Hmmm.
Does not disprove my hypothesis, but does not prove it either.
20081006 is a bit odd. If pmdumplog -l reports an end time then the
archive is not really busted. But pmdumplog -v is a strange beast, that
looks like a pmdumplog bug (sigh).
Just try pmdumplog 20081006 | head -100 ... the timestamps appear in the
pmResult dumps.
But 20081006 looks like it is for the correct day which undermines my
theory.
I believe the problem is in pmdate -1d when you're within an hour of
midnight and there was a DST adjustment in the previous 23-25 hrs ...
there is no good answer here, pmdate cannot produce the 100% right
answer no matter what.
I'm looking at reworking pmlogger_daily to remove all dependence on
pmdate, and this will give me a chance to re-instate the -o option ...
actually new behaviour to merge _all_ archives, not just yesterday's and
-o for the old (current) behaviour, so all of pmlogger_daily QA will
work again ... all of this was lost in one of the great rattle
filesystem corruption disasters.
On Tue, 2008-11-25 at 09:38 +1100, Nathan Scott wrote:
> On Tue, 2008-11-25 at 04:53 +1100, Ken McDonell wrote:
> > I may have a handle on this.
> >
> > Nathan, can you please check the following for me.
> >
> > 1. the summary archive for 20081004 is ok.
>
> Yep.
>
> > 2 the summary archive for 20081005 is missing.
>
> Yep.
>
> > 3. now can you check the summary archive for 20081006 ... I'm interested
> > in the actual dates for the label and the log records in that
> > archive ... I suspect it is really data for 20081005.
>
> See below, appears to really be data for 06 AFAICT.
>
> > 4. and if that is the case, I suspect you also have the component
> > archives for 20081006 that have not been merged (although these may well
> > have been culled by now).
>
> Yeah, no component archives left anywhere now that I can see.
>
> $ pmdumplog -l 20081004
> Log Label (Log Format Version 2)
> Performance metrics from host app4
> commencing Sat Oct 4 00:13:41.741 2008
> ending Sun Oct 5 00:13:26.753 2008
> $ pmdumplog -l 20081006
> Log Label (Log Format Version 2)
> Performance metrics from host app4
> commencing Mon Oct 6 00:12:14.239 2008
> ending Tue Oct 7 00:13:35.340 2008
>
> All options I tried to pmdumplog give timestamps relative
> to start of the day (and all appeared to be from the start
> of 06). Trying to get more info with -v was unhelpful - a
> bug lurking here too perhaps? All archives produce this:
>
> $ pmdumplog -v 20081006.0
> Raw dump of physical archive file "20081006.0" ...
> Dump ... record len: -2080374784 @ offset: 0
> Trailer botch: 36046160 != -2080374784
>
> Is there a better way to dump out the timestamps than with
> pmdumplog? (or an option I've missed? -Dsomething?)
>
> thanks!
>
> --
> Nathan
>
|