New exceptions: DatabaseBusyError, DatabasePermissionDeniedError,
DatabasePathInvalidError, DatabaseCorruptedError, DatabaseUnavailableError.
open_db creates parent directory if missing. Catches all aiosqlite errors
and maps to specific exception types.
get_db retries up to 3x on locked database with backoff.
Propagates specific exceptions instead of generic HTTPException.
Tests for all new error types and retry behavior.
- Remove structlog dependency from backend/pyproject.toml
- Add app.utils.logging_compat shim for keyword-arg logging API
- Add app.utils.json_formatter for JSON log output with extra fields
- Update all backend modules to use logging_compat.get_logger()
- Update docstrings in log_sanitizer.py and json_formatter.py
- Update test comment in test_async_utils.py
- Record 406 failing tests in Docs/Tasks.md for tracking
- All backend routers moved to /api/v1/ prefix
- Frontend BASE_URL updated to /api/v1
- Setup redirect middleware updated to redirect to /api/v1/setup
- Health router path fixed: prefix=/api/v1/health, @router.get('')
- conftest.py: set server_status=online for test fixture
- Created Docs/API_VERSIONING.md with deprecation policy
- Updated Docs/Backend-Development.md with versioning section
- Updated Instructions.md curl examples
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Adds support for gradual session secret rotation without forcing logout:
- Add BANGUI_SESSION_SECRET_PREVIOUS config field for rotation window
- Implement unwrap_session_token_with_rotation() to accept tokens signed with
either current or previous secret
- Update validate_session() to transparently accept old tokens during rotation
- Update logout() to accept tokens from both secrets
- Add comprehensive logging for rotation events and metrics
- Add 8 new tests covering all rotation scenarios
- Update documentation with step-by-step rotation strategy
- Update .env.example with previous secret field
Key features:
- No forced logout: old tokens continue working during rotation window
- Transparent validation: old tokens are automatically logged for monitoring
- Production-safe: can rotate secrets without service interruption
- Metrics-ready: logs track token rotation for observability
Rotation workflow:
1. Generate new secret and set BANGUI_SESSION_SECRET
2. Set BANGUI_SESSION_SECRET_PREVIOUS to old secret
3. Wait for old tokens to expire (≥ session_duration_minutes)
4. Unset BANGUI_SESSION_SECRET_PREVIOUS to complete rotation
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
CRITICAL FIX: Background tasks (especially blocklist_import) crashed mid-execution,
leaving partial state. On retry, the same bans were applied again, causing duplicates.
Solution: Content-hash based operation tracking for blocklist imports:
- Added import_runs table (migration 6) to track operations by source + content hash
- Before banning, check if this exact content has already been imported
- If completed: skip banning (already done), optionally re-warm cache
- If new or failed: proceed with ban and mark as completed or failed
Changes:
- Database: Migration 6 adds import_runs table with operation state tracking
- Model: Added ImportRunEntry for import run records
- Repository: New import_run_repo module with CRUD operations
- Workflow: Updated blocklist_import_workflow to check operation history before banning
- Dependencies: Registered import_run_repo for dependency injection
- Tests: Added test_import_source_idempotent_on_retry and test_import_source_different_content_not_reused
- Documentation: Added Task Idempotency section to Backend-Development.md
Verification:
- All 7 import tests pass (5 existing + 2 new idempotency tests)
- Type checking: mypy --strict ✅
- Linting: ruff ✅
- No API changes, backwards compatible via automatic migration
Fixes: Background tasks not idempotent #CRITICAL
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Move all repository imports (session_repo, blocklist_repo, import_log_repo,
settings_repo, history_archive_repo, geo_cache_repo, fail2ban_db_repo) and
service imports (auth_service, health_service, default_fail2ban_metadata_service)
to module level in app/dependencies.py.
This eliminates the pattern of local imports inside provider functions,
providing consistency and reducing import overhead. The from app.db import
open_db remains a local import since it's only used within get_db().
- Verified no circular dependencies exist
- All repository and service provider functions simplified to return modules
- Updated Architekture.md § 2.3 to document the module-level import pattern
- All tests pass (28 dependency + auth tests)
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Add comprehensive 'Dependency Wiring and Service Composition' section to
Architekture.md (§ 2.3) documenting:
* The lightweight FastAPI Depends() pattern used as composition root
* Service composition through explicit parameter passing
* Service context dependencies pattern (SessionServiceContext, etc.)
* Repository boundary enforcement
* Lifecycle and scope management
* Checklist for adding new services
- Update Backend-Development.md to reference the new Architecture section
from the 'Dependency Layering' section
- Enhance dependencies.py module docstring with clear explanation of:
* Composition root pattern
* Explicit over implicit principles
* Service context dependencies
* Repository boundary enforcement
This resolves issue #39 by providing clear guidance on dependency wiring
without over-engineering. The pattern uses FastAPI's built-in Depends()
framework and avoids heavyweight container libraries, keeping the solution
lightweight and maintainable.
Fixes: #39
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This commit enforces the repository boundary by eliminating direct database connection
dependencies (DbDep) from all routers. Routers now depend on service context dependencies
that combine the database connection with the related repositories.
Changes:
- Add 5 service context dependencies in dependencies.py:
* SessionServiceContext: db + session_repo
* BlocklistServiceContext: db + blocklist_repo + import_log_repo + settings_repo
* SettingsServiceContext: db + settings_repo
* BanServiceContext: db + fail2ban_db_repo
* HistoryServiceContext: db + fail2ban_db_repo + history_archive_repo
- Refactor all 9 routers (auth, bans, blocklist, config_misc, dashboard, geo,
history, jails, setup) to use service contexts instead of DbDep.
- Update Backend-Development.md with clear examples of the new pattern and
documentation of available service contexts.
Rationale:
- Enforces the repository boundary through the dependency system
- Makes database operations explicit and auditable
- Improves testability by allowing service contexts to be mocked
- Prevents accidental direct database access from routers
The deprecated DbDep remains available for backward compatibility with
services that have not yet been refactored, but routers can no longer import it.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Hide raw database connections (DbDep) from routers by removing from public exports
- Maintain DbDep as deprecated export for backward compatibility
- Add _DbDep internal dependency for use by other dependencies like require_auth
- Update module docstring to explain dependency layering rules
- Add comprehensive documentation section on dependency layering to Backend-Development.md
This enforces the architectural boundary where:
- Routers depend on repository dependencies (SessionRepoDep, BlocklistRepositoryDep, etc)
- Services orchestrate operations through repositories
- Only repositories execute SQL queries
The repository boundary is now technically enforced through the dependency injection
system, making it impossible for routers to accidentally bypass repositories and
access the database directly.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
TASK-004: Replace module-level mutable runtime flags in service layer with
injected state holder, eliminating hidden global state and improving testability
and synchronization boundaries.
Changes:
- Create JailServiceState dataclass in app/utils/runtime_state.py to hold
backend capability cache and synchronization lock
- Add JailServiceState as a field in RuntimeState (with default_factory)
- Remove module-level _backend_cmd_supported and _backend_cmd_lock from
jail_service.py
- Refactor _check_backend_cmd_supported() to accept state parameter
- Inject JailServiceState into list_jails() and _fetch_jail_summary() via
parameters
- Add get_jail_service_state() dependency provider in app/dependencies.py
- Add JailServiceStateDep type alias for router injection
- Update jails router to receive and pass state to service functions
- Update all tests to use jail_service_state fixture and pass state to functions
- Remove duplicate _MAX_PAGE_SIZE constant definition
- Document mutable state management in Backend-Development.md
- Update Architecture.md to describe JailServiceState and state nesting pattern
Benefits:
- Eliminates global mutable state and associated race conditions
- Makes state visible to callers (not hidden in module scope)
- Enables test isolation (each test gets fresh state)
- Prepares codebase for multi-worker deployments (state can be extracted to
shared backend)
- Synchronization boundaries are now explicit (state.get_backend_cmd_lock())
Compliance:
- All tests pass (17 passed in TestListJails, TestGetJail, TestLockInitialization)
- No ruff linting errors
- Type-safe: JailServiceState properly typed with asyncio.Lock, bool | None
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Remove hidden cross-service coupling by making dependencies explicit through
dependency injection while maintaining backward compatibility via lazy imports.
Key changes:
- history_service and ban_service: Removed direct module-level imports of
fail2ban_metadata_service, added optional service parameters to functions
- Added get_fail2ban_metadata_service() provider to dependencies.py
- Updated history router to inject Fail2BanMetadataService dependency
- history_service functions now use lazy imports in fallback paths for
backward compatibility when service is not explicitly injected
- All test patches updated to use internal _get_fail2ban_db_path() helper
- jail_config_service and jail_service already follow best practices
This pattern prevents circular imports, makes services testable via explicit
mocking, and documents service dependencies clearly.
Fixes: Instructions.md #2 - Hidden cross-service coupling
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Add in-memory rate limiter with per-IP deque tracking of attempt timestamps
- Limit login attempts to 5 per 60 seconds per IP, return 429 on excess
- Add Retry-After header to rate limit responses
- Implement IP extraction utility with proxy trust validation (prevent X-Forwarded-For spoofing)
- Integrate rate limiter into auth router and dependencies
- Add 10-second asyncio.sleep on failed login attempts to further slow brute-force
- Add comprehensive tests for rate limiting (9 new tests, all passing)
- Update Features.md to document login rate limiting
- Update Backend-Development.md with rate limiting conventions and design patterns
- Fix test infrastructure issues: update password to meet complexity requirements
- Fix TestValidateSession tests to use Bearer token authentication
- All tests passing: 23 auth tests + full test suite coverage
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Problem: Repository modules use structural typing to satisfy Protocol interfaces via
cast(). A function rename, parameter change, or signature mismatch would silently pass
mypy but fail at runtime.
Solution (Option B — minimal):
1. Aligned Protocol signatures in protocols.py with actual implementations:
- BlocklistRepository: dict[str, object] → dict[str, Any] (matches implementation)
- ImportLogRepository: dict[str, object] → ImportLogRow (typed model)
- GeoCacheRepository: dict[str, object] → GeoCacheRow; Iterable → Sequence
- HistoryArchiveRepository: dict[str, object] → dict[str, Any]
- ImportLogRepository: async compute_total_pages → sync (matches implementation)
2. Created CI validation script (backend/scripts/validate_repository_protocols.py)
that runs at build time to ensure all repository modules satisfy their Protocol
interfaces. Exit 0 if valid, 1 if any mismatch. Detects:
- Missing functions
- Parameter count mismatches
- Type annotation mismatches
- Return type mismatches
3. Updated backend/app/dependencies.py with explicit docstrings linking each
get_*_repo() provider to Backend-Development.md § 13.7.1, explaining the
module-as-Protocol pattern and that it is intentional and validated.
4. Documented the pattern in Backend-Development.md § 13.7.1:
'Repository Module Pattern — Module-as-Protocol Structural Compatibility'
explaining why the pattern works, risks (silent breakage), and how the
validation mitigates it.
5. Fixed type annotation in history_archive_repo.py:
- get_all_archived_history returns list[dict] → list[dict[str, Any]]
- Imported Any type
Benefits:
- Prevents silent breakage of repository interfaces
- Formalizes the module-as-Protocol pattern as intentional
- CI validation prevents regressions without refactoring cost
- All repository tests pass (53/53)
- mypy --strict passes on modified files
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Instead of returning a bound method (geo_cache.lookup_batch), now inject
the GeoCache instance directly into routers and services. This provides
proper runtime isolation since T-04 made GeoCache a proper object.
Changes:
- Remove get_geo_batch_lookup() dependency provider
- Add GeoCacheDep type alias for injecting GeoCache instances
- Update all routers (bans, blocklist, dashboard, jails) to use GeoCacheDep
- Update ban_service, blocklist_service, jail_service to accept GeoCache
- Update service protocols to match new signatures
- Update docstrings to reference GeoCache methods instead of module functions
All callers now call geo_cache.lookup_batch(...) directly instead of
geo_batch_lookup(...), providing real dependency injection with proper
testing isolation.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The AppState Protocol (lines 42-54) and ApplicationContext dataclass
(lines 57-69) described identical fields, creating maintenance burden.
The Protocol was only used for a single cast() in _build_app_context.
Changes:
- Remove AppState Protocol class
- Import ApplicationState from runtime_state.py
- Replace cast('AppState', request.app.state) with
cast(ApplicationState, request.app.state)
- Remove unused Protocol import
This eliminates the redundancy while maintaining the same typing
guarantees. request.app.state is set to ApplicationState instances
during app initialization in main.py.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Move session cache initialization from per-request _build_app_context to
startup lifespan handler. The session cache type is now decided once at app
startup based on settings, making _build_app_context pure (read-only).
Changes:
- Move cache initialization logic to new _update_session_cache() in main.py
- Call _update_session_cache() during lifespan startup to initialize cache
- Remove three if/elif/elif branches mutating state.session_cache from _build_app_context
- Add cache swap logic to set_runtime_settings() in runtime_state.py to handle
runtime settings changes (e.g., setup wizard updates)
- Keep app.state.session_cache initialization in create_app() for test compatibility
This ensures:
- _build_app_context is pure and doesn't mutate app state on each request
- Session cache configuration decisions are centralized at startup
- Settings changes during runtime (via setup wizard) also trigger cache swap
- Cache initialization logic is isolated in one place
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Create GeoCache class with all mutable state as instance attributes:
- _cache, _neg_cache, _dirty, _geoip_reader, _geoip_initialized, _cache_lock
- All public methods: lookup(), lookup_batch(), lookup_cached_only(), flush_dirty(), load_from_db(), clear(), etc.
Initialization & Dependency Injection:
- Instantiate GeoCache in startup.py and store on app.state.geo_cache
- Add get_geo_cache() dependency function in dependencies.py
- Inject into routes and tasks via FastAPI's dependency system
Backward Compatibility:
- Maintain module-level functions in geo_service.py as deprecated wrappers
- All old callers continue to work through _default_geo_cache instance
- Remove test-escape-hatch functions (clear_cache, clear_neg_cache moved to methods)
Background Tasks:
- Update geo_cache_flush.py and geo_re_resolve.py to receive GeoCache instance
- Tasks now operate on injected instance rather than module globals
Tests:
- Refactor test_geo_service.py with geo_cache fixture providing fresh instances
- Update patch paths to target GeoCache methods correctly
- Fix internal state assertions to access instance attributes
Documentation:
- Update Architekture.md to document GeoCache as managed stateful service
- Describe cache lifecycle (load on startup, flush periodically, re-resolve stale)
- Note process-local limitations for multi-worker deployments
Fixes violation of Single Responsibility Principle: module no longer owns both
lookup logic and cache lifecycle management. Cache is now a first-class
injectable service with transparent lifecycle.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Use NoOpSessionCache in backend/app/main.py and dynamically switch cache implementation in backend/app/dependencies.py so disabled cache mode remains safe while get_session_cache always returns a valid object.
- Add TYPE_CHECKING guards for runtime-expensive imports (aiohttp, aiosqlite)
- Reorganize imports to follow PEP 8 conventions
- Convert TypeAlias to modern PEP 695 type syntax (where appropriate)
- Use Sequence/Mapping from collections.abc for type hints (covariant)
- Replace string literals with cast() for improved type inference
- Fix casting of Fail2BanResponse and TypedDict patterns
- Add IpLookupResult TypedDict for precise return type annotation
- Reformat overlong lines for readability (120 char limit)
- Add asyncio_mode and filterwarnings to pytest config
- Update test fixtures with improved type hints
This improves mypy type checking and makes type relationships explicit.
- Cache setup_completed flag in app.state._setup_complete_cached after
first successful is_setup_complete() call; all subsequent API requests
skip the DB query entirely (one-way transition, cleared on restart).
- Add in-memory session token TTL cache (10 s) in require_auth; the second
request with the same token within the window skips session_repo.get_session.
- Call invalidate_session_cache() on logout so revoked tokens are evicted
immediately rather than waiting for TTL expiry.
- Add clear_session_cache() for test isolation.
- 5 new tests covering the cached fast-path for both optimisations.
- 460 tests pass, 83% coverage, zero ruff/mypy warnings.
