Re: monit monitor.h p.y validate.c

[Jan-Henrik Haukeland]

Rory Toma <address@hidden> writes:

> Log message:
>       Make the dependency not matter on the ordering in monitrc,
>       things should only get validated once.
>       Please test.

Looks like you nearly nailed it :-) Except, that the start order is
reversed. If you have a dependency A->B then A is always started
before B allthough it should be the other way around I think. I have
the following dependency:

apache -> sendmail

and get this when I run ./monit with no processes started

[CET Nov 10 17:42:26] 'apache' is not running.
[CET Nov 10 17:42:26] start: (apache) /usr/local/apache/bin/http
[CET Nov 10 17:42:26] start: (sendmail) /home/hauk/bin/tst

Also if I run 'monit stop' stop sendmail is called twice:

[CET Nov 10 17:41:09] stop: (sendmail) /home/hauk/bin/tst
[CET Nov 10 17:41:09] stop: (apache) /usr/local/apache/bin/http
[CET Nov 10 17:41:09] stop: (sendmail) /home/hauk/bin/tst

Calling 'monit restart' doesn't work properly, it depend in the order
I have written the entries into the monitrc file:

If I have:

 check sendmail (..) 
 check apache (..) depend sendmail

I get this, i.e. on-off-on for sendmail:

[CET Nov 10 17:45:47] stop: (sendmail) /home/hauk/bin/tst
[CET Nov 10 17:45:51] stop: (apache) /usr/local/apache/bin/http
[CET Nov 10 17:45:57] start: (sendmail) /home/hauk/bin/tst
[CET Nov 10 17:45:59] stop: (sendmail) /home/hauk/bin/tst
[CET Nov 10 17:46:01] start: (apache) /usr/local/apache/bin/http
[CET Nov 10 17:46:01] start: (sendmail) /home/hauk/bin/tst

and if I have this order

 check apache (..) depend sendmail
 check sendmail (..) 

I get this which is slightly better, the stop order is fine but the
start order is reversed.

[CET Nov 10 17:45:47] stop: (sendmail) /home/hauk/bin/tst
[CET Nov 10 17:45:51] stop: (apache) /usr/local/apache/bin/http
[CET Nov 10 17:46:01] start: (apache) /usr/local/apache/bin/http
[CET Nov 10 17:46:01] start: (sendmail) /home/hauk/bin/tst


In the depend validate function I think you must test for loops and
validate the process when you backtrack to get the correct start/stop
order. Something like this (only a draft and not tested, and
p->visited should be set to FALSE in p.y when monit starts):

void validate() {

  ..
  for (p= processlist; p; p= p->next) {

   if(p->visited) 
      continue;
   else {
    if (p->dependantlist != NULL) {
      depend_validate(p);
    } else {
       if(!do_not_validate(p))
        do_validate(p);
    }
   }

  }
 
 /* Reset the visit markers */
 for(processlist) set all processes p->visited= FALSE;

}

/**
 * Walk the process-list and validate each process
 */
void depend_validate(Process_T p) {

  p->visited= TRUE;

  if(p->dependantlist) {

    Process_T dp;
    Dependant_T d;

    /* Now check dependants */
    for (d= p->dependantlist; d; d= p->dependantlist->next) {
        if(d->visited) continue;
        d->visited= TRUE; // The dependant object must also have a visited 
marker
        dp= get_process(d->dependant);
        if(dp->visited) { 
            error("Got depenency loop"); exit(1); 
        }
        depend_validate(dp);
    }

  }
    
  /* This is executed when backtracking from the recursive call stack */
  if(!do_not_validate(p))
      do_validate(p);
  
}




-- 
Jan-Henrik Haukeland