Re: [PATCH] ICMP with optional host parameter

[Martin Pala]

Michel Marti wrote:
> Martin Pala wrote:
>
> > Thanks for patch, however i think you can reach the same result using 
> > either dependency on 'host' service or exec action. 
>
> Correct... However it complicates the monit-configuration a little bit. 
> In my opinion, the ICMP check is similar to other network tests (e.g. 
> "if failed host myhost port 1234 ..."). So it should aswell have a 
> "host" parameter.

You are rigth that the connection test could be used to check remote 
process even in monit 'process' service type container. This is however 
sort of special case (or side effect).

The idea is, that the process service type/container should allow to 
check all properties of the local process. One of such properties is 
that the proces is handling requests via appropriate socket. The socket 
is just transport layer and can be either unix (i.e. local only) or for 
example tcp/udp (which may be accessible from network).

So primarily the connection test in process service type should serve 
for local process testing - the host part of connection test should 
allow to check various virtual hostnames which the process may serve 
(e.g. apache virtual servers).

The remote host connection tests are preferred in host service type 
container.

ICMP echo on the other side, is not related to any local process - the 
response is handled by kernel => having icmp echo support directly in 
process container is not clean solution from my point of view ...

> > I think the process service container should be kept only for process 
> > properties (icmp testis is logicaly not related to the process).
>
> So is the "loadavg()" test but it can still be used for process services?!?

You are rigth again ;) We have currently no customisable "system" 
service type which can allow to group system-wide properties (such as 
load average, global memory and cpu usage, total process counts, etc.). 
This may change in the near future - the main part system service type 
is already here:

--
System 'unicorn'
  load average                      [0.48] [0.25] [0.14]
  cpu                               11.1%us 1.8%sy 0.0%wa
  memory usage                      91584 kB [17.7%]
  data collected                    Fri Jan 21 22:06:43 2005
--

Currently the only place where it is possible to define rule for load 
average is process service type. This allows to stop inimportant 
processes in the case that the load average exceeds the limit and start 
them again when it decreases (for example address@hidden ;)

When the system service container will be implemented, it may be 
possible to define related rules here ... (but maybe this rule is on the 
edge of process and system service type since process is participating 
on system load).

Martin