openstack team mailing list archive

[Sandy Walsh <sandy.walsh@xxxxxxxxxxxxx>]

Hey!

(sorry for the top-posting, crappy web client)

There is a periodic task already in the compute manager that can handle this:
https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L3021

There seems to be some recent (to me) changes in the manager now wrt the resource_tracker.py and stats.py files about how this information gets relayed. Now it seems it only goes to the db, but previously it was sent to a fanout queue that the schedulers could use. 

Regardless, this is done at a high enough level that it doesn't really care about the underlying virt layer, so long at the virt layer supports the get_available_resource() method.

https://github.com/openstack/nova/blob/master/nova/compute/resource_tracker.py#L152
https://github.com/openstack/nova/blob/master/nova/virt/xenapi/driver.py#L392
https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L2209

I'd add a hook in there do what we want with this data. Write it to the db, send it over the wire, whatever. If there is additional information required, it should go in this dictionary (or we should define a format for extensions to it).

The --periodic_interval value is meant to be the "fastest tick" approach and the individual methods have to deal with how many multiples of the base tick it should use. So you can have different data reported at different intervals.

Now, the question of polling vs. pushing shouldn't really matter if the sampling rate is predetermined. We can push when the sample is taken or we can read from some other store from an external process ... but the sampling should only be done in one place, once. 

Hope I answered your question? If not, just repeat it in another way and I'll try again :)

-S



________________________________________
From: openstack-bounces+sandy.walsh=rackspace.com@xxxxxxxxxxxxxxxxxxx [openstack-bounces+sandy.walsh=rackspace.com@xxxxxxxxxxxxxxxxxxx] on behalf of Eoghan Glynn [eglynn@xxxxxxxxxx]
Sent: Wednesday, November 07, 2012 4:32 PM
To: OpenStack Development Mailing List
Cc: openstack@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Openstack] [openstack-dev] [metering][ceilometer] Unified Instrumentation, Metering, Monitoring ...

> Here's a first pass at a proposal for unifying StackTach/Ceilometer
> and other instrumentation/metering/monitoring efforts.
>
> It's v1, so bend, spindle, mutilate as needed ... but send feedback!
>
> http://wiki.openstack.org/UnifiedInstrumentationMetering

Thanks for putting this together Sandy,

We were debating on IRC (#heat) earlier the merits of moving the
ceilometer "emission" logic into the services, e.g. directly into the
nova-compute node. At first sight, this seemed to be what you were
getting at with the suggestion:

 "Remove the Compute service that Ceilometer uses and integrate the
  existing fanout compute notifications into the data collected by the
  workers. There's no need for yet-another-worker."

While this could be feasible for measurements driven directly by
notifications, I'm struggling with the idea of moving say the libvirt
polling out of the ceilometer compute agent, as this seems to leak too
many monitoring-related concerns directly into nova (cadence of polling,
semantics of libvirt stats reported etc.).

So I just wanted to clarify whether the type of low level unification
you're proposing includes both push & pull (i.e. notification & polling)
or whether you mainly had just former in mind when it comes to ceilometer.

Cheers,
Eoghan

_______________________________________________
Mailing list: https://launchpad.net/~openstack
Post to     : openstack@xxxxxxxxxxxxxxxxxxx
Unsubscribe : https://launchpad.net/~openstack
More help   : https://help.launchpad.net/ListHelp