On Tue, 20 Jun 2000, Michal Kara wrote:
> > >When I add it, you will be able to run:
> > >pcpmon -a day1=... -a day2=... -s -1d -a day3=... -s -2d
> >
> > Consider a syntax that will support wildcard expressions for the archive
> > files
> > (e.g. /var/log/pcp/pmlogger/{node}/2000*) and optionally several
> > directories to
> > search.
> >
>
> This wouldn't be easy to do in PCPMON (if you mean concateating more
> archives).
Assuming the archives are all collected from the same host ...
The semantics of the archives (and in particular the meta data
therein) make it difficult for a tool to process chronologically
adjacent (or even overlapping) archives. Rather it is better to
stitch the archives together with pmlogextract(1) and then point the
tool at the single resultant archive.
If the archives are collected from different hosts then you have a
different set of curly issues.
> > Given that the archives contain date/time stamp info, one could envision a
> > PCPMON syntax that allows the user to specify a date/time range (e.g. -beg
> > 2000/06/01:09:00 -end 2000/06/30:17:00 -day 08:00-17:00).
> >
> This may be interesting option, to limit times... I will thing about
> this.
Check out the various options under TIME WINDOW SPECIFICATION in the
pcpIntro(1) man page. Most of the common perversions for setting
time windows are supported there, and there is PCP library support
for parsing options in these formats.
> > Perhaps the ability to handle compressed formats (gzip,compress) would be
> > helpful since pmlogger_daily does this.
> >
> Isn't this responsibility of PCP library?
Don't go there. If/when you understand interp.c in libpcp, you know this
is a bad plan. PCP archives may be accessed more or less at random, and
read forwards or backwards and any combination of the above. So uncompress
on the fly is not possible. If you uncompress when the archive is opened
you have to worry about
- disk space
- recompressing
- cleanup on exit
on balance I decided these are all tasks better handled by humans via
the shell.
|