[rrd-developers] Re: Restarting devices

[Bert Driehuis]

[ apologies if you see this twice -- I got bitten by the spam check ]

Alex van den Bogaerdt wrote:

> (note:
>      I assume here that sysUptime is in hundreds of seconds.
>      is this always true?)

Yes -- the designers of SNMP inconsiderately didn't take 60Hz into
acccount :-)

> To signal the unknown interval caused by the reset, the front end should
> insert "U" into the database.
>
> If the current time is 948669019, just insert "U" one second before
> and insert the correct value with the current timestamp.

[snip]

Just for kicks, I implemented precisely this for Cricket, the diffs are in
the mailing list archives for Cricket.

> Real brave people use sysUptime and current time to estimate when the
> reset took place.  Assuming sysUptime had a value of 10000, you could
> do:
> 
>    previous:  rrdtool update my.rrd 948668705:1000
>    current:    rrdtool update my.rrd 948668918:U 948668919:0 948779019:500
>    future:     rrdtool update my.rrd 948669309:1200
> 
> This minimizes the unknown data.

Cool idea. And rather obvious, now that you mention it. Hmmm, it means
I'll have to find out which data sources actually are counters and which
are not. As it is, the current patch to Cricket just inserts an all-U
record into RRD; the dummy record that you suggest should be zero for
counters, and either U or an approximation (average, last sample,
whatever) for GAUGEs.

Cheers,

				-- Bert

Bert Driehuis -- driehuis at playbeing.org -- +31-20-3116119
Hi! I'm a signature virus! Copy me to your .signature and help me spread!


--
Unsubscribe mailto:rrd-developers-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-developers-request at list.ee.ethz.ch?subject=help
Archive     http://www.ee.ethz.ch/~slist/rrd-developers