My primary goal in design of this solution to the assignment was to provide a clean Go implementation that provides an easy way to expand the scope into additional statistics beyond a simple average.
Calculating statistics on a set of action/timing data (which I refer to as an "event" in the code) requires the same basic steps for any basic stats. This implementation focuses on those needs and abstracts the actual calculations into a simple interface.
I chose to demonstrate the flexibility by implementing additional statistics beyond just an average.
To install:
$ go get -u -v github.com/akjc/jumpcloud
Basic usage is demonstrated in a small application in the examples/random package.
$ cd $GOPATH/src/github.com/akjc/jumpcloud
$ go run examples/random/random.go
Unit tests were written before much of the code was completed to verify accuracy and provide a basic test-driven workflow.
I chose not to implement unit tests for the additional statistics beyond average for the sake of brevity. This is the reason for the lower than 100% coverage.
$ go test -cover github.com/akjc/jumpcloud/event
$ go test -cover github.com/akjc/jumpcloud/event/stats
In order to keep the scope small enough for a simple exercise I left out some things that would likely be part of a true implementation:
- Basic stats might be desireable for more than simple action/time pairs
- The ActionEvent struct could be abstracted into something more generic to accomplish this
- Potential overflow is ignored, but noted
- Scale is assumed to be small enough to be handled within a single application
- I had to fight my background in "big data" big time
- If "web scale" were desired, the implementation for Tracker could include a number of improvements
- Sharding of keys (actions) into separate maps with separate mutexes to minimize waiting on locks
- Clustering of multiple instances to support a large number of distinct actions
- A map/reduce style approach to storing events in order to scale writes and calculations for a particular action type