[rrd-users] graph/rrd def for small increment monitoring

[A Darren Dunham]

On Thu, Sep 03, 2009 at 04:08:48PM -0400, Kevin P. Foote wrote:
> I'm trying to monitor a (hopefully) low value attribute within mysql,
> slow_query count.
> 
> I have setup my rra like the following basing my input on results from this mysql
> command (SHOW GLOBAL STATUS like 'Slow_queries') this is an increasing
> count of slow queries on this server.
> 
> 
> filename = "my_slow_query.rrd"
> rrd_version = "0003"
> step = 300
> last_update = 1252007704
> ds[count].type = "COUNTER"

Okay.

> The data at start of polling this value was 95 .. it has increased to
> 139 over the past 4 hours or so .. with most entries into the rra being
> the same (but using type COUNTER this should be a LAST of 0) ...

139 - 95 => 44
44 / (4 hours) => average of .003/second or 3m (milli) queries / second.

> When I get the graph back the resolution on the Y axis is showing 80 m
> ... this should rather be something like 8 I would think... And the
> Maximal value Im GPRINTing is showing 76.75m ..

Since you have some zeros, and the average is 3m, I think a few peaks of
75m doesn't seem unreasonable.  Remember, it's not showing values at a
point in time, it's showing rates between points in time.

Maybe slow queries per second isn't a useful number for you.  Perhaps
multiplying by a factor and showing slow queries per minute or per hour
would be better?

Or also, it could be that graphing short periods of time isn't useful
and it would be better to only graph bigger buckets (like hours).

-- 
Darren