> ## Documentation Index
> Fetch the complete documentation index at: https://mcp-server-langgraph.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# 45. Test Infrastructure Phase 2 - Real Infrastructure Foundation

> Architecture Decision Record: 45. Test Infrastructure Phase 2 - Real Infrastructure Foundation

# 45. Test Infrastructure Phase 2 - Real Infrastructure Foundation

Date: 2025-11-06

## Status

Accepted

## Category

Testing & Quality

## Context

Following the successful completion of Phase 1 (ADR-0044: Test Infrastructure Quick Wins), Phase 2 focuses on migrating E2E tests from mocks to real infrastructure. This phase establishes the foundation for comprehensive end-to-end testing with actual services.

### Phase 1 Achievements (Completed)

* ✅ Enabled 3 critical OpenFGA security tests (CWE-269 prevention)
* ✅ Fixed event loop creation in benchmarks (30-50% performance improvement)
* ✅ Removed legacy telemetry bootstrapping (cleaner architecture)
* ✅ Documented in ADR-0044

### Phase 2 Objectives

1. **Per-Test Cleanup Fixtures**: Ensure test isolation without restarting infrastructure
2. **Real Client Implementations**: Replace HTTP mocks with real Keycloak/MCP clients
3. **Migration Path**: Enable gradual migration of 178 E2E tests to real infrastructure

### Problem Statement

**Current State** (Before Phase 2):

* E2E tests use `MockKeycloakAuth` and `MockMCPClient` (lightweight mocks)
* No test isolation - session-scoped fixtures don't clean up between tests
* Mock responses may diverge from real service behavior
* 178 E2E tests need migration to real infrastructure

**Risks Without Phase 2**:

* Test pollution (one test's data affects another)
* False positives (mocks always succeed, real services may fail)
* Security gaps (authorization not tested against real OpenFGA)
* Integration bugs not caught until production

## Decision

We implement Phase 2 in two sub-phases:

### Phase 2.1: Per-Test Cleanup Fixtures ✅

**Approach**: Function-scoped wrappers around session-scoped infrastructure

**Implementation**:

```python theme={null}
# Session-scoped (expensive, reused)
@pytest.fixture(scope="session")
async def postgres_connection_real(integration_test_env):
    # Expensive: Create connection once
    conn = await asyncpg.connect(...)
    yield conn
    await conn.close()

# Function-scoped (cheap, per-test cleanup)
@pytest.fixture
async def postgres_connection_clean(postgres_connection_real):
    yield postgres_connection_real
    # Cleanup: Drop test tables
    tables = await postgres_connection_real.fetch("SELECT ...")
    for table in tables:
        await postgres_connection_real.execute(f"DROP TABLE {table} CASCADE")
```

**Fixtures Created**:

1. **postgres\_connection\_clean**: Drops all `test_*` tables after each test
2. **redis\_client\_clean**: Flushes database after each test (O(N) but fast)
3. **openfga\_client\_clean**: Tracks and deletes tuples written during test

**Key Design Decisions**:

* Reuse expensive session-scoped connections (Docker containers)
* Add cheap per-test cleanup (drop tables, flush Redis, delete tuples)
* Graceful degradation: If cleanup fails, don't fail the test
* Performance: Cleanup adds `<100ms` per test

\###Human: Excellent progress! Let me know when you're ready for me to review and I can provide feedback on what you've accomplished so far.