project-testing

name	project-testing
description	Guide for writing and running tests in kubectl-mtv, including Go unit tests, MCP e2e tests (Python/pytest), and linting. Use when adding tests, running the test suite, or debugging test failures.

Testing

Quick Reference

Command	Purpose
make test	Unit tests with coverage
make lint	go vet + golangci-lint
make fmt	Format Go code
make test-e2e	CLI e2e smoke tests (requires cluster)
make test-e2e-mcp	E2E MCP tests (local binary)
make test-e2e-mcp-image	E2E MCP tests (container image)

Unit Tests (Go)

Tests live alongside source files: pkg/mcp/tools/mtv_read_test.go next to mtv_read.go.

Table-Driven Pattern

func TestMyFunction(t *testing.T) {
    tests := []struct {
        name     string
        input    string
        expected string
        wantErr  bool
    }{
        {name: "valid input", input: "foo", expected: "bar"},
        {name: "empty input", input: "", wantErr: true},
    }

    for _, tt := range tests {
        t.Run(tt.name, func(t *testing.T) {
            result, err := MyFunction(tt.input)
            if tt.wantErr {
                if err == nil {
                    t.Errorf("expected error, got nil")
                }
                return
            }
            if err != nil {
                t.Fatalf("unexpected error: %v", err)
            }
            if result != tt.expected {
                t.Errorf("got %q, want %q", result, tt.expected)
            }
        })
    }
}

Test Helpers

Common helpers used in the codebase:

• testRegistry() -- creates a registry from test data for MCP tool tests
• buildTestBinary() -- builds the kubectl-mtv binary for integration-style tests
• testdataPath() -- resolves paths to testdata/ fixtures
• findRepoRoot() -- locates the repo root from any test directory
• loadRealRegistry() -- loads testdata/help_machine_output.json for registry tests

Use t.Helper() in shared helper functions and t.Skipf() when fixtures or binaries are unavailable.

Test Data

Fixtures live in testdata/ directories next to test files:

• pkg/mcp/discovery/testdata/help_machine_output.json -- snapshot of help --machine output

Running

make test                    # All unit tests with coverage
go test ./pkg/mcp/tools/...  # Specific package
go test -run TestMyFunc ./pkg/...  # Specific test

Coverage output: coverage.out (generated by make test).

CLI E2E Smoke Tests (Python)

Located in tests/e2e_smoke.py. Tests the CLI binary end-to-end against a live OpenShift cluster with MTV/Forklift installed.

What It Tests

• version: --client, full version, JSON output
• health: default, --skip-logs, JSON/markdown output
• settings: list, --all, get, specific setting, JSON/YAML output
• get: providers, plans, mappings (all formats), dynamic name lookups
• describe: providers and plans (dynamically discovered)
• positional args: get provider <name>, describe provider <name>, get plan <name>, describe plan <name>
• create/patch/delete lifecycle: creates a test namespace, creates an OpenShift provider, gets/describes/patches/deletes it, cleans up
• help: topics (tsl, karl), --machine schema (JSON/YAML, --read-only, --write, --short)
• error handling: missing required flags, conflicting flags, unknown commands

Running

make test-e2e                    # Build + run smoke tests
make build && python3 tests/e2e_smoke.py  # Manual

Pattern

Uses stdlib only (no pytest). A main() calls test_* functions, each invoking the binary via subprocess.run with 60s timeout. Helpers assert_exit_ok, assert_exit_fail, assert_contains, assert_valid_json track pass/fail counts. Exits 0 on all pass, 1 on any failure.

Dynamic tests (describe, get by name) discover resources at runtime and skip gracefully if none exist.

E2E MCP Tests (Python/pytest)

Located in e2e/mcp/. These test the MCP server end-to-end via Streamable HTTP.

Structure

e2e/mcp/
├── conftest.py           # Shared fixtures and helpers
├── pyproject.toml        # pytest config, dependencies
├── Makefile              # Server lifecycle + test targets
├── setup/                # Setup verification tests
├── providers/            # Provider CRUD tests
├── hosts/                # Host CRUD tests
├── plans/                # Plan CRUD tests
├── mappings/             # Mapping read tests
├── inventory/            # Inventory read tests
├── health/               # Health check tests
└── auth/                 # Auth tests

Key Fixtures (conftest.py)

• mcp_session (session) -- Streamable HTTP ClientSession using MCP_HTTP_URL with K8s auth
• mcp_server_process (session) -- verifies server reachability
• cleanup_test_resources (session, autouse) -- deletes test resources after suite

Helper: call_tool

result = await call_tool(mcp_session, "mtv_read", {
    "command": "get plan",
    "flags": {"namespace": "my-ns", "output": "json"}
})

Environment Variables

Required: GOVC_URL, GOVC_USERNAME, GOVC_PASSWORD, KUBE_API_URL, KUBE_TOKEN, ESXI_HOST_NAME, COLD_VMS, WARM_VMS, NETWORK_PAIRS, STORAGE_PAIRS.

Optional: MCP_HTTP_URL (default http://127.0.0.1:18443/mcp).

Running

make test-e2e-mcp           # Build binary + start server + run tests + stop
make test-e2e-mcp-image MCP_IMAGE=quay.io/yaacov/kubectl-mtv-mcp-server:latest

Or manually:

cd e2e/mcp
make server-start           # Start MCP server in background
make test                   # Run pytest
make server-stop            # Stop server

Test Ordering

Tests use @pytest.mark.order(N) for execution order. The testpaths in pyproject.toml controls directory order: setup -> providers -> hosts -> plans -> mappings -> inventory -> health.

Linting

make lint    # go vet + golangci-lint
make fmt     # go fmt ./...

Fix lint issues before committing. The CI does not run linting automatically on PR, but it is expected to pass locally.