Open kevinburke opened 10 years ago
StatusWolf does do it's own downsampling, I've had some strange results when pulling the data with downsampling done on the OpenTSDB end.
Essentially, for each of the available methods StatusWolf creates buckets for the interval and takes all the points in that bucket and applies whichever method you choose, resulting in one point per bucket. There's also now the option to choose "None" as the downsampling method, but that one requires some care as it's very easy to pull in enough data points that your browser will become unresponsive or crash trying to graph them - my testing puts that at somewhere around the 5,000 data point threshold for really bad things to start happening. Keep in mind also that your drawing a graph in a limited amount of pixel space, depending on the resolution of your screen it'll likely be a maximum of ~1500 pixels wide for a maximized graph. Trying to display more data points than you have available pixels isn't going to enhance the information at all, it's better to downsample so you can clearly see the trends in the data.
Would be nice if this clarification was available via a help dialog or page in the UI.
That is a good point, and I know that so far documentation has been the ignored child… My immediate goal is to get docs completed and in the wiki here, from there I'll add help links in the interface to the relevant docs.
Using downsampling will allow you to get less points, and so decrease the load for Statuswolf to graph them. Various functions for downsampling can be used : sum or average. Instead of getting 10 points for example, we average them and only get 1 point.
I don't remember if Statuswolf does its own dowsampling, or if it uses the Opentsdb native downsampling. I'll check that.
More information here for the OpenTSDB downsampling part : http://opentsdb.net/metrics.html