Migrate from Rocket to Axum

[RemiBardon]

I don't want to finish #71 using Rocket while we're supposed to migrate to Axum after that. Let's migrate now.

• REST API
• Smoke tests
• Integration tests
• ADR (see ADR: Use Axum instead of Rocket)
• Compilation time comparison
• Size comparison

Compile time and size comparisons

After rewriting the API using Axum and fixing all smoke tests and integration tests, I executed a custom script which would build the API before and after the rewrite, in release mode, to compare the compile times and image sizes.

I executed the script on my machine with no other app running in the foreground, and made sure to purge all Docker caches to get a complete build.

The script

#!/bin/bash

SUMMARY_FILE="summary-$(date +%s).txt"
TIME_LOG_FILE='build-time.log'
IMAGE_NAME='proseim/prose-pod-api:local'

try() {
	local ref="${1:?"Pass a git revision (hash or tag)"}"
	echo "# ${ref:?}" >> "${SUMMARY_FILE:?}"

	git -C "${PROSE_POD_API_DIR:?}" checkout "${ref:?}"

	docker system prune -a -f
	/usr/bin/time -o "${TIME_LOG_FILE:?}" task --taskfile "${PROSE_POD_API_DIR:?}"/Taskfile.yaml build-image

	build-time >> "${SUMMARY_FILE:?}"
	image-size >> "${SUMMARY_FILE:?}"
	echo '' >> "${SUMMARY_FILE:?}"

	rm "${TIME_LOG_FILE:?}"
}

build-time() {
	cat "${TIME_LOG_FILE:?}"
}

image-size() {
	docker image ls "${IMAGE_NAME:?}" --format "{{.Size}}"
}

try v0.4.0
try axum

After 18 minutes, it yielded the following:

# v0.4.0
      606.08 real         4.27 user         5.88 sys
25.2MB

# axum
      410.05 real         3.33 user         5.11 sys
24.1MB

We can see that the build took 32% less time using Axum ($(606-410)/606=0.32$), and 22% less User CPU time ($(4.27-3.33)/4.27=0.22$), which is even better than I expected! System CPU time also went down by 13% ($(5.88-5.11)/5.88=0.13$).

The resulting binary is almost the same size, with a small 4% reduction ($(25.2-24.1)/25.2=0.04$). At least it didn't increase.

---

NOTE: I will fix release builds in a separate PR. (Already done in #96)