Typesafe Config Json parsing performance

[Michael Li]

Has anyone done performance comparison between Typesafe config with other Json parser like Json4s-jackson? Is it good enough to be used as general json parser in a heavy load API service? 

[Yann Simon]

I made some tests here: https://github.com/yanns/scala-json-parsers-performance

I advise you to look at the JMH test results.

Le mar. 5 avr. 2016 à 21:05, Michael Li <m...@ooyala.com> a écrit :

Has anyone done performance comparison between Typesafe config with other Json parser like Json4s-jackson? Is it good enough to be used as general json parser in a heavy load API service? 

--
You received this message because you are subscribed to the Google Groups "scala-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scala-user+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Michael Li]

Thanks for your post. I was asking typesafe config's performance, but I can't find it in your list.

[info] Parameters(parser -> argonaut): 0.194616
[info] Parameters(parser -> jackson): 0.058777
[info] Parameters(parser -> json4sJackson): 0.224313
[info] Parameters(parser -> json4sNative): 0.23916
[info] Parameters(parser -> playJson): 0.245464
[info] Parameters(parser -> sphereJson): 0.143125
[info] Parameters(parser -> sprayJson): 0.198715

[Yann Simon]

sorry, I saw JSON and misunderstood your request.

[Julien L.]

The HOCON format has much more feature than plain JSON (variable substitution, expression, merging...), so it will be slower than any JSON parsing library.

Besides, the configuration file is parsed one time when the app start, so performance is not the priority.

I'll not use the typesafe config parser for an heavy load API. Go to Json, with for example Circe or any API backed by Jawn.

BTW, Yet another JSON benchmark https://github.com/studiodev/json-bidrequests-benchmark-scala

[Naftoli Gugenheim]

On Wed, Apr 6, 2016, 5:24 AM Julien L. <yotsu...@gmail.com> wrote:

The HOCON format has much more feature than plain JSON (variable substitution, expression, merging...), so it will be slower than any JSON parsing library.

Besides, the configuration file is parsed one time when the app start, so performance is not the priority.

That's up to your application.

[Glen Marchesani]

We have a workflow app where a lot of the extended features are exposed via HOCON stored in a database table.  Under load we are parsing a 10 to 20 HOCON docs (10 to 100 lines each) a second just fine.  We also know if we need to get more performance we can easily add an MRU caching layer to give us an order of magnitude boost.  So I would say it really is a balance between need for the extended capabilities of HOCON over json and high performance parsing.  We targeted 100 / 100 line json docs per second and felt we could achieve that with HOCON (noting that translates into like 50K requests per second on our app).

Note #1: we also looked at and were impressed with https://hjson.org/ 

Note #2: we actually parse the HOCON then process the HOCON layers into a json4s JValue (so the performance is for doing all of that not just the HOCON parse, the HOCON parse takes the majority of the cycles).