trink / streaming_algorithms Goto Github PK

Track a histogram of the data

Track the uniqueness of the field using hyperloglog

Alerts should be ignored while the time_series_length of data is being initially populated. i.e. mp[1] = INF

Note: this should be a cfg option. It is noisy during the initial load but it has also detected a noteworthy change in behavior (in a lot of cases it may be better to leave it on and simply review the alerts/load progress)

This is needed to store the normalized histogram values

@trink just making a quick note of this here,

It seems with the default build this ends up in /usr/lib64 where LSB extensions end up in /usr/lib; not immediately obvious why but I had to adjust some configuration files to look in the alternate location so there may be an inconsistency somewhere

This file is quickly switching from a debug aid to a data source for other tools.

We are seeing too many alerts for what appears to be normal but relatively large fluctuations; investigate and find a better way to handle the unique data types

If the receiving time series is a different size but the resolution is the same or greater merge the data.

Analyze the counts of the set as an enumerated histogram