Running the reference, start to finish.

Required tooling

• SDKMAN — manages JDK 25, Maven 3.9.x, JBang, and the Quarkus CLI
• Quarkus CLI — quarkus create, quarkus dev
• Docker Engine and Docker Compose — the v2 CLI plugin
• Camel CLI & the Camel MCP server — Camel-on-Quarkus route prototyping and validation
• Demo command-line tools — curl, jq, grpcurl, and kcat (Fedora: sudo dnf install kcat); demos/walkthrough.sh checks for them before it starts
• The Quarkus Agent MCP server — dev-loop tooling for Claude Code (project create, skills, docs search, logs)

Installing the toolchain via SDKMAN

curl -s "https://get.sdkman.io" | bash
source "$HOME/.sdkman/bin/sdkman-init.sh"
sdk install java 25-tem
sdk install maven 3.9.9
sdk install quarkus

Container engine sizing

Docker Engine on Fedora or RHEL (bare metal or VM) is required. Everything here runs on that one engine: the Compose stack, Dev Services and Testcontainers during mvn verify, the Mandrel builder image for native builds, and minikube, which runs as a container under its Docker driver. The engine uses all host CPUs and memory and stores images under /var/lib/docker, so there is nothing to size beyond the host itself.

Install Docker Engine and the Compose plugin for Fedora or RHEL, then enable it and point the CLI at it:

sudo systemctl enable --now docker
sudo usermod -aG docker "$USER"   # log out and back in
docker context use default
docker info

Membership in the docker group is root-equivalent on the host.

Workload	CPUs	Memory	Disk
Compose stack, dev mode, mvn verify, native build	8	16 GB	100 GB
Everything above plus minikube at the bootstrap defaults (24 GB)	8+	28 GB	200 GB

Disk runs out first; the native-build demo and the Ollama image pull for the AI service tests need room. Budget host memory the same way: at least 32 GB for the minikube path, leaving room for the OS, IDE, and browser.

A VM-based engine such as Docker Desktop also works; size its VM larger than the node and set MINIKUBE_CPUS / MINIKUBE_MEMORY to fit.

Fitting minikube to the engine

scripts/bootstrap.sh starts minikube with 24 GB, 16 CPUs, and and about 100 GB free under the Docker data root (/var/lib/docker; the docker driver ignores --disk-size). minikube cannot request more CPUs or memory than the engine has, so override the defaults to fit. On an 8-CPU host:

MINIKUBE_CPUS=8 ./scripts/bootstrap.sh
# also available: MINIKUBE_MEMORY (default 24g); MINIKUBE_DISK is ignored by the docker driver

Host access

The minikube profile publishes the platform NodePorts on 127.0.0.1 when it is created, so the cluster is reached at the same localhost URLs as the Compose stack:

Service	URL
Grafana	http://localhost:3000
OTLP gRPC / HTTP	localhost:4317 / http://localhost:4318
Mimir	http://localhost:9009
Loki	http://localhost:3100
Tempo	http://localhost:3200
Kiali	http://localhost:20001/kiali
Apicurio	http://localhost:8084/apis/registry/v3

./scripts/show-endpoints.sh prints the status table. The ports are fixed when the profile is created; ./scripts/setup-profile.sh --replace recreates it, which deletes the cluster, so re-run ./scripts/bootstrap.sh afterwards.

Compose and the cluster cannot run together: the Compose lgtm container uses ports 3000, 3100, 3200, 4317 and 4318. Stop the cluster with minikube stop -p datamesh before running Compose demos.

Optional: OpenShift Local (CRC)

The appendix Running on OpenShift Local runs the seven core services, Kafka, Postgres and Apicurio on a single-node OpenShift 4.22 VM. It needs crc with a Red Hat pull secret, oc, helm, jq and Maven with JDK 25; images are built inside the cluster, so no container engine is involved. It was verified with 6 vCPUs, 20 GB of memory and an 80 GB disk. Run it instead of minikube, not alongside it, and finish with ./openshift/teardown.sh, which removes everything it installed and stops CRC.

Verifying a service

mvn verify -f examples/pom.xml   # unit + integration tests, every service
mvn package -Pnative              # native build (GraalVM/Mandrel)
quarkus build --native            # equivalent, via the Quarkus CLI
./scripts/run-all-tests.sh        # single entrypoint for the full test suite