[rrd-users] Strange activity after a DST change

[Alex van den Bogaerdt]

On Tue, Mar 11, 2008 at 09:59:46AM -0500, Smith, Kyle A. (Security) wrote:
> Howdy all,
> 	I am a relatively new to rrdtool, but have setup a few
> re-occuring charts for system metrics for 2 systems.  The process
> (gathering the metrics from the systems, sending them to the 'local'
> system housing rrdtool, updating the rrd db's, and generating the
> graphs) has been working for a few months now without a hitch.. until
> this past weekend when the system 'corrected' its time for local
> daylight-savings.

Time isn't corrected, only its representation to the user.  Internally
time just continues, as can be seen here:

>             <!-- 2008-03-09 00:00:00 CST / 1205042400 --> <row><v>
> 1.4726296296e-01 </v></row>
> 		<!-- 2008-03-09 01:00:00 CST / 1205046000 --> <row><v>
> 1.0902040816e-01 </v></row>
>             <!-- 2008-03-09 03:00:00 CDT / 1205049600 --> <row><v> NaN
> </v></row>

Timestamp 1205042400
Timestamp 1205046000 = 1205042400 + 3600
Timestamp 1205049600 = 1205046000 + 3600, one hour difference, not two.
And it should be that way!


You aren't indicating that errors occur, but from past experience I
learnt the hard way I have to ask: do you get any errors, most notably
such as "ERROR: illegal attempt to update using ti"... ?
Or are you perhaps sending all those messages to /dev/null ?

Things to look at:
1: is the device measured still sending data, or do you get unknowns
   to begin with, in which cases it's quite normal for the RRD to show
   unknowns

2: has someone or something fiddled with the computer's clock to
   'correct' for DST, and has the database been filled with a bunch
   of unknowns as a result. Check rrdtool last on that file against
   the current system time as seen by date +%s. If last updated is
   in the future, someone set the clock forward, e.g. to 2008-09-03
   instead of 2008-03-09.

-- 
Alex van den Bogaerdt
http://www.vandenbogaerdt.nl/rrdtool/