libmetrics/linux/metrics.c and network counter bit width

[ganglia]

libmetrics/linux/metrics.c and network counter bit width

[avinson]

thanks for posting this.. it's pretty annoying when this happens because it ruins all of your aggregate graphs

[bkurtz]

I was lucky and caught this very shortly after I first installed ganglia on a 32-bit host, so things hadn't been averaged out (in time) too much yet. I cleaned up my RRD files by doing (approximately)

cd /var/lib/ganglia/rrds/CLUSTER/HOST
rrdtool dump bytes_in.rrd > /tmp/bytes_in.xml
sed -E -e 's/e\+1[567]/e+01/' -i /tmp/bytes_in.xml
rrdtool restore -f /tmp/bytes_in.xml bytes_in.rrd

(basically, find any network levels greater than 10^15 and drop them down so the exponent is 1 instead of 15, 16, or 17, i.e. basically set to zero). It won't be 100% accurate, but it's an easy way to do the change, and 100% accuracy doesn't really matter so long as you don't have those 10^17s getting averaged in to your data and messing up the scaling on the graphs.