Hi, Mark -
> Hi Frank, I like this API - have a few questions and suggestions.
Thanks a lot for looking into it!
> To begin with, for the case where we make several calls to
> pmExtendFetchGroup_indom for several different metrics with the same
> instance domain and the same fetch group, the returned instance
> domains should match up, right?
Yes, generally.
> e.g. for proc metrics, which are pretty dynamic, that doesn't seem
> to be happening - but maybe I'm missing something ...
I ran your program, ramped up proc/pid churn, and saw a few
FAIL("nrss != nvss")
entries hit. Run under the debugger, it's clear that pmFetch
itself returned resultsets with different numbers of values
(see *fg->prevResult->vset[0] vs. [1] vs [2]). The inconsistent
results originated from pmdaproc. fetchgroup can't be blamed.
pmdaproc's "refresh" logic is probably to blame here, if indeed
there is any blame to place.
- FChE
|