Re: pmrep: add support for swapping columns and rows in stdout output

On Tue, Nov 8, 2016 at 12:04 AM, Marko Myllynen <[email protected]> wrote:

> On 2016-11-04 18:14, Marko Myllynen wrote:
> > On 2016-11-03 03:05, Mark Goodwin wrote:
> >>
> >> I love it! Testing it out, could we consider adding an argument to the
> >> -X flag to label the instance name column, and maybe some way to specify
> >> it's width (independently of the width for the data columns)?
> >
> > Yeah, I noticed that it felt a bit of compromise to use the timestamp
> > width as the default width, perhaps indeed we could add some sort of
> > method to specify width for that, either a separate "field" or something
> > like disk.dev.util,'%util',,,6:10, what do you think?
>
> Hmm, specifying this on per-metric basis of course doesn't make much
> sense, not sure what I was thinking. Perhaps we could have -X to take an
> argument which could be < 0 turn turn this feature off (the default), 0
> use timestamp width, and > 0 to explicitly specify the width for the
> instance column.
>

yes that would work, or perhaps a quoted string argument could determine
the column width,
and also provide a heading for the instance column. e.g.
    -X "             Device Name"

(one problem with dm devices is they tend to be very long (especially
docker / thin-pool LVs).


> What do you think? Ok, defining something to 0 to enable it is not the
> most intuitive way but I'd like to avoid a separate configuration knob
> for the instance width, if possible.
>

yeah, we're getting into option overload territory a bit


>
> >> Also, when metrics with different instance domains are specified, maybe
> >> we could have a 'filler' such as '-' or '.' or something for the columns
> >> which don't make sense on the line for each instance. This would make it
> >> easier to parse and post-process because we'd have a fixed number of
> >> columns to pick out in the report.
> >
> > Would setting the column delimiter help, e.g., "pmrep -l : ..."? But
> > yes, printing "-" or such wouldn't be unreasonable either.
>
> Testing this with "-" I think it's aesthetically ok to be used as the
> default and not even worth a separate config option (so probably will
> add a constant and use it).


yep sounds good. Maybe not '-' since it's a valid numerical prefix (unary
negation).
And "." would have the same problem. I guess ':' or '_' might be OK?

BTW the existing column separator option is also OK, e.g. pmrep -l: ... |
awk -F: ....

Thanks, let's get this merged asap :)

CHeers