On 06/10/15 10:36, Nathan Scott wrote:
...
So, there's at least one problem here - firstly, we should have a fallback for
hertz if we cannot fetch it. But, the archive from this test *does* actually
have a kernel.all.hz in the 2nd record (see pmdumplog -a qa/archives/pcp-atop)
- a discrete metric too. I'll add some default-hertz code, but any ideas as to
what's going on when fetching that metric? (and why it might work for me but
not you?)
When I fetch it there are no values for _any_ of the "host" metrics from
this call:
fetch_metrics("host", HOST_NMETRICS, pmids, &result);
and this would appear to be because this archive fetch is being done in
interp mode and the current time is 11:03:19.770 which is _before_ the
data record that contains the host metrics at 11:03:19.827.
So interp.c is doing the "right" thing.
I have no clue why it is working for you!!
But if I change the fetch mode to be PM_MODE_FORW from the start of the
archive for just this case (as I think it should be) ...
1. the FPE error goes away (hertz is 100 as expected)
2. qa/785 generates different output (looks more correct to me, but what
would I know?) ... see attached 785.out.bad
3. and qa/1079 fails ... similar to qa/785 with additional information
being found for the header lines ... see attached 1079.out.bad
I've also included my code change patch in the event that that helps.
But still, "working" for you is not explained.
785.out.bad
Description: Text document
1079.out.bad
Description: Text document
atop.patch
Description: Text Data
|