Simon Hobson wrote: > Joachim Schrod wrote: > >> >> rrdtool create test.rrd --start '00:00 24.11.2009' \ >> >> --step 86400 DS:count:GAUGE:172800:0:U RRA:MAX:0.9:1:30 > > Note that all sample periods start at n*S from unix epoch (where > s is sample period)
Thanks - that's the sentence that I would have needed in the documentation to make me understand the behaviour. This is neither noted in rrdcreate man page nor in any of the tutorials that I read. For my actual task, taking measurements in UTC is OK, I need measurement-in-day accuracy, but should not loose data points. So I'll step up my heartbeat massively to avoid the problem of interrupting my time series. > - so specifying a full day (86400s) means sample periods will > ALWAYS start/finish at midnight UTC. > > You should read Alex's tutorial on normalisation. > http://www.vandenbogaerdt.nl/rrdtool/ I've read it now, and quite honestly, I wouldn't have understood its implications for my problem. It's not explained how the normalization target intervals are defined (which was the crucial missing information), only that they exist and that they are "well defined". It is also not explained that if my source interval only covers a small part of my target interval and the rest is unknown, my measurement is dropped. (I thought so at some point and changed my RRA xff ratio to 0.99999999, but that didn't change anything.) So thanks again for taking time to explain the basics. Joachim -- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Joachim Schrod Email: [email protected] Roedermark, Germany _______________________________________________ rrd-users mailing list [email protected] https://lists.oetiker.ch/cgi-bin/listinfo/rrd-users
