feat(cloud-api): expose health readiness

apps/cloud-api/cloud_api/app.py

@@ -7,7 +7,8 @@ from contextlib import asynccontextmanager
 from dataclasses import dataclass
 from typing import Any
 
-from fastapi import FastAPI
+from fastapi import FastAPI, status
+from fastapi.responses import JSONResponse
 
 from cloud.auth import create_auth_provider
 from cloud.config import CloudConfig
@@ -90,12 +91,15 @@ def create_app(
         app.state.cloud_config = control_config
         app.state.database = database
         app.state.cloud_services = services
+        app.state.startup_complete = False
+        app.state.schema_ready = False
         stop_workers = asyncio.Event()
         worker_tasks: list[asyncio.Task[None]] = []
         try:
             repository.health_check()
             if control_config.environment == "production":
                 require_current_schema(control_config.database_url)
+            app.state.schema_ready = True
             worker_tasks = [
                 asyncio.create_task(
                     _run_scheduler_loop(
@@ -116,8 +120,10 @@ def create_app(
                 ),
             ]
             app.state.worker_tasks = tuple(worker_tasks)
+            app.state.startup_complete = True
             yield
         finally:
+            app.state.startup_complete = False
             stop_workers.set()
             try:
                 if worker_tasks:
@@ -129,6 +135,37 @@ def create_app(
                     repository.unbind()
 
     app = FastAPI(title="Device Cloud API", lifespan=lifespan)
+
+    @app.get("/health/live")
+    def health_live() -> dict[str, str]:
+        return {"status": "live"}
+
+    @app.get("/health/ready")
+    def health_ready():
+        checks = {
+            "configuration": bool(getattr(app.state, "startup_complete", False)),
+            "database": False,
+            "schema": bool(getattr(app.state, "schema_ready", False)),
+            "workers": False,
+        }
+        if checks["configuration"]:
+            try:
+                services.repository.health_check()
+                checks["database"] = True
+            except Exception:
+                checks["database"] = False
+        worker_state = getattr(app.state, "worker_tasks", ())
+        checks["workers"] = bool(worker_state) and all(
+            not task.done() for task in worker_state
+        )
+        ready = all(checks.values())
+        return JSONResponse(
+            status_code=(
+                status.HTTP_200_OK if ready else status.HTTP_503_SERVICE_UNAVAILABLE
+            ),
+            content={"status": "ready" if ready else "not_ready", "checks": checks},
+        )
+
     app.include_router(
         create_cloud_router(
             pool=pool,