trk

trk is an event based stats library for node.js that uses redis as a data store.

depending on the project, it can either be a total replacement for map/reduce routines, or, at the very least, a catalystic compliment that allows you to strike a performant balance between real-time data and hierarchical time bucket(y) reporting.

key concepts

• events are recorded as objects (.record()), the only requirement is it has an event property
• redis counters are modified on the fly into different areas as events come in
• the time granularity supported is daily
• there are 4 types of configurable properties for recording events
  1. bmp ("bitmap") counts unique values occurrences for this key
  2. add counts number of occurrences for each value of this key
  3. addv ("add value") sum of the values for all occurrences of this key
  4. top similar to add but returns the top occurrence values in order and limits the result set to 250
• compound keys are supported by mapping them with a ~ delimiter
• an interface to query data over date ranges is supplied (.query(),.query_days()), see examples
• querying does not require a config and can be done from other machines, recording requires a map configuration

the data structure in redis ends up looks something like this

and for compound key combinations, if you choose to use them, like this

benefits

• recording an event is :zap: fast and can be done in a fire-and-forget fashion
• queries are :zap: fast and can be done in real-time with no need for caching in most cases

limitations

• no regression. the raw event is never stored, it is broken down and digested into specific locations based on configuration and then is hair-flipped

install

using npm

npm i trk --save

usage

examples are located in src/examples

recording speed

build/examples/record-events.iced on a mbp i7(3.1)/16gb

$ node record-events.js
Finished recording 10000 events (series) in 7046ms
Series events digested/sec: 1419.244961680386
Finished recording 10000 events (parallel_limit) in 6449ms
Parallel events digested/sec: 1550.6280043417585

inspiration

• statsd

todo

• tests
• map.bmp elaboration (support compound fields)
• data inflation routine for compound key result sets
• winston implementation for debug logging