  int spawn_child_err (pool *, void (*)(void *), void *,
                 enum kill_conditions, FILE **pipe_in, FILE **pipe_out,
--- 227,233 ----
   * to call in the child, and an argument to be passed to the function.
   */
       
! enum kill_conditions { kill_never, kill_always, kill_after_timeout, 
just_wait};
  
  int spawn_child_err (pool *, void (*)(void *), void *,
                 enum kill_conditions, FILE **pipe_in, FILE **pipe_out,

*** alloc.c     Fri Feb 28 17:42:27 1997
--- alloc.c.old Fri Feb 28 17:40:36 1997
***************
*** 1021,1028 ****
  #endif
  
    for (p = procs; p; p = p->next) {
!     if ((p->kill_how == kill_after_timeout) ||
!       (p->kill_how == kill_only_once)) {
        /* Subprocess may be dead already.  Only need the timeout if not. */
        if (kill (p->pid, SIGTERM) != -1)
        need_timeout = 1;
--- 1021,1027 ----
  #endif
  
    for (p = procs; p; p = p->next) {
!     if (p->kill_how == kill_after_timeout) {
        /* Subprocess may be dead already.  Only need the timeout if not. */
        if (kill (p->pid, SIGTERM) != -1)
        need_timeout = 1;

						Thanks, Stanley.

Re: Restart problems

[Dean Gaudet <dg...@arctic.org>]

An old one.  +1 on this going into HEAD. 

Dean

On Fri, 28 Feb 1997, Stanley Gambarin wrote:

> 	Greeting, Apache developers.
> 
> 	I currently maintain the mod_fastcgi (http://www.fastcgi.com) module
> for the Apache web server and we have been experiencing certain problems.
> Allow me to extrapolate on the cause of the problem:
> 	- in our init() function, we spawn off another process which performs
> certain computation, including starting other processes using fork()/exec()
> mechanism.  When the clear_pool() function is invoked upon restart of the
> server, it calls the free_proc_chain() function.  The latter function sends
> a SIGTERM signal to our main process, sleeps for 3 seconds and then sends
> SIGKILL signal.  Even though the main process catches SIGTERM, sometimes
> it does not have time to kill all fork()ed/exec()ed applications, which 
> results in the parentless apps being run on the machine, just wasting 
> resources.  
> 	- my suggestion is to introduce another kill_condition, which would
> make free_proc_chain function just issue a SIGTERM, without issuing SIGKILL
> later.  I think the above might also be related to the 3 second delay in the
> execution of the certain CGI programs.
> 
> 	The diffs are attached below:
> *** alloc.h     Fri Feb 28 17:41:14 1997
> --- alloc.h.old Fri Feb 28 17:40:41 1997
> ***************
> *** 227,234 ****
>    * to call in the child, and an argument to be passed to the function.
>    */
>        
> ! enum kill_conditions { kill_never, kill_always, kill_after_timeout, 
> just_wait,
> !                      kill_only_once };
>   
>   int spawn_child_err (pool *, void (*)(void *), void *,
>                  enum kill_conditions, FILE **pipe_in, FILE **pipe_out,
> --- 227,233 ----
>    * to call in the child, and an argument to be passed to the function.
>    */
>        
> ! enum kill_conditions { kill_never, kill_always, kill_after_timeout, 
> just_wait};
>   
>   int spawn_child_err (pool *, void (*)(void *), void *,
>                  enum kill_conditions, FILE **pipe_in, FILE **pipe_out,
> 
> *** alloc.c     Fri Feb 28 17:42:27 1997
> --- alloc.c.old Fri Feb 28 17:40:36 1997
> ***************
> *** 1021,1028 ****
>   #endif
>   
>     for (p = procs; p; p = p->next) {
> !     if ((p->kill_how == kill_after_timeout) ||
> !       (p->kill_how == kill_only_once)) {
>         /* Subprocess may be dead already.  Only need the timeout if not. */
>         if (kill (p->pid, SIGTERM) != -1)
>         need_timeout = 1;
> --- 1021,1027 ----
>   #endif
>   
>     for (p = procs; p; p = p->next) {
> !     if (p->kill_how == kill_after_timeout) {
>         /* Subprocess may be dead already.  Only need the timeout if not. */
>         if (kill (p->pid, SIGTERM) != -1)
>         need_timeout = 1;
> 
> 						Thanks, Stanley.
> 
> 
>