[openstack-dev] [Ceilometer] [TripleO] adding process/service monitoring

Łukasz Jernaś deejay1 at srem.org
Sat Feb 1 13:12:55 UTC 2014


On Tue, Jan 28, 2014 at 6:49 PM, Ladislav Smola <lsmola at redhat.com> wrote:
> Might be good to monitor it via SNMPD. As this daemon will be
> already running on each node. And I see it should be possible, though
> not very popular.
>
> Then it would be nice to have the data stored in Ceilometer, as
> it provides generic backed for storing samples and querying them.
> (would be nice to have history of those samples) It should be enough
> to sent it in correct format to notification bus and Ceilometer will store
> it.
> For now, Tuskar would just grab it from Ceilometer.
>
> The problem here is that every node can have different services running
> so you would have to write some smart inspector that would know what
> is running where. We have been talking about exposing these kind of
> information in Glance, so it would return you list of services for image.
> Then you would get list of nodes for image and you can poll them via SNMP.
> This could be probably inspector of central agent, same approach as for
> getting the baremetal metrics.

Hi,

I'm a bit new here, so please excuse me if I state the obvious or
speak gibberish.

One problem with using only SNMPD for querying might be that by
default it only exposes the running process and it's state, which
might not fully describe if it's working at all, as some processes
tend to look alive in the ps output but aren't doing anything or are
stuck at some code path. So in an ideal world you'd require some other
form a health check which isn't as easily exposed via snmpd conf
changes and some boilerplate code to expose that data via an OID.

As for writing a separate daemon for that, I was under the impressions
that ceilometer agents on hosts (clarification: with hosst I mean the
actual hardware running nova,glance,etc) would provide those
capabilities via pollsters/plugins instead of having other entities to
keep track of (the usual who monitors the monitoring process problem),
which then in turn might be exposed via SNMP for other tools to
collect if needed.

But if expectations are wrong feel free to correct me.

Have a nice day,
-- 
Łukasz [DeeJay1] Jernaś



More information about the OpenStack-dev mailing list