feat: Redis caching + async queue for LLM scaling (v1.6.0)

- Add async Redis client singleton (redis_client.py) for caching and arq pool
- Add arq job functions (jobs.py) for background LLM processing
- Cache ask/explain LLM responses with TTL (1h ask, 24h explain)
- Add async mode to /api/ask: enqueue job, return job_id, poll /api/jobs/{id}
- Add GET /api/jobs/{job_id} endpoint for job status polling
- Add arq worker service to docker-compose (dev + prod)
- Switch from Redis to Valkey (BSD fork) in Docker Compose
- Add REDIS_URL config setting
- Add tests for cache hit, async mode, and job status

backend/tests/test_ask.py

@@ -350,3 +350,124 @@ class TestAskEndpoint:
         data = response.json()
         assert data["query_info"]["event_count"] == 1
         assert data["events"][0]["id"] == "evt-bob"
+
+
+class TestAskCaching:
+    def test_ask_cache_hit_returns_cached_answer(self, client, mock_events_collection, monkeypatch):
+        """If the answer is cached, the LLM should not be called."""
+        now = datetime.now(UTC)
+        mock_events_collection.insert_one(
+            {
+                "id": "evt-cache",
+                "timestamp": now.isoformat(),
+                "service": "Directory",
+                "operation": "Add user",
+                "result": "success",
+                "actor_display": "Alice",
+                "target_displays": ["USER-001"],
+                "display_summary": "summary",
+                "raw_text": "raw",
+            }
+        )
+
+        llm_called = False
+
+        async def fake_llm(question, events, total=None, excluded_services=None):
+            nonlocal llm_called
+            llm_called = True
+            return "This should NOT appear."
+
+        monkeypatch.setattr("routes.ask.LLM_API_KEY", "fake-key")
+        monkeypatch.setattr("routes.ask._call_llm", fake_llm)
+
+        # Pre-populate cache with a specific answer
+        class CachingFakeRedis:
+            def __init__(self):
+                self.store = {}
+
+            async def get(self, key):
+                return self.store.get(key)
+
+            async def setex(self, key, ttl, value):
+                self.store[key] = value
+
+        redis = CachingFakeRedis()
+        # Seed cache with the exact filters the endpoint will generate
+        import asyncio
+        from jobs import set_cached_ask
+        filters_snapshot = {
+            "services": None,
+            "actor": None,
+            "operation": None,
+            "result": None,
+            "start": None,
+            "end": None,
+            "include_tags": None,
+            "exclude_tags": None,
+        }
+        asyncio.run(set_cached_ask(redis, "What happened to USER-001?", filters_snapshot, [{"id": "evt-cache"}], {"answer": "Cached answer!", "llm_used": True, "llm_error": None}))
+
+        async def fake_get_arq_pool():
+            return redis
+
+        monkeypatch.setattr("routes.ask.get_arq_pool", fake_get_arq_pool)
+
+        response = client.post("/api/ask", json={"question": "What happened to USER-001?"})
+        assert response.status_code == 200
+        data = response.json()
+        assert data["answer"] == "Cached answer!"
+        assert data["llm_used"] is True
+        assert llm_called is False
+
+    def test_ask_async_mode_returns_job_id(self, client, mock_events_collection, monkeypatch):
+        """Async mode should return immediately with a job_id."""
+        now = datetime.now(UTC)
+        mock_events_collection.insert_one(
+            {
+                "id": "evt-async",
+                "timestamp": now.isoformat(),
+                "service": "Directory",
+                "operation": "Add user",
+                "result": "success",
+                "actor_display": "Alice",
+                "target_displays": ["USER-001"],
+                "display_summary": "summary",
+                "raw_text": "raw",
+            }
+        )
+
+        monkeypatch.setattr("routes.ask.LLM_API_KEY", "fake-key")
+
+        # Mock arq pool to capture enqueue_job call
+        class FakeArqPool:
+            def __init__(self):
+                self.enqueued = []
+
+            async def get(self, key):
+                return None
+
+            async def setex(self, key, ttl, value):
+                pass
+
+            async def enqueue_job(self, func, *args, **kwargs):
+                from unittest.mock import MagicMock
+                job = MagicMock()
+                job.job_id = "job-12345"
+                self.enqueued.append((func, args, kwargs))
+                return job
+
+        pool = FakeArqPool()
+
+        async def fake_get_arq_pool():
+            return pool
+
+        monkeypatch.setattr("routes.ask.get_arq_pool", fake_get_arq_pool)
+
+        response = client.post("/api/ask", json={"question": "What happened to USER-001?", "async_mode": True})
+        assert response.status_code == 200
+        data = response.json()
+        assert data["job_id"] == "job-12345"
+        assert data["llm_used"] is False
+        assert "being processed" in data["answer"]
+        assert len(pool.enqueued) == 1
+        assert pool.enqueued[0][0] == "process_ask_question"