Re: [pcp] pcp updates: pmlogger semantics change, 1st round of strtoint64 changes

[Ken McDonell <kenj@xxxxxxxxxxxxxxxx>]

On 15/10/15 17:22, Marko Myllynen wrote:
> ...
> thanks for looking into this - I couldn't imagine how much work such a
> "simple" change would require!
It got a little messy because previously all the "log once" records 
magically appeared at the beginning of the archive, but now we need to 
force this to happen before any of the "now" samples for the log with 
delta metrics ... this is needed to keep lots of tools happy, e.g. 
expecting to find hinv.something before kernel.somethingelse in the archive.
And the QA fallout is all because now [simple explanation with one 
metric and one sample interval] if the sample delta is D seconds, and 
the archive contains K samples, then previously the time from the start 
to the end of the archive was K*D seconds, but now it is (K-1)*D seconds 
... and conversely if pmlogger stops after X seconds there is one more 
record in the archive now.
The change in the boundary conditions meant that some tests produce 
slightly different results now.