request-hammer

A Tool(kit) to Test Services with Defined Request Rates.

Note this is only a sketch. That being said, it seems to be good enough to hit an HTTP service with request rates of up to a few hundred requests per second.

Here is an example how you might construct your test:

lazyvalrequests:Seq[Request[Either[String, String], Any]] =
cycle(
Seq(
basicRequest.get(uri"https://myservice/testA"),
basicRequest.get(uri"https://myservice/testB"),
basicRequest.get(uri"https://myservice/testC"),
) )
valphases=Seq(
Phase(requestRatePerSec =2, durationSec =2, requests),
Phase(requestRatePerSec =10, durationSec =5, requests),
Phase(requestRatePerSec =20, durationSec =5, requests),
Phase(requestRatePerSec =30, durationSec =5, requests),
)
valscheduler=newHttpPhaseScheduler
phases.foreach( phase =>
println(scheduler.run(phase))
)
scheduler.close()

Which will print a result along the lines of the following:

rq/s total succs fails P90 P99 P999
2 4 4 0 183 183 183
10 50 50 0 43 58 58
20 100 100 0 62 130 130
30 150 150 0 60 130 155

Ideas/ Issues

• Strong timeout for individual requests, to avoid Future.sequence timeout.
• Write a couple of tests.
• Pass scoring function into the scheduler alongside the request.
• Introduce some sort of tagging for the request type, so that we can analyse latency some more.
• Facility to write full request and response log.
• Add simple commandline tool that can take requests from file.
• Add binary chop example to show how to determine where the system under test collapses.
• Tune http client to really understand what we are really doing
• Provide simple graphing (flexivis!).
• Use timer facility rather than Thread.sleep to do the scheduling.