What is latency and why do we we want to measure it?
Latency is the time between a user invoked action and the resulting response. It gives us an expression of how much time it takes for a packet of data to travel within a network.
To measure latency I will be using Prometheus as it offers a range of options. I have provided a very simple demonstration in a unit test on how you could gather data by adding simple Prometheus methods to your code in this repo.
Please bear in mind that there is a lot more to measuring latency than relying on a few methods to log invocation and response times. And there are plenty more ways to display that data. I would recommend watching a talk by "Gil Tene" here.
Prometheus has a range of metrics collection methods. For this particular scenario we will be using Histogram which
is commonly used to gather latency data.
In the example below I am simply importing the dependency and building a Histogram like so:
Histogram myHistogram = Histogram.build()
.name("API_REQUEST_LATENCY_MILLIS")
.help("REQUEST_LATENCY_IN_MILLISECONDS")
.labelNames("METHOD")
.buckets(0.5D, 1D, 2D, 3D, 4D)
.register();
Here, I am declaring a new Histogram with one label called method. Now the interesting part
is the buckets .
Buckets, as the name suggests collect data and provide you with a count. Using the prometheus
api you can easily access this data and display it on something like Grafana.
The above buckets are labelled in milliseconds and represent a double data type. Simply put, the first bucket will collect all the calls that have been made and completed in half a second and so forth.
Once you have instantiated a Histogram object then you can start a specific timer with a label like so:
Histogram.timer myTimer = myHistogram.labels(method.toString())
.startTimer();
This is followed by a method which observes the time that has lapsed and counts it against the bucket you have declared.
myTimer.observeDuration();
So now every time your code gets called the Prometheus timer will start logging.