Performance

The http4k server-backend modules provide a very thin adapter layer over the raw APIs of the underlying servers, so generally performs at a very low overhead compared to the raw server.

Tech Empower Benchmarks

We have entered http4k into the prominent Tech Empower Framework Benchmarks project, which assesses frameworks over a series of realistic tests.

For this benchmark, no customisation or performance tuning of the underlying servers is done - the default application HttpHandler is used which is then plugged into each custom backend, as below:

fun main() {
    Http4kBenchmarkServer(PostgresDatabase()).start(Undertow(9000))
}

Command-line JVM options, however, were tuned for the test to take advantage of various JVM features.

The full implementation of the benchmark can be found here.

Results - Round 22

Overall, http4k continues to do well in this round of benchmarking, placing 48/159 - especially considering that the ethos of the library is one of excellent Developer experience over and above high-end performance (which tends to result in less friendly APIs).

Rankings below are filtered for JVM libraries:

Composite ranking: results:

*Top rank: 13/41

DB query + HTML rendering: results:

Top rank: 25/146 - Apache backend

Database driver used is PostgreSql backed by a Hikari pool. Rocker templating engine used for rendering.

Multiple DB queries: results:

Top rank: 23/145 - Jetty Loom backend

Database driver used is Postgres Vertx Client backed by a Hikari pool.

Single DB query: results:

Top rank: 25/151 - Apache backend

Database driver used is PostgreSql backed by a Hikari pool.

Random DB updates: results:

Top rank: 41/138 - Jetty Loom backend

Database driver used is Postgres Vertx Client backed by a Hikari pool.

JSON Serialization: results:

Top rank: 59/152 - Netty backend

The standard Argo JSON module used for JSON creation and marshalling.

Plaintext pipelining: results:

Top rank: 84/153 - Netty backend

Recommendations

Benchmark your own app’s performance trying different engines if performance is critical. The Tech Empower benchmarks attempt to simulate simple real-world scenarios, but they can behave drastically different than your app. One other consideration is test time; some engines start up much faster than others.