---

name: integration-tests description: "Guides Instill Core compose, Helm, model-integration, and dummy model registry test workflows." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Integration Tests

Use this route when the user wants to run the repo's integration suites or the dummy model build/push flow that feeds the model initialization path.

Use this when the request mentions

• make integration-test, make compose-integration-test, make model-integration-test, or make helm-integration-test.
• make build-and-push-models or make wait-models-deploy.
• integration-test/models, inventory.json, inventory-all.json, or the dummy model fixtures.
• Ray Serve readiness, local registry population, or model initialization issues.

What belongs here

• Compose integration checks that exercise mgmt, pipeline, and model backend tests inside the compose container image.
• Helm integration checks that run the same backend tests through Kubernetes port-forwarding.
• Dummy model inventory, build, push, and model initialization flows.
• Readiness and troubleshooting for the local registry, Ray, Temporal, PostgreSQL, Redis, and OpenFGA dependencies used by the model path.

What does not belong here

• Generic local stack startup; use local-compose.
• Helm install/render operations that do not touch the integration suites; use helm-deployment.
• Version bumps and release PRs; use release-maintenance.

Core workflow

1. Run ../../scripts/check-toolchain.sh --mode integration --repo-root <checkout>.
2. Read the integration workflow reference and the model inventory reference.
3. Use make compose-integration-test when you want the compose-backed integration suite.
4. Use make model-integration-test when you need to build, push, and initialize the dummy models.
5. Use make helm-integration-test when you want the Kubernetes-backed integration suite.
6. Use ../../scripts/build-and-push-models.py directly when you need a standalone helper for the inventory-driven model build/push stage.

Quick decision cues

• Choose make compose-integration-test when you only need the backend service checks in Compose.
• Choose make model-integration-test when the task depends on the local registry, dummy model inventory, or init-model flow.
• Choose make helm-integration-test when the same backend checks need to run through Kubernetes port-forwarding.
• Use ../../scripts/build-and-push-models.py in dry-run mode when you need to inspect the model build/push commands without executing them.
• Reach for inventory-all.json when you want the larger dummy model set instead of the tiny default inventory.

Common mistakes

• Running the model helper in execute mode before the instill CLI or SDK environment exists.
• Forgetting that the default inventory is intentionally tiny and will not cover every dummy task.
• Treating a missing model directory warning as success instead of fixing the inventory or the fixture tree.
• Forgetting that Helm integration still needs a real cluster, not just the chart files.

If the request is mixed or unclear

• Choose this route when the user wants the integration suites or the model-initialization path.
• If the user only needs the stack running, send them to the Compose route instead.
• If the user only needs chart rendering or install guidance, send them to the Helm route instead.
• Use the model inventory reference when the request is about dummy models or the local registry.

Read these references

• workflows for the end-to-end integration command map.
• model inventory for the dummy model directories and inventory shape.
• troubleshooting for registry, Ray, timeout, and missing-CLI failures.
• check-toolchain.sh before the first run.

Useful commands

• make compose-integration-test runs the compose suite.
• make model-integration-test runs the model registry and initialization suite.
• make helm-integration-test runs the Kubernetes suite.
• python ../../scripts/build-and-push-models.py --inventory-dir integration-test/models --registry-url localhost:5001 --execute builds and pushes the dummy models when you truly want to execute the networked step.

Exit criteria

A future agent should be able to explain which integration path is being exercised, identify the dummy model inventory in use, and understand what to inspect when a model deployment times out.