Trying to write some web stuff with Spock and Haskell

Posted on February 2, 2019
Tags: haskell, web, benchmarks

By far Haskell is the prettiest of the programming languages. If I am in doubt about how an algorithm works, I often just find an example in Haskell to get enlightened quickly.

Have a look at the canonical quick sort

or merge sort, borrowed from here

Can´t be done better, can it?

So far I have not done any web stuff with Haskell. It would be fun to try out some of the solutions that exists for this.

After a quick look I decided to take a look at Spock. The server looks simple enought as well as easy to build out. These two features are by the way always two of the most important factors I look for, when evaluating techlogies.

Okay, so I create a stack project and add Spock to the build-depends in my cabal-files. When doing a “stack build” I see that the latest stack version of Spock does not like the current version stm-containers.

So I decide to learn a little more about how to use stackage.

I checkout version 1.1.02 of stm-containers and build - and a stack install.

I checkout latest version of Spock and Spock-core and bulids.

In my own stack project I have to add these two lines in the packages section:


  • /home/petter/dev/haskell/Spock/Spock
  • /home/petter/dev/haskell/Spock/Spock-core

In the extra-deps:


  • reroute-
  • stm-containers-
  • stm-hamt-
  • primitive-extras-0.7.1

And then my stack project uses my local repos - and (hey!) it works.

I am now able to run my project.

Creating hallo world with Spock is now as easy as writing this Main.hs

Okay, now running it with

stack exec testspock-exe

and it answers fine on port http://localhost:8080

Okay, now I am curious to do some benchmaking. I use wrk

wrk http://localhost:8080

Results are like this:

This are very nice numbers, as I am running on an old (but nice) Linux box.

One thing, however. Several minutes after running the benchmark the Spock sever still use 16% of the resurces in while doing a top.

I do a new an more aggressive benchmark

wrk -t 10 -c 50 http://localhost:8080

The results are a little lower, but cpu now seems stuck at 30% usage even hours after the benchmarks and when the server has nothing to do.

I am not sure what happends here. Could be a regression as I work on the bleeding edge of the git, or something that I have not configured right?

But for now I think I will test another framework.