Production TUI tests

Verify a Helm-installed control plane, native TUI interaction, and revision cutover on a disposable k3d cluster. Prepare the shared requirements first.

Production TUI Helm test environment

production-tui-k3d-real.test.mjs is the opt-in end-to-end production TUI proof. It installs the actual Helm chart into the selected disposable k3d cluster, starts task-owned PostgreSQL and HTTPS operator proxy Pods, provisions a Namespace and embedded OpenClaw Agent through the production API, drives the native TUI with a PTY, verifies revision cutover, and records nonsecret evidence as it progresses. The test sets agents.defaults.skipBootstrap to true in the disposable demo Agent Configuration so fresh-workspace BOOTSTRAP.md onboarding does not replace the nonce reply; existing workspaces with bootstrap files are unaffected. Missing prerequisites fail the selected test instead of skipping. Allow at least 4 GiB per gateway container because the test runs the TUI as a second OpenClaw process inside the embedded gateway.

Use the production TUI suite only with image references that already exist in the selected cluster, including the Node, PostgreSQL, controller, and runtime digests. Default cleanup uninstalls the Helm release and deletes only the task-owned namespaces. OCC_TEST_PRODUCTION_TUI_KEEP=1 changes that finalizer for operator rehearsal: it keeps the owned setup running, leaves an executable attach.sh for the final gateway TUI session, and writes proof.json with the cluster, image, Namespace, Agent, revision, Pod, and nonce-response evidence. Do not treat an in-progress run as passing live proof until the test completes.

When both upgrade image variables are set, the test runs a controller-only release and then a runtime-only release after the initial revision and model proof. It verifies that the controller release retains every Agent revision and that the runtime release keeps the controller digest, replaces two running Agents, preserves one stopped Agent, and completes a fresh model turn. Setting only one variable fails the selected test. Omitting both keeps the original installation and TUI coverage but does not prove either upgrade path.

Run the suite

Put the required inputs in the private environment file after importing the selected images into the disposable cluster:

OCC_TEST_PRODUCTION_TUI_REAL=1 \  node --env-file="$TEST_ENV_FILE" --test tests/integration/production-tui-k3d-real.test.mjs

Run an isolated upgrade qualification lane

Use the production-tui lane from the reviewed source revision, including the upgrade implementation under test. In the private environment file, set OPENAI_API_KEY, OCC_TEST_OPENAI_MODEL, OCC_TEST_PRODUCTION_POSTGRES_IMAGE, and OCC_TEST_PRODUCTION_NODE_IMAGE. Select the compatible prior release in OCC_TEST_PRODUCTION_CONTROLLER_IMAGE and OCC_TEST_KUBERNETES_RUNTIME_IMAGE, and the reviewed candidate in both OCC_TEST_PRODUCTION_UPGRADE_CONTROLLER_IMAGE and OCC_TEST_PRODUCTION_UPGRADE_RUNTIME_IMAGE. Each reference must use an immutable digest; each candidate must differ from its matching baseline. Verify image provenance separately. Supplying either candidate makes all four images mandatory and selects supplied images instead of building the current checkout.

With Docker, k3d, kubectl, Helm, yq, and the built OCC CLI available, run:

run_dir="$(mktemp -d)"node --env-file="$TEST_ENV_FILE" scripts/ci/prepare.mjs \  --lane production-tui --state "$run_dir/state.json"node --env-file="$TEST_ENV_FILE" scripts/ci/run-tests.mjs run production-tui \  --state "$run_dir/state.json" --results "$run_dir/results.json"install -m 600 "$run_dir/state.json" "$run_dir/state-before-cleanup.json"node scripts/ci/cleanup.mjs --state "$run_dir/state.json"

Run cleanup even if preparation or the test fails; retain the private state if cleanup reports a surviving resource. The private copy preserves the image mapping after successful cleanup removes its working state. Use the same image selection for both commands. Preparation creates a loopback-only disposable cluster, an isolated PostgreSQL service, and a logging backend. The test also creates a separate PostgreSQL Pod with migration and application roles. It imports each approved source digest into the owned k3d nodes and supplies an immutable local platform manifest reference to the test. The private state records the original source, host image ID, and imported reference; the local alias can differ from a registry index digest, so this is not proof of exact registry-reference deployment. Do not print or publish the private state or environment file.

Confirm the test's evidence includes both independent image upgrades and a fresh model reply. A passing TUI test with both candidate variables omitted does not prove upgrade behavior. The local cluster result is not live EKS proof.

Related

• Choose another test suite.
• Results, cleanup, and troubleshooting.