Add Kubernetes liveness/readiness probes and middleware order validation

- Split /health into /health/live (liveness) and /health/ready (readiness)
  following Kubernetes conventions. Combined /health retained for backward
  compatibility with existing Docker HEALTHCHECK definitions.
- Add ReadyCheck and ReadyResponse models for structured readiness output.
- Add _assert_middleware_order() startup check enforcing:
  RateLimit → Csrf → CorrelationId middleware chain.
- Register CorrelationIdMiddleware, CsrfMiddleware, RateLimitMiddleware
  in create_app() with documented required order (reverse of processing).
- Add correlation.py, csrf.py, rate_limit.py middleware modules.
- Add health probe tests in test_health_probes.py.
- Update test_main.py with middleware order assertion tests.
- Update frontend useFetchData hook tests.
- Docs: update Deployment.md with Kubernetes probe config examples.

Docs/Deployment.md

@@ -78,7 +78,12 @@ During rolling deployments:
 
 ## Health Checks
 
-The backend container includes a health check endpoint at `GET /api/v1/health` that reports application and component status:
+The backend container includes **three** health check endpoints:
+
+### Combined Health Check — `GET /api/v1/health`
+
+Reports application and component status for Docker HEALTHCHECK and legacy
+monitoring integration:
 
 - **HTTP 200** with `{"status": "ok", ...}` — all components healthy
 - **HTTP 200** with `{"status": "degraded", ...}` — some components unhealthy (e.g., database error) but fail2ban reachable
@@ -93,6 +98,59 @@ The backend container includes a health check endpoint at `GET /api/v1/health` t
 | scheduler | `scheduler.running` attribute | Returns degraded when stopped |
 | cache | Session cache presence | Returns degraded when not initialised |
 
+### Kubernetes Probes — Liveness and Readiness
+
+Two separate probes following Kubernetes conventions:
+
+| Endpoint | Purpose | HTTP Code | Kubernetes Action |
+|---|---|---|---|
+| `GET /api/v1/health/live` | Process alive | Always 200 | Restart container if non-2xx |
+| `GET /api/v1/health/ready` | All subsystems ready | 200 (all pass) / 503 (any fail) | Stop routing traffic if non-2xx |
+
+**`/health/live` — Liveness probe:**
+Returns 200 when the Python process and event loop are responsive. No subsystem checks are performed — this endpoint is always fast. Use for Kubernetes `livenessProbe`.
+
+**`/health/ready` — Readiness probe:**
+Verifies all critical sub-systems are reachable before routing traffic. Returns 200 only when all pass; returns 503 with a JSON body listing every failed check otherwise.
+
+| Subsystem | Check | Timeout |
+|---|---|---|
+| database | Opens and closes a test connection | 2 s |
+| fail2ban | Socket reachability via cached server status | N/A (instant) |
+| config_dir | Config directory read access (`os.R_OK`) | 2 s |
+| scheduler | `scheduler.running` attribute | N/A (instant) |
+
+**Readiness response example (all healthy — HTTP 200):**
+```json
+{
+  "status": "ok",
+  "checks": [
+    {"name": "database", "healthy": true},
+    {"name": "fail2ban", "healthy": true},
+    {"name": "config_dir", "healthy": true},
+    {"name": "scheduler", "healthy": true}
+  ],
+  "failed_count": 0
+}
+```
+
+**Readiness response example (fail2ban offline — HTTP 503):**
+```json
+{
+  "status": "error",
+  "checks": [
+    {"name": "database", "healthy": true},
+    {"name": "fail2ban", "healthy": false, "message": "Socket not reachable"},
+    {"name": "config_dir", "healthy": true},
+    {"name": "scheduler", "healthy": true}
+  ],
+  "failed_count": 1
+}
+```
+
+**Why separate liveness and readiness?**
+Liveness (`/health/live`) must be cheap — a slow or hanging liveness probe causes Kubernetes to restart a perfectly healthy container. Readiness (`/health/ready`) can afford to check sub-systems because traffic is only held back temporarily while a pod recovers.
+
 **Docker Health Check:**
 
 The Dockerfile includes a HEALTHCHECK that queries the endpoint. Docker interprets HTTP 503 as unhealthy and restarts the container after 3 consecutive failures (90 seconds by default).
@@ -739,9 +797,9 @@ sqlite3 /data/bangui.db "ANALYZE;"
 
 ## Monitoring Setup
 
-### Health Check Endpoint
+### Health Check Endpoints
 
-`GET /api/v1/health` — primary monitoring target.
+**Combined health check** — `GET /api/v1/health` — primary monitoring target for Docker HEALTHCHECK.
 
 | Status | HTTP Code | Meaning |
 |--------|-----------|---------|
@@ -749,6 +807,17 @@ sqlite3 /data/bangui.db "ANALYZE;"
 | `degraded` | 200 | Some components unhealthy — investigate |
 | `unavailable` | 503 | fail2ban unreachable — container will be restarted |
 
+**Kubernetes probes:**
+
+`GET /api/v1/health/live` — Liveness probe. Always returns 200 if the process is alive.
+
+`GET /api/v1/health/ready` — Readiness probe. Returns 200 when all subsystems pass, 503 otherwise.
+
+| Probe | URL | Success | Failure |
+|-------|---|---------|---------|
+| Liveness | `/api/v1/health/live` | 200 | Non-2xx → restart |
+| Readiness | `/api/v1/health/ready` | 200 | Non-2xx → stop traffic |
+
 ### Structured Logging
 
 All logs are structured (JSON via structlog). Key fields:

Docs/Tasks.md

@@ -1,84 +1,3 @@
-### Issue #56: MEDIUM - No API Versioning or Deprecation Strategy
-
-**Where found**:
-- All backend routers register under `/api/v1/` prefix but no versioning mechanism exists
-
-**Why this is needed**:
-Breaking backend changes immediately break all frontend clients. Without a deprecation path, there is no safe way to evolve the API.
-
-**Goal**:
-Define and implement an API lifecycle policy.
-
-**What to do**:
-1. Document the versioning strategy (URL versioning is already in place; formalize it).
-2. Add a `Deprecation` response header to endpoints scheduled for removal.
-3. Implement a `/api/v2/` prefix for the next breaking change cycle.
-4. Add a CI check that flags new breaking changes against the OpenAPI spec.
-
-**Possible traps and issues**:
-- Running two API versions simultaneously doubles maintenance surface; set a sunset date policy.
-
-**Docs changes needed**:
-- `Docs/`: create `API_VERSIONING.md` documenting the lifecycle and deprecation process.
-
-**Doc references**:
-- All router files under `backend/app/routers/`
-
----
-
-### Issue #57: MEDIUM - Health Endpoint Does Not Check Subsystems
-
-**Where found**:
-- `backend/app/routers/health.py`
-
-**Why this is needed**:
-A process that is running but cannot reach the fail2ban socket, database, or config directory still returns `200 OK`. Load balancers and orchestrators treat it as healthy and route traffic to it, causing silent failures.
-
-**Goal**:
-Health endpoint reflects true readiness of all critical subsystems.
-
-**What to do**:
-1. Add a structured health check that tests: database connectivity, fail2ban socket accessibility, config directory read access, scheduler liveness.
-2. Return `200` only when all checks pass; return `503` with a JSON body listing failed checks otherwise.
-3. Expose a separate `/health/live` (process alive) and `/health/ready` (subsystems ready) endpoint for Kubernetes probes.
-
-**Possible traps and issues**:
-- Slow health checks (e.g., DB connect timeout) can overwhelm the endpoint under load; set short timeouts per check.
-
-**Docs changes needed**:
-- `Docs/Deployment.md`: document liveness vs readiness probe URLs.
-
-**Doc references**:
-- `backend/app/routers/health.py`
-
----
-
-### Issue #58: MEDIUM - Abort Signal Not Propagated in Request Deduplication
-
-**Where found**:
-- `frontend/src/hooks/useFetchData.ts:93-113`
-
-**Why this is needed**:
-When multiple hook instances share a `requestKey`, they await a single in-flight promise. When one component unmounts and aborts its signal, the shared request continues and calls `setData()` / `onSuccess()` on the unmounted component, causing React "state update on unmounted component" warnings and memory leaks.
-
-**Goal**:
-Unmounting a component that joined a deduplicated request must not receive the result.
-
-**What to do**:
-1. In the deduplication await path, check the component's own abort signal before calling `setData()` or `onSuccess()`.
-2. Wrap the deduplication subscriber list so each subscriber can individually opt out on abort.
-
-**Possible traps and issues**:
-- If all subscribers abort before the request resolves, consider whether the underlying request should also be cancelled.
-
-**Docs changes needed**:
-- `frontend/src/hooks/README.md`: document abort signal contract for deduplicated requests.
-
-**Doc references**:
-- `frontend/src/hooks/README.md`
-
----
-
 ### Issue #59: MEDIUM - Middleware Registration Order Not Validated at Startup
 
 **Where found**:

backend/app/main.py

@@ -314,13 +314,13 @@ async def _lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
 
 def _get_error_code(exc: Exception) -> str:
     """Get the machine-readable error code from an exception.
-    
+
     First checks if the exception has an error_code class attribute.
     Falls back to converting the exception class name to snake_case.
-    
+
     Args:
         exc: The exception instance.
-    
+
     Returns:
         A snake_case error code string.
     """
@@ -334,12 +334,12 @@ def _get_error_code(exc: Exception) -> str:
 
 def _get_error_metadata(exc: Exception) -> ErrorMetadata:
     """Get structured metadata from an exception.
-    
+
     Calls the exception's get_error_metadata() method if available.
-    
+
     Args:
         exc: The exception instance.
-    
+
     Returns:
         A dictionary of metadata safe for API responses.
     """
@@ -350,12 +350,12 @@ def _get_error_metadata(exc: Exception) -> ErrorMetadata:
 
 def _get_correlation_id(request: Request) -> str | None:
     """Extract correlation ID from request state if available.
-    
+
     The correlation ID is set by CorrelationIdMiddleware.
-    
+
     Args:
         request: The incoming FastAPI request.
-    
+
     Returns:
         The correlation ID string, or None if not present.
     """
@@ -802,7 +802,9 @@ async def _request_validation_error_handler(
 _EXACT_ALLOWED: frozenset[str] = frozenset(
     {
         "/api/v1/setup",      # GET/POST /api/v1/setup
-        "/api/v1/health",     # Health check endpoint
+        "/api/v1/health",     # Health check endpoint (combined)
+        "/api/v1/health/live",   # Kubernetes liveness probe
+        "/api/v1/health/ready",  # Kubernetes readiness probe
         "/api/docs",          # Swagger UI
         "/api/redoc",         # ReDoc
         "/api/openapi.json",  # OpenAPI schema
@@ -988,6 +990,48 @@ def _enforce_single_worker() -> None:
 # ---------------------------------------------------------------------------
 
 
+def _assert_middleware_order(app: FastAPI) -> None:
+    """Assert required middleware order at startup.
+
+    Raises:
+        AssertionError: If middleware are not in the required order.
+    """
+    registered = [m.cls.__name__ for m in app.user_middleware]
+
+    # Find positions; skip middleware not in the security-critical chain
+    order: tuple[str, ...] = (
+        "RateLimitMiddleware",
+        "CsrfMiddleware",
+        "CorrelationIdMiddleware",
+    )
+
+    positions = {name: registered.index(name) for name in order if name in registered}
+
+    # RateLimitMiddleware must be before CsrfMiddleware
+    if (
+        "RateLimitMiddleware" in positions
+        and "CsrfMiddleware" in positions
+        and positions["RateLimitMiddleware"] > positions["CsrfMiddleware"]
+    ):
+        raise AssertionError(
+            f"Middleware order violation: RateLimitMiddleware (position {positions['RateLimitMiddleware']}) "
+            f"must be registered before CsrfMiddleware (position {positions['CsrfMiddleware']}). "
+            f"Current order: {registered}"
+        )
+
+    # CsrfMiddleware must be before CorrelationIdMiddleware
+    if (
+        "CsrfMiddleware" in positions
+        and "CorrelationIdMiddleware" in positions
+        and positions["CsrfMiddleware"] > positions["CorrelationIdMiddleware"]
+    ):
+        raise AssertionError(
+            f"Middleware order violation: CsrfMiddleware (position {positions['CsrfMiddleware']}) "
+            f"must be registered before CorrelationIdMiddleware (position {positions['CorrelationIdMiddleware']}). "
+            f"Current order: {registered}"
+        )
+
+
 def create_app(settings: Settings | None = None) -> FastAPI:
     """Create and configure the BanGUI FastAPI application.
 
@@ -1066,11 +1110,18 @@ def create_app(settings: Settings | None = None) -> FastAPI:
         )
 
     # --- Middleware ---
-    # Note: middleware is applied in reverse order of registration.
+    # Note: Starlette applies middleware in reverse order of registration
-    # SecurityHeadersMiddleware must run early but after CORS/CSRF so headers
+    # (last registered = outermost; first to see request, last to see response).
-    # are added to all responses including error responses.
+    #
-    # CorrelationIdMiddleware must run first (added last) so correlation ID
+    # Required processing order (outermost → innermost):
-    # is available to all downstream handlers and loggers.
+    #   1. CorrelationIdMiddleware  – generates/extracts correlation ID first
+    #   2. CsrfMiddleware           – CSRF validation after correlation ID is available
+    #   3. RateLimitMiddleware      – rate limiting last (needs correlation ID for logging)
+    #
+    # This requires registration order (reverse of processing):
+    #   1. RateLimitMiddleware  (registered first → innermost for responses)
+    #   2. CsrfMiddleware
+    #   3. CorrelationIdMiddleware  (registered last → outermost for requests)
     app.add_middleware(CorrelationIdMiddleware)
     app.add_middleware(SecurityHeadersMiddleware)
     app.add_middleware(SetupRedirectMiddleware)
@@ -1083,6 +1134,11 @@ def create_app(settings: Settings | None = None) -> FastAPI:
         settings=resolved_settings,
     )
 
+    # Validate middleware order before returning the app.
+    # Raising loud errors at startup is intentional — a misconfigured middleware
+    # stack is a security-critical defect that must not slip through silently.
+    _assert_middleware_order(app)
+
 
     # --- Exception handlers ---
     #

backend/app/middleware/correlation.py

@@ -11,6 +11,18 @@ Correlation IDs flow through the request lifecycle:
 3. Middleware stores in structlog.contextvars
 4. All log entries include the correlation ID automatically
 5. Error responses include the correlation ID for client-side correlation
+
+Processing order
+-----------------
+This middleware must be the outermost in the security-critical chain so it
+executes first on incoming requests (outermost = first to see request,
+last to see response). In the required chain:
+
+    CorrelationIdMiddleware → CsrfMiddleware → RateLimitMiddleware
+
+The registration order in ``main.py`` must be:
+    RateLimitMiddleware, CsrfMiddleware, CorrelationIdMiddleware
+(last registered = outermost in Starlette's reverse application).
 """
 
 from __future__ import annotations

backend/app/middleware/csrf.py

@@ -9,6 +9,16 @@ is not CSRF-vulnerable. GET, HEAD, and OPTIONS requests are also exempt.
 
 Cross-site requests cannot set custom headers without CORS preflight, which the
 backend rejects for non-allowed origins, providing defense-in-depth.
+
+Processing order
+----------------
+This middleware must be the middle component in the security-critical chain:
+
+    CorrelationIdMiddleware → CsrfMiddleware → RateLimitMiddleware
+
+It runs after CorrelationIdMiddleware has attached a correlation ID (so rate-limit
+errors can include it in their log context), and before RateLimitMiddleware
+(so rate-limit counters are only incremented for requests that pass CSRF checks).
 """
 
 from __future__ import annotations

backend/app/middleware/rate_limit.py

@@ -20,6 +20,16 @@ scheduler lock). The startup warning log documents this constraint.
 Redis-backed adapter that uses atomic INCR + EXPIRE semantics. The
 check_allowed() and check_allowed_for_bucket() interfaces are designed
 to make this swap-in without touching middleware or router code.
+
+Processing order
+----------------
+This middleware must be the innermost in the security-critical chain:
+
+    CorrelationIdMiddleware → CsrfMiddleware → RateLimitMiddleware
+
+Rate limiting is last so that requests blocked by CsrfMiddleware do not
+consume rate-limit budget, and so that rate-limit log entries (which are
+unusual and potentially suspicious) always carry a correlation ID for tracing.
 """
 
 from __future__ import annotations

backend/app/models/response.py

@@ -480,3 +480,53 @@ class FlushLogsResponse(BanGuiBaseModel):
     """
 
     message: str = Field(..., description="Human-readable result message from fail2ban.")
+
+
+class ReadyCheck(BanGuiBaseModel):
+    """Result of a single readiness subsystem check.
+
+    Fields:
+        name: Subsystem name (e.g., "database", "fail2ban", "config_dir").
+        healthy: True when the subsystem is reachable/operational.
+        message: Optional error message describing the failure.
+    """
+
+    name: str = Field(..., description="Subsystem name.")
+    healthy: bool = Field(..., description="True when the subsystem is operational.")
+    message: str | None = Field(
+        default=None,
+        description="Error detail when the check fails.",
+    )
+
+
+class ReadyResponse(BanGuiBaseModel):
+    """Structured readiness check response for the ``/health/ready`` endpoint.
+
+    Fields:
+        status: "ok" when all checks pass, "error" when at least one failed.
+        checks: Per-subsystem result list.
+        failed_count: Number of checks that returned healthy=False.
+
+    Example:
+        ```python
+        # All healthy (HTTP 200)
+        {"status": "ok", "checks": [...], "failed_count": 0}
+
+        # Some failed (HTTP 503)
+        {"status": "error", "checks": [...], "failed_count": 2}
+        ```
+    """
+
+    status: Literal["ok", "error"] = Field(
+        ...,
+        description="'ok' when all checks pass, 'error' when at least one fails.",
+    )
+    checks: list[ReadyCheck] = Field(
+        default_factory=list,
+        description="Per-subsystem check results.",
+    )
+    failed_count: int = Field(
+        ...,
+        ge=0,
+        description="Number of checks that returned healthy=False.",
+    )

backend/app/routers/health.py

@@ -1,27 +1,36 @@
 """Health check router.
 
-A lightweight ``GET /api/v1/health`` endpoint that verifies the application
+Two distinct probes following Kubernetes conventions:
-is running and can serve requests.  Also reports the cached fail2ban liveness
-state so monitoring tools and Docker health checks can observe daemon status
-without probing the socket directly.
 
-Comprehensive checks performed:
+* ``GET /api/v1/health/live`` — **Liveness** — checks that the Python process is
-- Database connectivity
+  alive and the event loop is responsive.  Always returns 200; a non-2xx answer
-- fail2ban socket reachability (via cached server_status)
+  tells Kubernetes to *restart* the container.
-- Background scheduler health
+
-- Session cache initialization
+* ``GET /api/v1/health/ready`` — **Readiness** — checks that all critical
+  sub-systems (database, fail2ban socket, config directory, scheduler) are
+  reachable.  Returns 200 only when all pass; returns 503 with a JSON body
+  listing every failed check otherwise.  A non-2xx answer tells Kubernetes to
+  *stop routing traffic* to the pod until it recovers.
+
+The combined ``GET /api/v1/health`` endpoint is retained for backward
+compatibility with existing Docker HEALTHCHECK definitions.
 """
 
 from __future__ import annotations
 
-from typing import Annotated, Literal
+import asyncio
+import os
+from typing import TYPE_CHECKING, Literal
 
 import structlog
 from fastapi import APIRouter, status
 from fastapi.responses import JSONResponse
 
 from app.dependencies import AppStateDep, ServerStatusDep
-from app.models.response import ComponentHealth, HealthResponse
+from app.models.response import ComponentHealth, HealthResponse, ReadyCheck, ReadyResponse
+
+if TYPE_CHECKING:
+    from collections.abc import Coroutine
 
 router: APIRouter = APIRouter(prefix="/api/v1/health", tags=["Health"])
 
@@ -142,3 +151,164 @@ async def health_check(
             components=components,
         ).model_dump(),
     )
+
+
+# --- Constants for subsystem checks ------------------------------------------ #
+
+SUBSYSTEM_TIMEOUT_SECONDS: float = 2.0
+
+
+# --- Helper: run a blocking check in a thread pool to avoid event-loop delays -- #
+
+async def _run_check(
+    name: str,
+    coro: Coroutine[object, object, None],
+    error_msg: str,
+) -> ReadyCheck:
+    """Run *coro* with a short timeout and return a ReadyCheck."""
+    try:
+        await asyncio.wait_for(coro, timeout=SUBSYSTEM_TIMEOUT_SECONDS)
+        return ReadyCheck(name=name, healthy=True)
+    except (OSError, TimeoutError, Exception) as exc:  # noqa: BLE001
+        log.warning("ready_check_failed", subsystem=name, error=str(exc))
+        return ReadyCheck(name=name, healthy=False, message=f"{error_msg}: {exc}")
+
+
+# --- Liveness probe ---------------------------------------------------------- #
+
+
+@router.get(
+    "/live",
+    summary="Process liveness probe",
+    response_model=ReadyResponse,
+    responses={
+        200: {"description": "Process is alive"},
+    },
+)
+async def liveness_probe() -> JSONResponse:
+    """Lightweight liveness check for Kubernetes.
+
+    Returns 200 when the Python process and event loop are responsive.
+    A non-2xx response tells Kubernetes to restart the container.
+    No subsystem checks are performed — this endpoint must be fast.
+    """
+    return JSONResponse(
+        status_code=status.HTTP_200_OK,
+        content=ReadyResponse(
+            status="ok",
+            checks=[ReadyCheck(name="process", healthy=True)],
+            failed_count=0,
+        ).model_dump(),
+    )
+
+
+# --- Readiness probe --------------------------------------------------------- #
+
+
+async def _check_database(app_state: AppStateDep) -> ReadyCheck:
+    """Check database connectivity with a short timeout."""
+    from app.config import Settings
+    from app.db import open_db
+
+    effective_settings: Settings = (
+        app_state.runtime_settings if app_state.runtime_settings is not None else app_state.settings
+    )
+
+    async def _probe() -> None:
+        test_db = await open_db(effective_settings.database_path)
+        await test_db.close()
+
+    return await _run_check(
+        "database",
+        _probe(),
+        "Connection failed",
+    )
+
+
+async def _check_fail2ban(app_state: AppStateDep, server_status: ServerStatusDep) -> ReadyCheck:
+    """Check fail2ban socket reachability using the cached server status."""
+    if server_status.online:
+        return ReadyCheck(name="fail2ban", healthy=True)
+    return ReadyCheck(name="fail2ban", healthy=False, message="Socket not reachable")
+
+
+async def _check_config_dir(app_state: AppStateDep) -> ReadyCheck:
+    """Check config directory read access."""
+    from app.config import Settings
+
+    effective_settings: Settings = (
+        app_state.runtime_settings if app_state.runtime_settings is not None else app_state.settings
+    )
+
+    async def _probe() -> None:
+        config_path = effective_settings.fail2ban_config_dir
+        # Quick read-test: list directory (checks both existence and readability)
+        await asyncio.to_thread(os.access, config_path, os.R_OK)
+
+    return await _run_check(
+        "config_dir",
+        _probe(),
+        "Config directory not readable",
+    )
+
+
+async def _check_scheduler(app_state: AppStateDep) -> ReadyCheck:
+    """Check scheduler liveness."""
+    try:
+        scheduler = app_state.scheduler
+        if scheduler is not None and getattr(scheduler, "running", False):
+            return ReadyCheck(name="scheduler", healthy=True)
+        elif scheduler is not None:
+            return ReadyCheck(name="scheduler", healthy=False, message="Scheduler stopped")
+        else:
+            return ReadyCheck(name="scheduler", healthy=False, message="Not initialised")
+    except AttributeError:
+        return ReadyCheck(name="scheduler", healthy=False, message="Not accessible")
+
+
+@router.get(
+    "/ready",
+    summary="Subsystem readiness probe",
+    response_model=ReadyResponse,
+    responses={
+        200: {"description": "All subsystems healthy"},
+        503: {"description": "One or more subsystems unreachable"},
+    },
+)
+async def readiness_probe(
+    app_state: AppStateDep,
+    server_status: ServerStatusDep,
+) -> JSONResponse:
+    """Readiness check for Kubernetes.
+
+    Verifies all critical sub-systems are reachable:
+    - Database connectivity
+    - fail2ban socket (via cached server status)
+    - Config directory read access
+    - Background scheduler liveness
+
+    Returns HTTP 200 only when every check passes; returns HTTP 503 with a
+    JSON body listing every failed subsystem otherwise.  Each check has a
+    short per-subsystem timeout to prevent the endpoint from overwhelming the
+    system under load.
+    """
+    db_check, f2b_check, config_check, sched_check = await asyncio.gather(
+        _check_database(app_state),
+        _check_fail2ban(app_state, server_status),
+        _check_config_dir(app_state),
+        _check_scheduler(app_state),
+    )
+
+    checks: list[ReadyCheck] = [db_check, f2b_check, config_check, sched_check]
+    failed_count = sum(1 for c in checks if not c.healthy)
+
+    http_status = status.HTTP_200_OK if failed_count == 0 else status.HTTP_503_SERVICE_UNAVAILABLE
+
+    return JSONResponse(
+        status_code=http_status,
+        content=ReadyResponse(
+            status="ok" if failed_count == 0 else "error",
+            checks=checks,
+            failed_count=failed_count,
+        ).model_dump(),
+    )

backend/tests/test_main.py

@@ -12,7 +12,15 @@ from httpx import ASGITransport, AsyncClient
 from app.config import Settings
 from app.db import init_db
 from app.exceptions import ConfigValidationError, ConfigWriteError, JailNotFoundError
-from app.main import CORSMiddleware, _enforce_single_worker, _lifespan, create_app
+from app.main import (
+    CORSMiddleware,
+    _assert_middleware_order,
+    _enforce_single_worker,
+    _lifespan,
+    create_app,
+)
+from app.middleware.correlation import CorrelationIdMiddleware
+from app.middleware.rate_limit import RateLimitMiddleware
 from app.services import setup_service
 
 
@@ -450,14 +458,23 @@ async def test_startup_loads_geo_cache_from_persisted_runtime_database(tmp_path:
     exit_stack.enter_context(patch("app.services.geo_cache.GeoCache.load_cache_from_db", new=load_cache))
     exit_stack.enter_context(patch("app.services.geo_cache.GeoCache.count_unresolved", new=AsyncMock(return_value=0)))
     exit_stack.enter_context(patch("app.services.setup_service.is_setup_complete", new=AsyncMock(return_value=True)))
-    exit_stack.enter_context(patch("app.services.setup_service.get_runtime_database_path", new=AsyncMock(return_value=runtime_db_path)))
+    exit_stack.enter_context(patch(
-    exit_stack.enter_context(patch("app.services.setup_service.get_persisted_runtime_settings", new=AsyncMock(return_value={
+        "app.services.setup_service.get_runtime_database_path",
-        "database_path": runtime_db_path,
+        new=AsyncMock(return_value=runtime_db_path),
-        "fail2ban_socket": "/tmp/persisted.sock",
+    ))
-        "timezone": "Europe/Berlin",
+    exit_stack.enter_context(patch(
-        "session_duration_minutes": 123,
+        "app.services.setup_service.get_persisted_runtime_settings",
-    })))
+        new=AsyncMock(return_value={
-    exit_stack.enter_context(patch("app.services.setup_service.get_fail2ban_db_path", new=AsyncMock(return_value="/tmp/fail2ban/banned.tar.bz2")))
+            "database_path": runtime_db_path,
+            "fail2ban_socket": "/tmp/persisted.sock",
+            "timezone": "Europe/Berlin",
+            "session_duration_minutes": 123,
+        }),
+    ))
+    exit_stack.enter_context(patch(
+        "app.services.setup_service.get_fail2ban_db_path",
+        new=AsyncMock(return_value="/tmp/fail2ban/banned.tar.bz2"),
+    ))
     exit_stack.enter_context(patch("app.tasks.health_check.register"))
     exit_stack.enter_context(patch("app.tasks.blocklist_import.register"))
     exit_stack.enter_context(patch("app.tasks.geo_cache_flush.register"))
@@ -466,8 +483,9 @@ async def test_startup_loads_geo_cache_from_persisted_runtime_database(tmp_path:
 
     with exit_stack:
         async with _lifespan(app):
-            loaded_db_path = load_cache.call_args.args[0]
+            runtime_connections = [
-            runtime_connections = [conn for path, conn in opened_connections if path == runtime_db_path]
+                conn for path, conn in opened_connections if path == runtime_db_path
+            ]
             assert runtime_connections, "Expected runtime database to be opened"
 
             assert app.state.runtime_settings is not None
@@ -538,6 +556,91 @@ async def test_concurrent_requests_use_request_scoped_db_connections(tmp_path: P
     assert all(connection.close.await_count == 1 for connection in connections)
 
 
+# ---------------------------------------------------------------------------
+# Middleware order validation
+# ---------------------------------------------------------------------------
+
+
+def _make_settings(tmp_path: Path) -> Settings:
+    """Return a minimal Settings object with a temporary fail2ban config dir."""
+    fail2ban_config_dir = tmp_path / "fail2ban"
+    fail2ban_config_dir.mkdir()
+    return Settings(
+        database_path=str(tmp_path / "bangui.db"),
+        fail2ban_socket="/tmp/fake_fail2ban.sock",
+        fail2ban_config_dir=str(fail2ban_config_dir),
+        session_secret="test-secret-key-do-not-use-in-production",
+        session_duration_minutes=60,
+        timezone="UTC",
+        log_level="debug",
+    )
+
+
+def test_create_app_raises_on_incorrect_middleware_order(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """_assert_middleware_order() raises AssertionError when middleware order is wrong.
+
+    The security-critical chain requires:
+    RateLimitMiddleware → CsrfMiddleware → CorrelationIdMiddleware
+    in user_middleware (processing order: outermost → innermost).
+    """
+    monkeypatch.setenv("TESTING", "1")
+    settings = _make_settings(tmp_path)
+    app = create_app(settings=settings)
+    # Swap CorrelationIdMiddleware and RateLimitMiddleware to break the order.
+    user_mw = app.user_middleware
+    corr_idx = next(i for i, m in enumerate(user_mw) if m.cls.__name__ == "CorrelationIdMiddleware")
+    rate_idx = next(i for i, m in enumerate(user_mw) if m.cls.__name__ == "RateLimitMiddleware")
+    user_mw[corr_idx], user_mw[rate_idx] = user_mw[rate_idx], user_mw[corr_idx]
+    with pytest.raises(AssertionError, match="must be registered before"):
+        _assert_middleware_order(app)
+
+
+def test_middleware_order_validation_passes_for_correct_order(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """_assert_middleware_order() does not raise when middleware order is correct."""
+    monkeypatch.setenv("TESTING", "1")
+    settings = _make_settings(tmp_path)
+    app = create_app(settings=settings)
+    _assert_middleware_order(app)  # Should not raise
+
+
+def test_create_app_validates_middleware_order_at_startup(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """create_app() raises immediately if middleware registration order is incorrect.
+
+    This test verifies the integration: _assert_middleware_order is called at the
+    end of create_app, so a fresh app with deliberately wrong middleware order
+    (simulated by patching add_middleware during creation) raises AssertionError.
+    """
+    monkeypatch.setenv("TESTING", "1")
+    settings = _make_settings(tmp_path)
+
+    from starlette.applications import Starlette
+
+    original_add = Starlette.add_middleware
+
+    def swapping_add(self, middleware_cls: type, **kwargs: object) -> None:
+        """Patched add_middleware that swaps CorrelationId and RateLimit."""
+        if middleware_cls is CorrelationIdMiddleware:
+            pass  # Skip CorrelationId
+        elif middleware_cls is RateLimitMiddleware:
+            original_add(self, RateLimitMiddleware, **kwargs)
+            original_add(self, CorrelationIdMiddleware)
+        else:
+            original_add(self, middleware_cls, **kwargs)
+
+    with patch.object(Starlette, "add_middleware", swapping_add), \
+         pytest.raises(AssertionError, match="must be registered before"):
+        create_app(settings=settings)
+
+
 # ---------------------------------------------------------------------------
 # Single-worker enforcement
 # ---------------------------------------------------------------------------

backend/tests/test_routers/test_health_probes.py

@@ -0,0 +1,130 @@
+"""Tests for the health-check router — liveness and readiness probes."""
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+from httpx import AsyncClient
+
+from app.models.server import ServerStatus
+from app.models.response import ReadyCheck
+
+
+# ---------------------------------------------------------------------------
+# GET /health/live — liveness probe
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_liveness_returns_200(client: AsyncClient) -> None:
+    """``GET /api/v1/health/live`` must always return HTTP 200."""
+    response = await client.get("/api/v1/health/live")
+    assert response.status_code == 200
+
+
+@pytest.mark.asyncio
+async def test_liveness_body_is_ready_response(client: AsyncClient) -> None:
+    """Response body must be a ReadyResponse."""
+    response = await client.get("/api/v1/health/live")
+    data: dict[str, object] = response.json()
+    assert data["status"] == "ok"
+    assert data["failed_count"] == 0
+    assert "checks" in data
+    assert isinstance(data["checks"], list)
+
+
+@pytest.mark.asyncio
+async def test_liveness_includes_process_check(client: AsyncClient) -> None:
+    """Liveness response must include a 'process' check."""
+    response = await client.get("/api/v1/health/live")
+    data: dict[str, object] = response.json()
+    checks: list[dict[str, object]] = data["checks"]  # type: ignore[assignment]
+    assert any(c.get("name") == "process" and c.get("healthy") is True for c in checks)
+
+
+# ---------------------------------------------------------------------------
+# GET /health/ready — readiness probe
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_readiness_returns_200_when_all_pass(client: AsyncClient) -> None:
+    """``GET /api/v1/health/ready`` must return 200 when all subsystems pass."""
+    with patch("app.routers.health._run_check", side_effect=lambda n, c, e: ReadyCheck(name=n, healthy=True)):
+        response = await client.get("/api/v1/health/ready")
+    assert response.status_code == 200
+
+
+@pytest.mark.asyncio
+async def test_readiness_returns_503_when_subsystem_fails(client: AsyncClient) -> None:
+    """``GET /api/v1/health/ready`` must return 503 when at least one check fails."""
+    # Force fail2ban offline
+    client._transport.app.state.server_status = ServerStatus(online=False)
+    response = await client.get("/api/v1/health/ready")
+    assert response.status_code == 503
+
+
+@pytest.mark.asyncio
+async def test_readiness_body_is_ready_response(client: AsyncClient) -> None:
+    """Response body must be a ReadyResponse."""
+    response = await client.get("/api/v1/health/ready")
+    data: dict[str, object] = response.json()
+    assert data["status"] in ("ok", "error")
+    assert "failed_count" in data
+    assert "checks" in data
+    assert isinstance(data["checks"], list)
+
+
+@pytest.mark.asyncio
+async def test_readiness_includes_all_subsystems(client: AsyncClient) -> None:
+    """Readiness response must include checks for all four subsystems."""
+    response = await client.get("/api/v1/health/ready")
+    data: dict[str, object] = response.json()
+    checks: list[dict[str, object]] = data["checks"]  # type: ignore[assignment]
+    names = {c["name"] for c in checks}
+    assert names == {"database", "fail2ban", "config_dir", "scheduler"}
+
+
+@pytest.mark.asyncio
+async def test_readiness_status_ok_when_all_healthy(client: AsyncClient) -> None:
+    """``status`` must be 'ok' when all checks pass."""
+    with patch("app.routers.health._run_check", side_effect=lambda n, c, e: ReadyCheck(name=n, healthy=True)):
+        response = await client.get("/api/v1/health/ready")
+    data: dict[str, object] = response.json()
+    assert data["status"] == "ok"
+    assert data["failed_count"] == 0
+
+
+@pytest.mark.asyncio
+async def test_readiness_status_error_when_fail2ban_offline(client: AsyncClient) -> None:
+    """``status`` must be 'error' when fail2ban is offline."""
+    client._transport.app.state.server_status = ServerStatus(online=False)
+    response = await client.get("/api/v1/health/ready")
+    data: dict[str, object] = response.json()
+    assert data["status"] == "error"
+    assert data["failed_count"] > 0
+
+
+@pytest.mark.asyncio
+async def test_readiness_includes_failed_subsystem_detail(client: AsyncClient) -> None:
+    """When fail2ban is offline the fail2ban check must include an error message."""
+    client._transport.app.state.server_status = ServerStatus(online=False)
+    response = await client.get("/api/v1/health/ready")
+    data: dict[str, object] = response.json()
+    checks: list[dict[str, object]] = data["checks"]  # type: ignore[assignment]
+    f2b = next(c for c in checks if c["name"] == "fail2ban")
+    assert f2b["healthy"] is False
+    assert f2b["message"] is not None
+
+
+@pytest.mark.asyncio
+async def test_readiness_content_type_is_json(client: AsyncClient) -> None:
+    """``/api/v1/health/ready`` must set the ``Content-Type`` header to JSON."""
+    response = await client.get("/api/v1/health/ready")
+    assert "application/json" in response.headers.get("content-type", "")
+
+
+@pytest.mark.asyncio
+async def test_readiness_live_content_type_is_json(client: AsyncClient) -> None:
+    """``/api/v1/health/live`` must set the ``Content-Type`` header to JSON."""
+    response = await client.get("/api/v1/health/live")
+    assert "application/json" in response.headers.get("content-type", "")

frontend/src/hooks/README.md

@@ -53,3 +53,28 @@ With drift correction:
 - Total time from poll start to next poll start is always ~5 seconds
 - Fetch duration doesn't affect the long-term polling rate
 - Bandwidth and CPU usage remain consistent
+
+## Request Deduplication with Abort Signal Handling
+
+When `useFetchData` is called with a `requestKey`, multiple hook instances sharing the same key coalesce around a single in-flight request. Each instance is tracked as a **subscriber**.
+
+### How it works
+
+- A module-level `Map<requestKey, InFlightRequest>` holds in-flight requests.
+- Each `InFlightRequest` contains a `subscribers` map: `subscriberId → { signal, onSetData, onSuccess }`.
+- When a new hook instance joins a deduplicated request, it registers a subscriber entry and attaches a `once` abort listener that removes it from the subscriber list.
+- When the underlying request resolves, **each subscriber's signal is checked before calling `setData` or `onSuccess`**. Aborted subscribers are skipped, preventing state updates on unmounted components.
+- When all subscribers have aborted, the underlying `AbortController` is signalled, cancelling the in-flight fetch.
+
+### Guarantees
+
+| Scenario | Result |
+|---|---|
+| Subscriber aborts before request resolves | Subscriber receives no `setData`/`onSuccess`. Entry removed from subscriber list. |
+| Last subscriber aborts before request resolves | Underlying request is cancelled via `AbortController`. |
+| Request resolves, subscriber already aborted | `signal.aborted` checked before `setData`/`onSuccess` — no-op. |
+| Request errors | Non-abort errors passed to `handleFetchError`. Each subscriber's error state updated independently. |
+
+### When to use deduplication
+
+Use `requestKey` when multiple components may request the same resource simultaneously. The shared promise avoids redundant network calls and prevents race conditions from out-of-order responses.

frontend/src/hooks/__tests__/useFetchData.test.ts

@@ -1,8 +1,13 @@
-import { describe, it, expect, vi } from "vitest";
+import { describe, it, expect, vi, beforeEach } from "vitest";
 import { renderHook, act } from "@testing-library/react";
-import { useFetchData } from "../useFetchData";
+import { useFetchData, _resetInFlightRequests } from "../useFetchData";
 import type { FetchError } from "../../types/api";
 
+// Clear the module-level inFlightRequests map between tests to prevent state leakage
+beforeEach(() => {
+  _resetInFlightRequests();
+});
+
 describe("useFetchData", () => {
   it("fetches and selects data on mount", async () => {
     const fetcher = vi.fn().mockResolvedValue({ value: "test" });
@@ -406,4 +411,123 @@ describe("useFetchData", () => {
     expect(fetcher).toHaveBeenCalledTimes(2);
     expect(result.current.data).toBe("second");
   });
+
+  it("aborted subscriber in deduplication does not receive setData/onSuccess", async () => {
+    let resolveFirst: ((value: { value: string }) => void) | null = null;
+    const fetcher = vi.fn().mockImplementation(
+      () =>
+        new Promise((resolve) => {
+          resolveFirst = resolve;
+        })
+    );
+    const selector = vi.fn((response: { value: string }) => response.value);
+    const onSuccess1 = vi.fn();
+    const onSuccess2 = vi.fn();
+
+    const hook1 = renderHook(() =>
+      useFetchData({
+        fetcher,
+        selector,
+        errorMessage: "Failed to load",
+        requestKey: "abort-subscriber-test",
+        onSuccess: onSuccess1,
+      })
+    );
+
+    const hook2 = renderHook(() =>
+      useFetchData({
+        fetcher,
+        selector,
+        errorMessage: "Failed to load",
+        requestKey: "abort-subscriber-test",
+        onSuccess: onSuccess2,
+      })
+    );
+
+    await act(async () => {
+      await Promise.resolve();
+    });
+
+    expect(fetcher).toHaveBeenCalledTimes(1);
+
+    // Unmount hook2 (simulates component unmounting and aborting its signal)
+    hook2.unmount();
+
+    // Allow the request to resolve
+    await act(async () => {
+      resolveFirst?.({ value: "shared-data" });
+      await Promise.resolve();
+    });
+
+    // hook2 should not have received the data (it aborted)
+    expect(hook2.result.current.data).toBeUndefined();
+    expect(onSuccess2).not.toHaveBeenCalled();
+    // hook1 should still have the data
+    expect(hook1.result.current.data).toBe("shared-data");
+    expect(onSuccess1).toHaveBeenCalledWith({ value: "shared-data" });
+  });
+
+  it("last subscriber abort cancels underlying request", async () => {
+    let resolveFirst: ((value: { value: string }) => void) | null = null;
+    const abortSignals: AbortSignal[] = [];
+    const fetcher = vi.fn().mockImplementation((signal: AbortSignal) => {
+      abortSignals.push(signal);
+      return new Promise((resolve) => {
+        resolveFirst = resolve;
+      });
+    });
+    const selector = vi.fn((response: { value: string }) => response.value);
+
+    const hook1 = renderHook(() =>
+      useFetchData({
+        fetcher,
+        selector,
+        errorMessage: "Failed to load",
+        requestKey: "cancel-underlying-test",
+      })
+    );
+
+    await act(async () => {
+      await Promise.resolve();
+    });
+
+    expect(fetcher).toHaveBeenCalledTimes(1);
+
+    // Mount second subscriber
+    const hook2 = renderHook(() =>
+      useFetchData({
+        fetcher,
+        selector,
+        errorMessage: "Failed to load",
+        requestKey: "cancel-underlying-test",
+      })
+    );
+
+    await act(async () => {
+      await Promise.resolve();
+    });
+
+    // Both subscribers now sharing one request
+    expect(fetcher).toHaveBeenCalledTimes(1);
+
+    // Unmount first subscriber (doesn't cancel underlying request yet)
+    hook1.unmount();
+
+    await act(async () => {
+      await Promise.resolve();
+    });
+
+    // Underlying request should NOT be cancelled yet (hook2 still waiting)
+    expect(abortSignals[0]?.aborted).toBe(false);
+
+    // Unmount second (last) subscriber — should cancel the underlying request
+    hook2.unmount();
+
+    await act(async () => {
+      await Promise.resolve();
+    });
+
+    // Underlying request should now be cancelled
+    expect(abortSignals[0]?.aborted).toBe(true);
+  });
 });

frontend/src/hooks/useFetchData.ts

@@ -19,16 +19,33 @@ import type { FetchError } from "../types/api";
 
 /**
  * Module-level cache for in-flight requests.
- * Maps requestKey to { promise, controller } to enable deduplication
+ * Maps requestKey to { promise, controller, subscribers } to enable deduplication
  * across multiple hook instances.
  */
+interface Subscriber<TResponse> {
+  /** This instance's abort signal. */
+  signal: AbortSignal;
+  /** Callbacks registered by this subscriber. */
+  onSetData?: (response: TResponse) => void;
+  onSuccess?: (response: TResponse) => void;
+}
+
 interface InFlightRequest<TResponse> {
   promise: Promise<TResponse>;
   controller: AbortController;
+  /** Map of subscriberId -> Subscriber.  Cleared when subscriber aborts. */
+  subscribers: Map<string, Subscriber<TResponse>>;
+  /** True when initiator has cleaned up but subscribers remain. */
+  initiatorDone: boolean;
 }
 
 const inFlightRequests = new Map<string, InFlightRequest<unknown>>();
 
+/** Visible for testing only.  Clears all in-flight request state. */
+export const _resetInFlightRequests = (): void => {
+  inFlightRequests.clear();
+};
+
 export interface UseFetchDataOptions<TResponse, TData> {
   /** Async function that accepts an AbortSignal for cancellation. */
   fetcher: (signal: AbortSignal) => Promise<TResponse>;
@@ -86,18 +103,57 @@ export function useFetchData<TResponse, TData>(
   const [error, setError] = useState<FetchError | null>(null);
   const abortRef = useRef<AbortController | null>(null);
   const localControllerRef = useRef<AbortController | null>(null);
+  const subscriberAbortRef = useRef<AbortController | null>(null);
+  /** Unique ID for this instance, used to track its subscription to deduplicated requests. */
+  const subscriberIdRef = useRef<string | null>(null);
 
   const refresh = useCallback((): void => {
+    // Abort any previous request from this hook instance
+    abortRef.current?.abort();
+    localControllerRef.current = new AbortController();
+    abortRef.current = localControllerRef.current;
+
     // If using request deduplication via requestKey and a request is already in-flight,
-    // wait for it to complete instead of launching a duplicate
+    // subscribe to it instead of launching a duplicate
     if (requestKey && inFlightRequests.has(requestKey)) {
       const inFlight = inFlightRequests.get(requestKey)! as InFlightRequest<TResponse>;
-      inFlight.promise
+
-        .then((response: TResponse) => {
+      // Create per-instance abort controller for this subscription.
+      // Do NOT abort previous subscription signals here - each subscriber's signal
+      // only aborts when THAT subscriber unmounts. Aborting old signals would
+      // incorrectly remove other active subscribers from the list.
+      subscriberAbortRef.current = new AbortController();
+
+      // Register this instance as a subscriber.
+      // Always generate a new ID - each hook instance must have its own subscriber ID
+      // so that unmounting one subscriber does not remove another subscriber's entry.
+      subscriberIdRef.current = crypto.randomUUID();
+      const sid = subscriberIdRef.current;
+      const subscription: Subscriber<TResponse> = {
+        signal: subscriberAbortRef.current.signal,
+        onSetData: (response: TResponse) => {
           setData(selector(response));
           if (onSuccess) {
             onSuccess(response);
           }
+        },
+        onSuccess,
+      };
+      inFlight.subscribers.set(sid, subscription);
+
+      // When this instance aborts, remove it from the subscriber list.
+      // Note: we do NOT abort the underlying controller here - that belongs to
+      // the initiator and should only be aborted by the initiator itself.
+      const cleanup = (): void => {
+        inFlight.subscribers.delete(sid);
+      };
+      subscription.signal.addEventListener("abort", cleanup, { once: true });
+
+      inFlight.promise
+        .then((response: TResponse) => {
+          // Check whether this subscriber has already aborted before propagating result
+          if (subscription.signal.aborted) return;
+          subscription.onSetData?.(response);
         })
         .catch((err: unknown) => {
           // Only handle non-abort errors; abort errors are silently ignored
@@ -139,7 +195,6 @@ export function useFetchData<TResponse, TData>(
         if (!controller.signal.aborted) {
           setLoading(false);
         }
-        // Clear cache entry when response arrives
         if (requestKey) {
           inFlightRequests.delete(requestKey);
         }
@@ -150,6 +205,8 @@ export function useFetchData<TResponse, TData>(
       inFlightRequests.set(requestKey, {
         promise: responsePromise,
         controller,
+        subscribers: new Map(),
+        initiatorDone: false,
       });
     }
   }, [fetcher, selector, errorMessage, onSuccess, requestKey]);
@@ -158,7 +215,25 @@ export function useFetchData<TResponse, TData>(
     refresh();
 
     return (): void => {
-      abortRef.current?.abort();
+      if (subscriberIdRef.current !== null) {
+        // Subscriber cleanup
+        const inFlight = requestKey ? inFlightRequests.get(requestKey) : undefined;
+        subscriberAbortRef.current?.abort();
+        // If initiator already done and no subscribers left, cancel the request
+        if (inFlight && inFlight.initiatorDone && inFlight.subscribers.size === 0) {
+          inFlight.controller.abort();
+        }
+      } else {
+        // Initiator cleanup
+        const inFlight = requestKey ? inFlightRequests.get(requestKey) : undefined;
+        // Mark initiator as done so last subscriber knows to cancel
+        if (inFlight) {
+          inFlight.initiatorDone = true;
+        }
+        if (!inFlight || inFlight.subscribers.size === 0) {
+          abortRef.current?.abort();
+        }
+      }
     };
   }, [refresh]);