Jan-Henrik Haukeland wrote:
On 28. jun. 2006, at 00.24, Martin Pala wrote:
Jan-Henrik Haukeland wrote:
On 27. jun. 2006, at 15.36, Martin Pala wrote:
Description of the patch:
When some action is requested via http, the http thread just sets
the "doaction" task on the given service thread and wakes up the
main thread. The main thread then performs normal testing cycle and
when it founds the doaction flag on any service, it performs the
requested action.
From the patch
+++ validate.c 27 Jun 2006 11:38:58 -0000
+ if(s->doaction != ACTION_IGNORE) {
+ control_service(s->name, s->doaction);
+ s->doaction = ACTION_IGNORE;
+ } else if(s->monitor && !check_skip(s) && !check_timeout(s)) {
I think it may be better to run a new for-loop first and only check
and act on the doaction flag or merge the if-else. In the code
above, services with this flag set is excluded from the standard
test (in the else part), which probably is not what we want.
Good topic :) I'll try to explain the cause - maybe i'm wrong:
It was based on the assumption, that when the action is scheduled by
the http thread (such as stop/restart/start) it has higher precedence
for particular service then the monitoring and it is not needed to
check the service in the same cycle.
For example, when the service is to be restarted, it will be
initializing after the restart and when we will try monitor it in the
same cycle, it may not be fully initialized yet and monit can produce
false alarms, thus i think it is better to perform the restart and
postpone the monitoring on the next cycle.
Ah yes, initializing. For restart and start (after stop) you need to
wait one poll cycle anyway, otherwise we must add more complicated code
and test if and when a service is ready to be tested. I see your point
and agree that it is easier and simpler to postpone the service check
until the next cycle. I don't remember and don't bother to check the
code now, but this is probably the case also when turning monitoring on
after unmonitoring?
In the unmonitor->monitor case makes sense to check in the same cycle.
The patch works the same with all action tasks (i.e. postpone the
monitoring on next cycle).
Maybe we can live with common behavior for all actions including
unmonitor->monitor?
Ok, my unconditional +1 for your patch then :) Good work!
Thanks :)
Martin
_______________________________________________
monit-dev mailing list
[email protected]
http://lists.nongnu.org/mailman/listinfo/monit-dev
