[pcp] Derived Metrics and pmlogsummary.. ? Is it supposed to work?

Ken McDonell kenj at internode.on.net
Thu Sep 8 05:15:34 CDT 2011


On Thu, 2011-09-08 at 16:57 +1000, Nathan Scott wrote:
> ----- Original Message -----
> > ...
> > Why wouldn't it work in pmlogsummary?
> 
> I suspect its to do with the way pmlogsummary is using a
> fairly low level interface (see pmFetchArchive(3)) to scan
> through every data point in the archive with no changes (ie.
> no interpolation, rate conversion, etc) ... this scanning
> is all done upfront before we look at the requested metrics
> at all.  Ken might have a cunning plan to resolve this, but
> I suspect it wont be easy unfortunately.

Paul,

Nathan is correct ... derived metrics were intended for use with pmFetch
and pmFetchArchive is outside the scope of the design (much less the
implementation).

I've checked in the "cunning plan" cupboard, and can see a possible
tacky algorithm in the "hack" bin ... I need to take some time to
consider this matter.

> Might be worth considering alongside that other missing piece
> of log summary functionality - calculating the Nth percentile.
> That also would need some design-level surgery for logsummary,
> though probably they are orthogonal.

Nod to the orthogonal comment.




More information about the pcp mailing list