rrdtool graph without downsampling data?

Question

I'd like to leverage the beautiful (in my opinion) graphing abilities of RRDtool with data stores that don't down sample (influxdb, statsd, etc).

From my reading it seems like the only way to do this is to synthetically recreate an RRDtool database from the full resolution time series data, and use that as the foundation for my graphing?

I might have thought that some best-practices, or guides for gnuplot or similar to render graphs in the RRDtool style might exist, but I have not found anything.

Answer 1

You can use RRDTool with the sql// pseudo datasource as defined in the manual here.

This is rather complex, but allows the use of the rrdtool graph command with external DBI-based SQL queries in the DEF declaration. In this you can specify connection criteria, the fields holding the timestamp and value, and any additional WHERE clauses for the SQL query.

Performance is not as good as with native RRD files of course.

EG:

DEF:myvar=sql//mysql/host=127.0.0.1/dbname=rrd/username=rrd/password=passwd//RRDValue/UnixTimeStamp/value/RRDKeyID=1141942900757789274:min:AVERAGE

See the manual page for details.