WARNING: the below is very raw. Not really for public consumption until I have some time to clean things up. (I wrote the original a few years ago in a combination of R and Ruby. I'm going to write a couple of new implementations in a single language, which hopefully should make it more straightforward to use.)
Trendy is a simple disk trend reporter and alerter. I created it as a supplement to traditional threshold-based alerting. Sometimes alerts are timely, but other times they can be unnecessary (slow growing large disk), or totally missed (small, fast growing disk).
The motivation was a couple of articles
http://lpenz.github.com/articles/df0pred-1/index.html
There are 2 phases:
- collection
- trend analysis
Collection consists of gathering the statistics. Right now, it's a fairly small, simple wrapper around Ruby. You can have it collect statistics for a given machine only, or you can have it run against multiple machines. By default, it is in "localhost" mode. Eventually, there will be a way of pulling statistics out of OpenTSDB or InfluxDB.
The trend analysis stage uses a Monte Carlo simulation.
These scripts are developed in a literate programming style.
Specifically, the Ruby is written using:
http://rtomayko.github.com/rocco/
And R is written using
Perhaps.
To rebuild the documentation:
groc scripts/*.rb README.md
For some example output, see ...
This is a placeholder to a mini-presentation describing my motivation.
Is this you?
Tale of 2 servers
Since there may or may not be code shared with df0-pred, we'll just go with it being under the same terms for now...
- increase linear component. For example, past 12 hours / samples. That way if we have a runaway process, we can estimate when that will cause us to run out of disk.