stackvia-bench

The standardized Spring Boot app we deploy to every platform for StackVia hosting benchmarks. Same app, same Dockerfile, everywhere → honest apples-to-apples numbers.

It's open source (MIT) so our benchmarks are fully reproducible — clone it, deploy it, and check our numbers yourself. Spotted something we could measure better? PRs welcome.

• Spring Boot 3.3.5, Java 21, spring-boot-starter-web + actuator.
• GET / → JSON: version (for redeploy checks) + uptime + how much RAM/CPUs the JVM actually sees (proves whether a plan can really run a JVM).
• GET /actuator/health → health probe for platform health checks.
• Binds to $PORT (falls back to 8080).

Build & run

mvn -q clean package -DskipTests      # -> target/bench.jar
java -jar target/bench.jar            # local run on :8080
# or, exactly how the platforms build it:
docker build -t bench . && docker run -p 8080:8080 -e PORT=8080 bench

Local baseline startup: ~4s (fast desktop). Expect more on constrained cloud plans.

---

Benchmark protocol

Deploy this repo to each platform, then record the fields below. Measure identically.

Order (easiest → most manual)

1. Railway — New Project → Deploy from GitHub repo → it detects the Dockerfile and injects $PORT. Nearly one-click.
2. DigitalOcean — App Platform → create from GitHub repo → Dockerfile deploy (this is the managed/PaaS experience, comparable to Railway). (Droplet + docker run is the raw-VPS alternative — worth a note if you have time.)
3. Vultr — raw VPS: smallest Cloud Compute instance → install Docker → docker build + docker run -p 80:8080 -e PORT=8080 bench. Most manual; time the steps.

Measure per platform

Field	How to get it
deployMethod	Dockerfile auto-detected? buildpack? manual? Any Java-specific friction.
deployMinutes	Wall-clock from "create" to a live URL responding (first deploy).
redeploySeconds / redeployAutoOnPush	Bump version.txt, git push, poll GET / until version changes = push-to-live time. Note whether it auto-deploys on push or needs a manual trigger.
startingPricePerMonthUsd	Cheapest plan that actually RUNS it. If the smallest OOMs, note it and size up — that's gold for "cheapest way to host a Java app".
planRam / planCpu	RAM + vCPU at that price (JVM is RAM-hungry — this is the story).
springStartupSeconds	From logs: Started BenchApplication in X seconds.
coldStartMs	If the plan scales-to-zero/sleeps: curl -w "%{time_total}" after idle. Else N/A.
jvmDetectedMaxHeapMb / jvmDetectedProcessors	jvmMaxMemoryMB and availableProcessors from GET /. Compare to the advertised RAM/vCPU — the gap ("plan says 0.5 GB / 1 vCPU, JVM sees ~700 MB / 2") is prime page material.
freeTier	Yes/No + limits (trial credit, sleep, hours). Can it run the JVM app?
httpsIncluded	Automatic HTTPS + custom domain?
dxNotes	Logs, rollbacks, CLI, dashboard, restart behaviour.
gotchas	OOM, build timeouts, port issues, anything that bit you.
screenshots	Dashboard, deploy logs, running /, pricing page.

Recording template (copy one block per platform)

platform: railway
planName:                 # e.g. Free, Hobby ($5)
startingPricePerMonthUsd: # cheapest plan that actually RUNS the JVM
planRamMb:                # advertised RAM
planVcpu:                 # advertised vCPU (e.g. 0.5, 1)
jvmDetectedMaxHeapMb:     # jvmMaxMemoryMB from GET /
jvmDetectedProcessors:    # availableProcessors from GET /
jvmIdleUsedMb:            # jvmUsedMemoryMB from GET /
springStartupSeconds:     # logs: "Started BenchApplication in X seconds"
coldStartMs:              # or N/A if always-on
deployMinutes:
deployMethod:             # Dockerfile / buildpack / manual
redeployAutoOnPush:       # does git push auto-trigger a deploy?
redeploySeconds:          # bump version.txt, push -> version changes in GET /
freeTier:                 # yes/no + limits
httpsIncluded:
dxNotes:
gotchas:
screenshots:              # file names / links (page assets, not a data field)
measuredOn: 2026-07-08

Fill three of these (Railway, DigitalOcean, Vultr), hand them back, and they become best-host-for-spring-boot.