[rrd-users] [rrd] Can't I turn off data smoothing for GAUGE?

[Alex van den Bogaerdt]

If I may chime in for a moment...

I have read (most of) the contributions, and I will comment on the tread,
not just on tobi's post.

Due to other activities I will not be able to keep participating; I hope
my contribution is useful and is accepted as-is.

>>  Why? It just GAUGE data, the only thing that we are doing differently
>> is storing the data as if it had occurred *exactly* on time.
>> As far as processing/graphing goes, I don't see any difference.
>
> to implement this, you would create a new datasource type ...

I respectfully disagree. If what they want is a timestamp which is
n*{amount_of_seconds_per_interval} then basically all what's needed is an
option to normalize the timestamp. That doesn't even have to be used with
GAUGE, it would also work with COUNTER.

A command line option, or even a rrdtool tune-able bitfield, rrdtool
update looks for this, adjusts $timestamp so that it looses whatever
remainder there is after dividing by $step, and you're done.

Next is people complaining about what happens after consolidation, either
because they ask for it in their RRA, or at graph time.

Some random food for thought, for all participants:

if at 00:01:00 there are 10 people in the room, and at 00:01:02 there are
11 people in the room, how many people entered the room _per_second_ ? 
Does that mean you can have half people?  Why would storing "1" be the
correct thing to do here? Because that's what is being asked for.

If I read my speedometer, a gauge, does it show "90" or does it show 90
kilometer_per_hour?  If you don't see why that's relevant: GAUGE is for
numbers which are already a rate, do not touch. Do_not_touch is about
converting to a rate, not about normalizing and consolidating. If RRDtool
would look at the difference over time, it would be converting the speed
measurement to an acceleration measurement.

Can the people who need integers please experiment with a step size of 1
second, LAST as the consolidation function, a large enough heartbeat, and
multiple PDPs per CDP?  You may still need to massage timestamps for be
exactly on the interval, meaning you still need to use something different
than "now", due to the sub-second precision built into rrdtool some time
ago.

HTH
Alex