[openstack-dev] [Ceilometer] Aggregation discussion

Jay Pipes jaypipes at gmail.com
Mon Jan 13 15:51:57 UTC 2014


On Mon, 2014-01-13 at 13:39 +0400, Nadya Privalova wrote:
> Jay,
> 
> 
> Thanks for comments!
> 
> The question you raised was discussed several times within Ceilometer
> team but as I understand there is no official resolution yet. 
> 
> I agree with you that statistics' collection is the main Ceilometer's
> goal. But at the same time there should be a way to visualize the
> result of Ceilometer's work.
> 
> My opinion is that the current set of data queries (get samples, get
> statistics) is the minimum set and it's ok to keep it as a part of
> Ceilometer's functionality. We need it at least for UI.

Yes, I've seen the (excellent) wireframes that Jarda and others have
been working on, and I can certainly see where the feature request comes
from.

> So, my proposal is to make this existing queries faster. Looks like
> our vision are the same :)

Certainly :)

>         Pre-calculate when? :) During processing of samples, or during
>         some
>         periodic job?
> 
> 
> It should be a periodic job, right.

I'm actually not so sure. If you spread the work of updating the
aggregate into the (multi-processed) collectors, then updating the
aggregates become less resource-intensive.

It's the same idea behind putting a SQL trigger on a table in an RDBMS
that updates a rolling aggregate. Doing the update work one little piece
at a time can be more efficient than doing a large update on a periodic
job -- particularly when, as is the case here, the long-running periodic
update job would need to take locks on the base fact tables.

> 
>         The term "aggregate" really just means a generic grouping or
>         summarization. If you are looking for a term that represents
>         the
>         rules/heuristics for maintaining rolling calculations, perhaps
>         the term
>         "report" is better?
> 
> 
> Hmm, I think that 'aggregate' is ok. In "my terminology" an aggregate
> is a ready set of statistics for concrete meter, period and query.
> Anyway, will think about it.

Sure, not a big deal. Was just throwing out ideas, really. "Aggregate"
is fine.

Best,
-jay





More information about the OpenStack-dev mailing list