Vapor benchmark - Spring-boot edition

What it is

Vapor is a Web framework written in Swift. The authors of this benchmarking article put out a call for help getting some benchmarks run using various popular Web frameworks from other languages; this is the one for Java's Spring Boot framework.

Test Methodology

This is the description of the benchmark tests each framework was to run. You can see that these each became a simple request handler in ApplicationController.java:

Plaintext

Request: GET /plaintext HTTP/1.1

Response: HTTP/1.1 200 OK

Hello, World!

JSON

Request: GET /json HTTP/1.1

Response: HTTP/1.1 200 OK

{
  "array": [1, 2, 3]
  "dict": {"one": 1, "two": 2, "three": 3}
  "int": 42,
  "string": "test",
  "double": 3.14,
  "null": null
}

SQLite Fetch

Grab a random user from the 3 users in the provided test.sqlite database:

sqlite> select * from users;
1|Foo|foo@gmail.com
2|Bar|bar@gmail.com
3|Baz|baz@gmail.com

This will require randomly choosing from 1, 2, or 3.

Request: GET /sqlite-fetch HTTP/1.1

Response: HTTP/1.1 200 OK

{
  "id": 2,
  "name": "Bar",
  "email": "bar@gmail.com"
}

Benchmark

wrk -d 10 -t 4 -c 128 http://<host>:<port>/plaintext
wrk -d 10 -t 4 -c 128 http://<host>:<port>/json
wrk -d 10 -t 4 -c 128 http://<host>:<port>/sqlite-fetch

How it is built

I started by cloning the Spring REST service starter guide to get a REST service up and running as quickly as possible. I added code and tests for the three REST endpoints described above, then ran it locally in IntelliJ IDEA. Once I was satisfied that the code was working and had good-enough test coverage, I uploaded it to the server and ran it with

java -Dspring.profiles.active=prod -jar ./vapor-benchmark-0.1.0.jar

You can see in application.yml that I define a dev profile that creates a fresh SQLite DB locally, and a prod profile that expects to find the one provided for the benchmark.

Challenges

The single biggest challenge was working with SQLite instead of a "blessed" low-footprint DB like Derby or H2. In the end I had to crib from a StackOverflow answer to get a Hibernate dialect that was compliant with Hibernate 4; the SQLite dialects you can find on Maven Central are not up to date, and lead to a runtime crash with a suitably inscrutable exception.

Conclusion

I have over a decade of experience writing Web and backend processing code in Java. It would be foolish to dismiss the power and battle-hardened nature of the JVM as a platform, but the language itself has never been anything but needlessly verbose and painful, with situations just like this being the norm rather than an surprising outlier. In the last few years I have turned my attention gratefully to Clojure and, to a lesser degree, Scala, as "escape hatches" that let me leverage the JVM while still working in a language that satisfies my desire for beauty and concision. I have very high hopes for Swift, and Vapor, to take a place among the top tier of beautiful, powerful tools for doing real work beyond the confines of iOS development.