wrk-bench

wrk-bench is a library to make creating wrk jobs that recur immediately. This is good for having tests that collect stats for a couple seconds, record them, then does it again. This assumes you'll pipe the results into something that makes sense.

usage

Since this is just a light wrapper around wrk the node module, which is just a light wrapper around wrk the command line tool. You'll just pass the options that you'd normally pass to wrk to wrk-bench like so:

var bench = require('wrk-bench')
 
var job = bench({
    url: 'http://localhost:3000',
    threads: 2,
    connections: 5,
    duration: '10s'
})
 
job.pipe(process.stdout)

Which might result in output that looks like the following:

{
  "transferPerSec": 355090,
  "requestsPerSec": 228.54,
  "requestsTotal": 2286,
  "durationActual": 10000,
  "transferTotal": 3470000,
  "latencyAvg": 17.62,
  "latencyStdev": 4.29,
  "latencyMax": 73.93,
  "latencyStdevPerc": 98.67,
  "rpsAvg": 115.89,
  "rpsStdev": 15.94,
  "rpsMax": 171,
  "rpsStdevPerc": 79.32,
  "options-threads": 2,
  "options-connections": 5,
  "options-duration": "10s",
  "options-url": "http://localhost:3000"
  "timestamp": 1413854725961
}

Quantities of data are measured in bytes, quantities of time are measure in milliseconds.

Notice how the options is merged into the object. This makes it so you don't have to track each job yourself. Also if you add any other properties to the option argument they will be merged in and if wrk doesn't know about them, they wont be used. This is handy if you want to name jobs or add in any interesting version data or something of that sort.

Also a timestamp of when the test finished is added to the object as well. I am up for debate on this and could also see collecting start/end and reporting them, or maybe not having a timestamp on at all and leaving that up to a user of the API

theoretical changes

• Possibly take an array of jobs
• Delay between jobs seems like a good idea but I don't have a use case yet.