README
¶
Brewlet integration tests
Cross-component end-to-end harness for the Brewlet monorepo. This directory owns test orchestration and fixture applications. Production code, Kubernetes manifests, and specifications remain in their owning monorepo directories.
Layout
e2e/ runner, reset helper, shared library, and tiers 1-17
fixtures/ harness-owned Java demo applications and PetClinic build fixture
Running from the monorepo
From the repository root, run:
integration-tests/e2e/run.sh --tier 1 --tier 2
The harness defaults to core/ and kubernetes/ in the monorepo. The
BREWLET_CORE_DIR and BREWLET_KUBERNETES_DIR overrides remain available for
testing external checkouts.
Running
integration-tests/e2e/run.sh # all 17 tiers
integration-tests/e2e/run.sh --list # tier catalog
integration-tests/e2e/run.sh --tier 4 # one tier
integration-tests/e2e/run.sh --reset # remove Brewlet test state
integration-tests/e2e/run.sh --reset --tier 10
E2E_WORK=/tmp/brewlet-e2e integration-tests/e2e/run.sh --tier 9
Tiers skip when an optional host prerequisite is unavailable and fail only when
an exercised capability fails. PR CI sets E2E_REQUIRE_ALL=true, which also
fails skipped assertions and tiers with no passing assertions. Tier 2 tests
Maven goals without repeating the unit suite; run make maven-plugin-check
for Maven unit coverage. The suite covers:
| Tier | Scope | Primary components |
|---|---|---|
| 1 | Go unit/component suites | core + Kubernetes |
| 2 | CLI push, inspect, run, bundle, classpath, and JPMS; Maven plugin config, inspect, build, appcds, and push |
core + maven-plugin + fixtures |
| 3 | shim to runc under Linux cgroups | core + fixtures |
| 4 | operator control plane and Helm packaging | Kubernetes |
| 5-6 | host and in-cluster admission webhooks | Kubernetes |
| 7 | Spring PetClinic artifact, layered deployment, runc, reconcile | both + fixtures |
| 8-9 | AppCDS lifecycle, cross-namespace cache isolation, and serving through kubelet/CRI | core + fixtures |
| 10-11 | installed Helm stack and webhook resilience | Kubernetes |
| 12 | runnable image pulled and unpacked by kubelet | core + fixtures |
| 13 | NodeProfile lifecycle | Kubernetes |
| 14 | retired: every JDK is NodeProfile-sourced, so its coverage moved to tier 18 and NodeProfile envtests | — |
| 15 | live opt-in Prometheus metrics through Helm, provisioner, shim, and exporter | both + fixtures |
| 16 | SPECIFICATION §14 failure modes on a live node | core + fixtures |
| 17 | default-enabled runnable-stage GC: fresh activation, installation safety blocking, reference protection, and reclamation | both + fixtures |
| 18 | containerd restart rollback, sourced JDK + jaz launcher provisioning, and patched JDK digest rollout under a live workload |
both + fixtures |
| 19 | dependency CVE remediation across environments | both + fixtures |
Run Tier 17 alone against a dedicated fresh node. It refuses existing shim,
installation-record, or nonempty staging-root state; --reset only resets
Kubernetes test state and does not make a used node fresh. A multi-tier run
(including the default all-tiers run) therefore runs Tier 17 before every other
Kubernetes tier, uses the first fresh node, and reports SKIP with instructions
if no node is fresh; run alone, Tier 17 FAILs on a used node. The CI matrix
runs it separately on a new kind cluster. Never clear retained host files or
manufacture a safety record to satisfy this preflight.
See AGENTS.md for cluster requirements, cleanup, and troubleshooting.
PR selection is documented in Contributing.
python3 integration-tests/e2e/live/smoke.py runs one install/provision/publish/
deploy/serve cycle against its own disposable cluster using checkout-built
components and the Maven plugin's generated manifest. It is the only cluster
smoke in PR CI; exhaustive cluster tiers and live scenarios run nightly or by
explicit E2E dispatch, including pre-merge validation of high-risk changes.
It shares the strict live fixtures' prerequisites, cleanup, evidence, and
cold-start GC deferral; it is not a replacement for the full nightly scenarios.
Monitoring runs from the GitHub Copilot app
Enter /e2e in a GitHub Copilot app session for this repository to open or
focus the Brewlet E2E Tests canvas. The command is a repository skill in
.github/skills/e2e/SKILL.md; if it was added during an existing session, use
/skills reload to discover it. Opening the panel does not start, reset, or
stop tests. Select a suite and start a run from the canvas when ready.
The project extension in .github/extensions/e2e-monitor/ adds an
Integration tests canvas. It starts run.sh tiers, the live HPA/admission
scenarios, or the offline safeguard tests as detached processes, and shows
per-tier progress and ETA, assertion results, failures, a live log, and the
E2E_WORK artifacts. Run records are kept in the session's state directory
under e2e-runs/, so they survive extension reloads.
When the selected kube context is a live cluster (anything other than Docker
Desktop, kind, k3d, minikube, Rancher Desktop, OrbStack or Colima), the canvas
shows a Node pool picker listing the cluster's pools and their Ready nodes,
defaulting to javaworkers. The run gets E2E_NODE_ACCESS=kubectl and
E2E_POOLS=<pool> automatically. Values set in Environment overrides win.
Kubernetes CLI API integration
The process/API integration suite lives in the existing Kubernetes Go module so it can reuse envtest and the production controllers:
make -C kubernetes test-cli-integration
It builds the CLI and invokes real kubectl/Helm against a fresh API server and etcd using only private fixture kubeconfigs. No Docker daemon, existing cluster, or tier reset is used. Missing prerequisites fail this explicit target. The Kubernetes pull-request CI job runs the same tests.
See CLI integration tests for coverage and limitations. In particular, fixture readiness and node inventory are simulated; this is not coverage of host provisioning or JVM execution.
Directories
¶
| Path | Synopsis |
|---|---|
|
e2e
|
|
|
live
command
This deliberately untrusted competing verifier is used only to prove that its success cannot replace Brewlet verification and its failure cannot veto it.
|
This deliberately untrusted competing verifier is used only to prove that its success cannot replace Brewlet verification and its failure cannot veto it. |