All,
  I'm trying to apply Holt-Winters to a few of our rrd's and have run into a problem which seems specific to the __SummaryInfo__ rrds. When I apply the Holt-Winters algorithm to both the host rrds as well as the __SummaryInfo__ rrds, the host rrds work fine however the __SummaryInfo__ do not. This note will focus on the __SummaryInfo__ rrds.

These metrics are stored in an rrd of type COUNTER. I have also observed this behavior on the same metric set to type DERIVE. 

I am using the rrd_hwreapply.pl script (http://rrfw.sourceforge.net/rrdman/rrd_hwreapply.pod.html) to apply holt-winters to the rrds and for simplicity right now I'm only using the following command line:

$ sudo -u nobody cp cron.webServiceRequestCounter.Counter.rrd cron.webServiceRequestCounter.Counter.rrd.bak && sudo -u nobody rrd_hwreapply cron.webServiceRequestCounter.Counter.rrd.bak cron.webServiceRequestCounter.Counter.rrd --defaults --force

As soon as I run this command, the primary metric (cron.webServiceRequestCounter.Counter) will stop rendering new values on the graph. It appears, however, that updates are still being stored in the rrd except in a different column in each row - I couldn't find much info about what the two different columns in the rrdtool dump output are for. 

I have placed images, rrdtool dump output & version info into a github repository here - just view the readme

https://github.com/adnichols/ganglia-hw-graph-problem

I dug through the rrdtool changes in 1.4.6 & 1.4.7 and didn't see anything that looked like this problem, however if someone can point me to something that suggests rrdtool is the issue I can update. It's just a fair amount of work that I'd like to avoid if it's not the issue. 

Thanks,
Aaron

--
Twitter: @anichols
Blog: http://www.opsbs.com
Pro: http://www.linkedin.com/in/anichols