Re: monit doesn't poll all services at once?

[Jan-Henrik Haukeland]

On 9. okt. 2005, at 05.56, Joe S. wrote:

> I am wondering if its possible to monitor all services at once,  
> rather than one by one, I like to monitor all services remotely,  
> but when it goes down the list, it eventually takes around 30  
> minutes to check if all services are online before it goes back to  
> #1 to check.


Oj, that sounds much. Is it possible for you to post your monitrc  
file here so we can have a closer look.

> is it possible for it instead going down the list,  but to poll  
> everything, all at once, and continue from there?
>
> I know it will eat alot of resources, but its needed to speed  
> things up a bit.

It can be a solution to fire of each monitoring in it's own thread.  
But this may not even solve the problem in the long run. The stuff  
that (probably) take time is network monitoring since monit has a  
default 5 sec timeout on each connect, read and write. A better  
solution than using threads may be to use level triggered event  
notification (see e.g. man epoll on Linux and kqueue on BSD and  
http://www.monkey.org/~provos/libevent/ for a general library). We  
could also fork of a monitor process per service monitored like  
apache (1.x) does.

I do feel a chill going down my spine for the extra complexity any of  
these solutions will add :) But maybe we simply *must* rewrite monit  
to scale better if more users are experience the same problem. (I  
only monitor around 10 services and have never had this problem.)

--
Jan-Henrik Haukeland
Mobil +47 97141255