Nagios vs Juju monitoring

[Tom Haddon]

On 26/11/12 13:32, Gustavo Niemeyer wrote:
> On Fri, Nov 23, 2012 at 5:38 PM, Clint Byrum <clint at ubuntu.com> wrote:
>> Indeed, the charm just refreshes the entire nagios config on depart or
>> broken. I actually never really understood that departed meant two things
>> (admin removal or box down). There absolutely must be a way to distinguish
>> them for nagios or any monitoring system to be able to do anything sane.
> 
> This is indeed super error prone and, as we discussed in the meetings
> pre-UDS, the behavior has been changed in the Go port to become more
> reliable. The relation-departed hook is *only* called when the user
> kill the relation/service/unit for real. A box-down will not trigger
> it anymore. That may come back in the future as a specific hook.

Fwiw, we've worked around this by having a script that parses the output
of `juju status` to create a list of a hosts that we're connecting to in
Nagios (an external nagios instance, using a custom subordinate charm),
and we then copy that to an "approved" list, so that if the output of
`juju status` changes we're still checking for the list of approved
hosts. This is obviously working outside of the Juju framework, but we
need to do that currently to support multiple environments in any case.