[hobbit] hobbit-clients.cfg

[Henrik Stoerner]

On Thu, Jul 06, 2006 at 10:52:42AM -0400, Sean Hennessey wrote:
> Say I have for example
> 
> HOST=pdcd20-ic1
>  PROC sm_server 1 1 RED
> HOST=pdcd20-ic2
>  PROC sm_server 2 2 RED
> 
> Pdcdd20-ic1 sends its update and it has 5 sm_server process, can you tell me
> the flow that happens?
> 
> Does it take the hostname as sent (pdcd20-ic1) and then look that host up in
> the clients.cfg file to see it should have only 1 sm_server or does it look
> at all the PROC directives first and then look at the host?

Mostly it's the second way you describe, but it's actually a two-step process.

hobbitd_client parses the hobbit-clients.cfg file, and stores it in
an internal data-structure - really just a linked list with a parsed
copy of the rules. I.e. there are the strings or regular expressions 
to match hostnames/testnames etc against, the thresholds and so on. At
this point, there is no relation to any particular host, it's just a 
list of the rules in hobbit-clients.cfg.

When a client message arrives, the rule-list is scanned for PROC rules
that match the particulars for this host, and each line in the "ps"
output is tested against the process-name that the rule is about.
While doing this, it merely counts how many matches are found in the
"ps" output for each of the relevant PROC rules. So one line from the
ps output can add a count to several PROC rules, not just one.

Finally, all of the relevant PROC rules are examined, and the actual
process count is compared to the thresholds listed in the rule. Based on
this, the "procs" status color is decided, and the "procs" status text
is generated with info about the rules that failed or succeeded.


So in your example, it would first count the number of "sm_server"
processes, and both of the PROC rules would get a count of 5 active 
processes matching the rule. So when the "procs" status is generated,
you should actually see two rules listed as "red" - one saying that
hobbit found 5 sm_server processes while expecting just 1, and the
other saying it found 5 sm_server processes while expecting only 2.


So the problem with your proposal is that by the time the final step
happens - when hobbit goes through all of the PROC rules to see which
ones has the right count of active processes - by then any association
with a specific process name has been lost. So there is no way that 
Hobbit knows that one rule comes before another, and relates to the
same process. In your example configuration, they both happen to
look for an "sm_server" process; we *could* choose to make this special
setup for rules that look for the exact same processname. But what if
you wrote the rules as

   HOST=bla
      PROC %[st]m_server 1 1
   HOST=bla
      PROC sm_server 2 2

and there is 1 sm_server processes active ? The two patterns are
different, so how can hobbit tell that it should use the first one
and ignore the second one ? And what should it do if there is a
tm_server process running ? Use only the first rule and ignore the
second?


> Great product by the way.

After all :-))

Thanks,
Henrik