Three layers, cheapest first: unit, integration, and cluster-backed end to end. Put each
test in the lowest layer that can express the failure. Broad cluster-backed coverage of
these pages lives in the kubevirt-ui Playwright suite, which drives them through the real
console against a real cluster.
npm test # Both Jest projects plus the sanitizer tests
npm run test:coverage # Run with coverage report
npm run test:updateSnapshot # Update snapshotsUnit files are *.test.ts(x) / *.spec.ts(x); the unit project's testMatch in
jest.config.ts excludes *.integration.test.*. The SDK is mocked via src/__mocks__/, so
these tests need no network.
The real component tree, hooks and selectors are rendered against test doubles for
@openshift-console/dynamic-plugin-sdk and react-router; MSW answers the
console-proxy requests (/api/kubernetes/...) that the SDK double in
src/__mocks__/integration/dynamic-plugin-sdk.tsx issues. src/__mocks__/i18n.ts (unit) and
src/__mocks__/integration/i18n.ts (integration) initialize i18next from
locales/en/plugin__nmstate-console-plugin.json, so t and Trans interpolate and
pluralize against the real English catalog.
Files are *.integration.test.tsx. jest.config.ts declares the second project with
testEnvironment: 'jest-fixed-jsdom' and
setupFilesAfterEnv: ['<rootDir>/src/__mocks__/integration/setup.ts'].
The integration project runs as ESM: Jest starts with --experimental-vm-modules (see
npm run test:integration), jest.config.ts treats .ts/.tsx as ESM for that project, and
ts-jest emits ESM. msw 3 is ESM-only and needs Node 22.12 or newer, so the runtime is required
rather than optional; the unit project keeps the CommonJS transform.
Two consequences of that runtime:
- jsdom leaves out Node's
setImmediate, and msw's interceptor schedules request handling with it, sosrc/__mocks__/integration/nodeGlobals.tssupplies it throughsetupFiles. @patternflypackages are CommonJS-typed but ship ESM sources, and the CommonJS build of@patternfly/react-data-viewrequiresreact-tokens/dist/esm/*; that scope is transformed instead of loaded as-is, and ts-jest emits each file in the format Jest loads it in. Fixtures and other JSON import withwith { type: 'json' }, since Jest treats a bare JSON import as deprecated under ESM.
npm run test:integration # Integration project only
npm run test:unit # Unit project onlyPayloads come from src/__mocks__/integration/recordings/*.json, recorded from a live
cluster, never hand-written. paths.ts holds consoleApiPath and the baseline empty-list
handlers for nodes, nodenetworkconfigurationpolicies,
nodenetworkconfigurationenactments and nodenetworkstates, so an ambient watch answers
empty instead of failing; onUnhandledFrame: 'error' keeps anything outside those four
loud.
Menus, drawers and the state-list filters are driven through the real PatternFly controls;
assert what the user sees, not component props. The policy-list filter is the exception: the
SDK double renders ListPageFilter as plain checkboxes. A healthy run prints no console
output — a React, PatternFly, MSW or YAML warning is a defect in the test or the markup, not
background noise.
scripts/record-integration-fixtures.mjs reads the current oc context and rewrites
src/__mocks__/integration/recordings/{nodes,nns,nncps,nnces}.json plus
policy-{uplink,settings,editable}.json. --context <name> pins a cluster; --only <items>
narrows the run and accepts nodes,nns,nncps,nnces,shapes.
Set the cluster once, as CTX below, and pass it to the fixture policy, the recording and
the cleanup, so those steps can never touch different clusters.
Policy shapes come from oc create --dry-run=server: the API server returns the object it
would persist and nothing is created.
The recorder sanitizes as it writes — node names to master-N/worker-N, IP/MAC into
documentation ranges, domains derived from node names to example.com, host identifiers
(machineID, systemUUID, bootID, providerID, osImage, image lists) dropped,
per-operator and environment labels trimmed — and refuses to overwrite nncps.json/nnces.json
with an empty list. Only node-name domains are rewritten, so review a fresh recording for
internal domains and other identifying data before committing it.
It rewrites all four list files, so capture while the fixture policy the suites assert on is
applied: a pw-int-fixture policy that brings dummy0 up and reports three available
enactments. Use a disposable cluster and confirm the names are unused first: the policy
changes node network state, and the cleanup removes dummy0 from every selected worker.
CTX=<cluster-name>
oc --context "$CTX" get nncp pw-int-fixture # expect: Error from server (NotFound)
oc --context "$CTX" get nns -o json | grep -c '"dummy0"' # expect: 0
oc --context "$CTX" apply -f - <<'EOF'
apiVersion: nmstate.io/v1
kind: NodeNetworkConfigurationPolicy
metadata:
name: pw-int-fixture
spec:
nodeSelector:
node-role.kubernetes.io/worker: ""
desiredState:
interfaces:
- name: dummy0
state: up
type: dummy
EOF
oc --context "$CTX" wait --for=condition=Available nodenetworkconfigurationpolicies.nmstate.io/pw-int-fixture --timeout=5m
node scripts/record-integration-fixtures.mjs --context "$CTX"Deleting an NNCP removes its enactments but does not revert what it applied: apply a
cleanup policy that sets the interface to state: absent first, wait for it to converge,
then delete the policies, all with --context "$CTX".
Re-recording on a different cluster changes node counts and enactment totals, so update the literal expectations in the specs with the recordings.
LLDP coverage uses the checked-in cypress/fixtures/NodeNetworkStatusWithLLDP.json,
sanitized in place by scripts/sanitize-cypress-fixtures.mjs.
npm run test:coverage reports only the files the suites import. For whole-source numbers:
NODE_OPTIONS=--experimental-vm-modules npx jest --coverage \
--collectCoverageFrom='src/**/*.{ts,tsx}' \
--collectCoverageFrom='!src/**/*.test.{ts,tsx}' \
--collectCoverageFrom='!src/**/__mocks__/**' \
--collectCoverageFrom='!src/**/*.d.ts'Topology layout and zoom have no meaningful jsdom assertion, and nmstate apply and converge semantics, RBAC and non-privileged paths, console shell navigation and plugin registration, and LLDP discovery across multiple nodes all need a real cluster; that is where they stay. What belongs here is "given this API payload, is this rendering correct" — it runs on every PR without a cluster.