Re: [pcp] pmie support for string-typed metrics

[Ken McDonell <kenj@xxxxxxxxxxxxxxxx>]

On 02/05/16 11:36, Mark Goodwin wrote:
> ...
> How about "async" logging frequency
>
> e.g. :
>
> log advisory on async {
>          hinv
>          kernel.uname
>          filesys.mountdir
>          filesys.blocksize
>          filesys.capacity
> }
If I am understanding the discussion so far, the request is for
(a) don't log unless changed, and
(b) log no more frequently than some delta

The (b) part provides the "rate limitation" and tells pmlogger how 
frequently to check for a change (this will be much simpler to explain 
and implement than checking on every single pmFetch for any metric).
So I think it needs to be something like ...

log [advisory|mandatory] on [default|every N timeunits] [if changed] {
        metric ...
}

And this can only apply to discrete or instanteous metrics (it would 
change the semantics for counters during replay).