feat(matrix-bridge-dagi): add backpressure queue with N workers (H2)

Reader + N workers architecture:
  Reader: sync_poll → rate_check → dedupe → queue.put_nowait()
  Workers (WORKER_CONCURRENCY, default 2): queue.get() → invoke → send → audit

Drop policy (queue full):
  - put_nowait() raises QueueFull → dropped immediately (reader never blocks)
  - audit matrix.queue_full + on_queue_dropped callback
  - metric: matrix_bridge_queue_dropped_total{room_id,agent_id}

Graceful shutdown:
  1. stop_event → reader exits loop
  2. queue.join() with QUEUE_DRAIN_TIMEOUT_S (default 5s) → workers finish in-flight
  3. worker tasks cancelled

New config env vars:
  QUEUE_MAX_EVENTS (default 100)
  WORKER_CONCURRENCY (default 2)
  QUEUE_DRAIN_TIMEOUT_S (default 5)

New metrics (H3 additions):
  matrix_bridge_queue_size (gauge)
  matrix_bridge_queue_dropped_total (counter)
  matrix_bridge_queue_wait_seconds histogram (buckets: 0.01…30s)

/health: queue.size, queue.max, queue.workers
MatrixIngressLoop: queue_size + worker_count properties

6 queue tests: enqueue/process, full-drop-audit, concurrency barrier,
graceful drain, wait metric, rate-limit-before-enqueue
Total: 71 passed

Made-with: Cursor

services/matrix-bridge-dagi/app/config.py

@@ -29,6 +29,11 @@ class BridgeConfig:
     rate_limit_room_rpm: int       # max messages per room per minute
     rate_limit_sender_rpm: int     # max messages per sender per minute
 
+    # H2: Backpressure queue
+    queue_max_events: int          # max pending items (drops oldest on full)
+    worker_concurrency: int        # parallel invoke workers
+    queue_drain_timeout_s: float   # graceful shutdown drain timeout
+
     # Service identity
     node_id: str
     build_sha: str
@@ -62,6 +67,9 @@ def load_config() -> BridgeConfig:
         bridge_allowed_agents=allowed,
         rate_limit_room_rpm=int(_optional("RATE_LIMIT_ROOM_RPM", "20")),
         rate_limit_sender_rpm=int(_optional("RATE_LIMIT_SENDER_RPM", "10")),
+        queue_max_events=max(1, int(_optional("QUEUE_MAX_EVENTS", "100"))),
+        worker_concurrency=max(1, int(_optional("WORKER_CONCURRENCY", "2"))),
+        queue_drain_timeout_s=max(1.0, float(_optional("QUEUE_DRAIN_TIMEOUT_S", "5"))),
         node_id=_optional("NODE_ID", "NODA1"),
         build_sha=_optional("BUILD_SHA", "dev"),
         build_time=_optional("BUILD_TIME", "local"),

services/matrix-bridge-dagi/app/ingress.py

@@ -1,32 +1,41 @@
 """
-Matrix Ingress + Egress Loop — Phase M1.4 + H1/H3
+Matrix Ingress + Egress Loop — Phase M1.4 + H1 + H2 + H3
 
-Polls Matrix /sync for new messages, invokes DAGI Router for mapped rooms,
-sends agent replies back to Matrix, writes audit events to sofiia-console.
+Architecture (H2):
+  Reader task  → asyncio.Queue(maxsize) → N Worker tasks
+  ─────────────────────────────────────────────────────────
+  Reader:
+    sync_poll() → extract_room_messages()
+    → rate_limit check (H1)
+    → mark_seen / dedupe
+    → queue.put_nowait() or DROP (audit matrix.queue_full + metric)
 
-Pipeline:
-  sync_poll() → extract_room_messages()
-  → for each message:
-    1. rate_limit check (room + sender)  ← H1
-    2. dedupe (mark_seen)
-    3. audit: matrix.message.received
-    4. invoke DAGI Router  (timed → on_invoke_latency)  ← H3
-    5. send_text() reply   (timed → on_send_latency)    ← H3
-    6. audit: matrix.agent.replied | matrix.error
+  Workers (N concurrent):
+    queue.get() → measure wait latency (H3)
+    → audit matrix.message.received
+    → invoke Router  (timed, H3)
+    → send_text()    (timed, H3)
+    → audit matrix.agent.replied | matrix.error
 
-Graceful shutdown via asyncio.Event.
+Shutdown:
+  1. stop_event set → reader exits loop
+  2. queue.join() with drain_timeout → workers finish in-flight
+  3. worker tasks cancelled
+
+Queue entry: _QueueEntry(event, room_id, agent_id, enqueue_time)
 """
 
 import asyncio
 import logging
 import time
+from dataclasses import dataclass
 from typing import Any, Callable, Dict, List, Optional
 
 import httpx
 
 from .matrix_client import MatrixClient
 from .rate_limit import InMemoryRateLimiter
-from .room_mapping import RoomMappingConfig
+from .room_mapping import RoomMappingConfig, RoomMapping
 
 logger = logging.getLogger(__name__)
 
@@ -37,6 +46,17 @@ _INIT_RETRY_BACKOFF = 2.0
 _ROUTER_TIMEOUT_S = 45.0
 _AUDIT_TIMEOUT_S = 5.0
 _REPLY_TEXT_MAX = 4000
+_WORKER_GET_TIMEOUT_S = 1.0   # how long a worker waits on empty queue before re-checking
+
+
+# ── Queue entry ────────────────────────────────────────────────────────────────
+
+@dataclass
+class _QueueEntry:
+    event: Dict[str, Any]
+    room_id: str
+    agent_id: str
+    enqueue_time: float   # time.monotonic() at enqueue
 
 
 # ── Router invoke ──────────────────────────────────────────────────────────────
@@ -49,20 +69,13 @@ async def _invoke_router(
     prompt: str,
     session_id: str,
 ) -> str:
-    """
-    POST /v1/agents/{agent_id}/infer — returns response text string.
-    Field confirmed as 'response' on NODA1.
-    Raises httpx.HTTPError on failure.
-    """
+    """POST /v1/agents/{agent_id}/infer → response text. Raises httpx.HTTPError on failure."""
     url = f"{router_url.rstrip('/')}/v1/agents/{agent_id}/infer"
     payload = {
         "prompt": prompt,
         "session_id": session_id,
         "user_id": "matrix_bridge",
-        "metadata": {
-            "transport": "matrix",
-            "node_id": node_id,
-        },
+        "metadata": {"transport": "matrix", "node_id": node_id},
     }
     resp = await http_client.post(url, json=payload, timeout=_ROUTER_TIMEOUT_S)
     resp.raise_for_status()
@@ -74,9 +87,7 @@ async def _invoke_router(
         or data.get("message")
         or ""
     )
-    if not isinstance(text, str):
-        text = str(text)
-    return text.strip()
+    return (text if isinstance(text, str) else str(text)).strip()
 
 
 # ── Audit write ────────────────────────────────────────────────────────────────
@@ -95,13 +106,12 @@ async def _write_audit(
     duration_ms: Optional[int] = None,
     data: Optional[Dict[str, Any]] = None,
 ) -> None:
-    """Fire-and-forget audit write. Never raises."""
+    """Fire-and-forget. Never raises."""
     if not console_url or not internal_token:
         return
     try:
-        url = f"{console_url.rstrip('/')}/api/audit/internal"
         await http_client.post(
-            url,
+            f"{console_url.rstrip('/')}/api/audit/internal",
             json={
                 "event": event,
                 "operator_id": "matrix_bridge",
@@ -111,11 +121,7 @@ async def _write_audit(
                 "status": status,
                 "error_code": error_code,
                 "duration_ms": duration_ms,
-                "data": {
-                    "matrix_event_id": event_id,
-                    "matrix_room_id": room_id,
-                    **(data or {}),
-                },
+                "data": {"matrix_event_id": event_id, "matrix_room_id": room_id, **(data or {})},
             },
             headers={"X-Internal-Service-Token": internal_token},
             timeout=_AUDIT_TIMEOUT_S,
@@ -124,19 +130,25 @@ async def _write_audit(
         logger.warning("Audit write failed (non-blocking): %s", exc)
 
 
-# ── Ingress loop ───────────────────────────────────────────────────────────────
+# ── Ingress loop (reader + workers) ───────────────────────────────────────────
 
 class MatrixIngressLoop:
     """
-    Drives Matrix sync-poll → rate-check → router-invoke → Matrix send_text.
+    Drives the full Matrix → Router → Matrix pipeline with backpressure.
+
+    Reader task:   sync → extract → rate_check → dedupe → queue.put_nowait
+    Worker tasks:  queue.get → invoke → send → audit
 
     Metric callbacks (all optional, called synchronously):
       on_message_received(room_id, agent_id)
       on_message_replied(room_id, agent_id, status)
       on_gateway_error(error_type)
-      on_rate_limited(room_id, agent_id, limit_type)       ← H1
-      on_invoke_latency(agent_id, duration_seconds)         ← H3
-      on_send_latency(agent_id, duration_seconds)           ← H3
+      on_rate_limited(room_id, agent_id, limit_type)
+      on_queue_dropped(room_id, agent_id)
+      on_queue_size(current_size: int)
+      on_invoke_latency(agent_id, duration_seconds)
+      on_send_latency(agent_id, duration_seconds)
+      on_queue_wait(agent_id, wait_seconds)
     """
 
     def __init__(
@@ -150,12 +162,19 @@ class MatrixIngressLoop:
         sofiia_console_url: str = "",
         sofiia_internal_token: str = "",
         rate_limiter: Optional[InMemoryRateLimiter] = None,
+        queue_max_events: int = 100,
+        worker_concurrency: int = 2,
+        queue_drain_timeout_s: float = 5.0,
+        # Callbacks
         on_message_received: Optional[Callable[[str, str], None]] = None,
         on_message_replied: Optional[Callable[[str, str, str], None]] = None,
         on_gateway_error: Optional[Callable[[str], None]] = None,
         on_rate_limited: Optional[Callable[[str, str, str], None]] = None,
+        on_queue_dropped: Optional[Callable[[str, str], None]] = None,
+        on_queue_size: Optional[Callable[[int], None]] = None,
         on_invoke_latency: Optional[Callable[[str, float], None]] = None,
         on_send_latency: Optional[Callable[[str, float], None]] = None,
+        on_queue_wait: Optional[Callable[[str, float], None]] = None,
     ) -> None:
         self._hs_url = matrix_homeserver_url
         self._token = matrix_access_token
@@ -166,28 +185,52 @@ class MatrixIngressLoop:
         self._console_url = sofiia_console_url
         self._internal_token = sofiia_internal_token
         self._rate_limiter = rate_limiter
+        self._queue_max = queue_max_events
+        self._worker_count = worker_concurrency
+        self._drain_timeout_s = queue_drain_timeout_s
+        # Callbacks
         self._on_message_received = on_message_received
         self._on_message_replied = on_message_replied
         self._on_gateway_error = on_gateway_error
         self._on_rate_limited = on_rate_limited
+        self._on_queue_dropped = on_queue_dropped
+        self._on_queue_size = on_queue_size
         self._on_invoke_latency = on_invoke_latency
         self._on_send_latency = on_send_latency
+        self._on_queue_wait = on_queue_wait
         self._next_batch: Optional[str] = None
+        self._queue: Optional[asyncio.Queue] = None  # exposed for /health
 
     @property
     def next_batch(self) -> Optional[str]:
         return self._next_batch
 
+    @property
+    def queue_size(self) -> int:
+        return self._queue.qsize() if self._queue else 0
+
+    @property
+    def worker_count(self) -> int:
+        return self._worker_count
+
+    # ── Public run ─────────────────────────────────────────────────────────────
+
     async def run(self, stop_event: asyncio.Event) -> None:
-        backoff = _INIT_RETRY_BACKOFF
         logger.info(
-            "Matrix ingress/egress loop started | hs=%s node=%s mappings=%d",
+            "Matrix ingress loop started | hs=%s node=%s mappings=%d "
+            "queue_max=%d workers=%d",
             self._hs_url, self._node_id, self._room_map.total_mappings,
+            self._queue_max, self._worker_count,
         )
 
         if self._room_map.total_mappings == 0:
             logger.warning("No room mappings — ingress loop is idle")
 
+        queue: asyncio.Queue[Optional[_QueueEntry]] = asyncio.Queue(
+            maxsize=self._queue_max
+        )
+        self._queue = queue
+
         async with MatrixClient(self._hs_url, self._token, self._user_id) as client:
             for mapping in self._room_map.mappings:
                 if mapping.agent_id in self._room_map.allowed_agents:
@@ -197,29 +240,75 @@ class MatrixIngressLoop:
                         logger.warning("Could not join room %s: %s", mapping.room_id, exc)
 
             async with httpx.AsyncClient() as http_client:
-                while not stop_event.is_set():
-                    try:
-                        sync_resp = await client.sync_poll(since=self._next_batch)
-                        self._next_batch = sync_resp.get("next_batch")
-                        backoff = _INIT_RETRY_BACKOFF
-                        await self._process_sync(client, http_client, sync_resp)
-                    except asyncio.CancelledError:
-                        break
-                    except Exception as exc:
-                        logger.error("Ingress loop error (retry in %.1fs): %s", backoff, exc)
-                        if self._on_gateway_error:
-                            self._on_gateway_error("sync_error")
-                        try:
-                            await asyncio.wait_for(stop_event.wait(), timeout=backoff)
-                        except asyncio.TimeoutError:
-                            pass
-                        backoff = min(backoff * 2, _MAX_RETRY_BACKOFF)
+                # Start workers
+                worker_tasks = [
+                    asyncio.create_task(
+                        self._worker(queue, client, http_client),
+                        name=f"matrix_worker_{i}",
+                    )
+                    for i in range(self._worker_count)
+                ]
 
-        logger.info("Matrix ingress/egress loop stopped")
+                # Run reader until stop_event
+                await self._reader(client, queue, http_client, stop_event)
 
-    async def _process_sync(
+                # Drain: wait for all enqueued items to be processed
+                logger.info(
+                    "Reader stopped. Draining queue (%d items, timeout=%.1fs)...",
+                    queue.qsize(), self._drain_timeout_s,
+                )
+                try:
+                    await asyncio.wait_for(queue.join(), timeout=self._drain_timeout_s)
+                    logger.info("Queue drained successfully")
+                except asyncio.TimeoutError:
+                    remaining = queue.qsize()
+                    logger.warning(
+                        "Drain timeout (%.1fs): %d items not processed",
+                        self._drain_timeout_s, remaining,
+                    )
+
+                # Cancel workers
+                for task in worker_tasks:
+                    task.cancel()
+                results = await asyncio.gather(*worker_tasks, return_exceptions=True)
+                cancelled = sum(1 for r in results if isinstance(r, asyncio.CancelledError))
+                logger.info("Workers stopped (%d cancelled)", cancelled)
+
+        self._queue = None
+        logger.info("Matrix ingress loop stopped")
+
+    # ── Reader ─────────────────────────────────────────────────────────────────
+
+    async def _reader(
         self,
         client: MatrixClient,
+        queue: "asyncio.Queue[Optional[_QueueEntry]]",
+        http_client: httpx.AsyncClient,
+        stop_event: asyncio.Event,
+    ) -> None:
+        backoff = _INIT_RETRY_BACKOFF
+        while not stop_event.is_set():
+            try:
+                sync_resp = await client.sync_poll(since=self._next_batch)
+                self._next_batch = sync_resp.get("next_batch")
+                backoff = _INIT_RETRY_BACKOFF
+                await self._enqueue_from_sync(client, queue, http_client, sync_resp)
+            except asyncio.CancelledError:
+                break
+            except Exception as exc:
+                logger.error("Reader error (retry in %.1fs): %s", backoff, exc)
+                if self._on_gateway_error:
+                    self._on_gateway_error("sync_error")
+                try:
+                    await asyncio.wait_for(stop_event.wait(), timeout=backoff)
+                except asyncio.TimeoutError:
+                    pass
+                backoff = min(backoff * 2, _MAX_RETRY_BACKOFF)
+
+    async def _enqueue_from_sync(
+        self,
+        client: MatrixClient,
+        queue: "asyncio.Queue[Optional[_QueueEntry]]",
         http_client: httpx.AsyncClient,
         sync_resp: Dict[str, Any],
     ) -> None:
@@ -228,14 +317,15 @@ class MatrixIngressLoop:
                 continue
             messages = client.extract_room_messages(sync_resp, mapping.room_id)
             for event in messages:
-                await self._handle_message(client, http_client, event, mapping)
+                await self._try_enqueue(client, queue, http_client, event, mapping)
 
-    async def _handle_message(
+    async def _try_enqueue(
         self,
         client: MatrixClient,
+        queue: "asyncio.Queue[Optional[_QueueEntry]]",
         http_client: httpx.AsyncClient,
         event: Dict[str, Any],
-        mapping,
+        mapping: RoomMapping,
     ) -> None:
         event_id = event.get("event_id", "")
         sender = event.get("sender", "")
@@ -246,7 +336,7 @@ class MatrixIngressLoop:
         if not text:
             return
 
-        # ── H1: Rate limit check ───────────────────────────────────────────────
+        # H1: Rate limit (before mark_seen — don't charge quota on drop)
         if self._rate_limiter is not None:
             allowed, limit_type = self._rate_limiter.check(room_id=room_id, sender=sender)
             if not allowed:
@@ -266,12 +356,81 @@ class MatrixIngressLoop:
                 )
                 return
 
-        # Dedupe — mark seen before any IO
+        # Dedupe — mark before enqueue (prevents double-enqueue on retry)
         client.mark_seen(event_id)
 
+        # H2: Enqueue or drop
+        entry = _QueueEntry(
+            event=event,
+            room_id=room_id,
+            agent_id=agent_id,
+            enqueue_time=time.monotonic(),
+        )
+        try:
+            queue.put_nowait(entry)
+            qsize = queue.qsize()
+            logger.debug("Enqueued event=%s qsize=%d", event_id, qsize)
+            if self._on_queue_size:
+                self._on_queue_size(qsize)
+        except asyncio.QueueFull:
+            logger.warning(
+                "Queue full (max=%d): dropping event=%s room=%s agent=%s",
+                self._queue_max, event_id, room_id, agent_id,
+            )
+            if self._on_queue_dropped:
+                self._on_queue_dropped(room_id, agent_id)
+            await _write_audit(
+                http_client, self._console_url, self._internal_token,
+                event="matrix.queue_full",
+                agent_id=agent_id, node_id=self._node_id,
+                room_id=room_id, event_id=event_id,
+                status="error", error_code="queue_full",
+                data={"queue_max": self._queue_max, "sender": sender},
+            )
+
+    # ── Worker ─────────────────────────────────────────────────────────────────
+
+    async def _worker(
+        self,
+        queue: "asyncio.Queue[Optional[_QueueEntry]]",
+        client: MatrixClient,
+        http_client: httpx.AsyncClient,
+    ) -> None:
+        """Consume queue entries until cancelled."""
+        while True:
+            entry = await queue.get()   # blocks until item available; raises CancelledError on cancel
+            try:
+                await self._process_entry(client, http_client, entry)
+            except Exception as exc:
+                logger.error("Worker unhandled error: %s", exc)
+            finally:
+                queue.task_done()
+                if self._on_queue_size:
+                    self._on_queue_size(queue.qsize())
+
+    # ── Process (invoke + send + audit) ───────────────────────────────────────
+
+    async def _process_entry(
+        self,
+        client: MatrixClient,
+        http_client: httpx.AsyncClient,
+        entry: _QueueEntry,
+    ) -> None:
+        event = entry.event
+        event_id = event.get("event_id", "")
+        sender = event.get("sender", "")
+        text = event.get("content", {}).get("body", "").strip()
+        room_id = entry.room_id
+        agent_id = entry.agent_id
+
+        # H3: Queue wait latency
+        wait_s = time.monotonic() - entry.enqueue_time
+        if self._on_queue_wait:
+            self._on_queue_wait(agent_id, wait_s)
+
         logger.info(
-            "Matrix message: room=%s sender=%s agent=%s event=%s len=%d",
-            room_id, sender, agent_id, event_id, len(text),
+            "Processing: room=%s agent=%s event=%s len=%d wait=%.3fs",
+            room_id, agent_id, event_id, len(text), wait_s,
         )
 
         if self._on_message_received:
@@ -283,148 +442,119 @@ class MatrixIngressLoop:
             agent_id=agent_id, node_id=self._node_id,
             room_id=room_id, event_id=event_id,
             status="ok",
-            data={"sender": sender, "text_len": len(text)},
+            data={"sender": sender, "text_len": len(text), "queue_wait_ms": int(wait_s * 1000)},
         )
 
         session_id = f"matrix:{room_id.replace('!', '').replace(':', '_')}"
 
-        # ── H3: Invoke with latency measurement ───────────────────────────────
+        # H3: Invoke with latency
         t0 = time.monotonic()
         reply_text: Optional[str] = None
         invoke_ok = False
-        invoke_duration_s: float = 0.0
+        invoke_duration_s = 0.0
 
         try:
             reply_text = await _invoke_router(
-                http_client,
-                self._router_url,
-                agent_id=agent_id,
-                node_id=self._node_id,
-                prompt=text,
-                session_id=session_id,
+                http_client, self._router_url,
+                agent_id=agent_id, node_id=self._node_id,
+                prompt=text, session_id=session_id,
             )
             invoke_ok = True
             invoke_duration_s = time.monotonic() - t0
-            duration_ms = int(invoke_duration_s * 1000)
-
             if self._on_invoke_latency:
                 self._on_invoke_latency(agent_id, invoke_duration_s)
-
             logger.info(
-                "Router invoke ok: agent=%s event=%s reply_len=%d duration=%dms",
-                agent_id, event_id, len(reply_text or ""), duration_ms,
+                "Invoke ok: agent=%s event=%s reply_len=%d duration=%dms",
+                agent_id, event_id, len(reply_text or ""), int(invoke_duration_s * 1000),
             )
 
         except httpx.HTTPStatusError as exc:
             invoke_duration_s = time.monotonic() - t0
-            duration_ms = int(invoke_duration_s * 1000)
             logger.error(
-                "Router HTTP %d for agent=%s event=%s duration=%dms",
-                exc.response.status_code, agent_id, event_id, duration_ms,
+                "Router HTTP %d agent=%s event=%s duration=%dms",
+                exc.response.status_code, agent_id, event_id, int(invoke_duration_s * 1000),
             )
             if self._on_gateway_error:
                 self._on_gateway_error(f"http_{exc.response.status_code}")
             await _write_audit(
                 http_client, self._console_url, self._internal_token,
-                event="matrix.error",
-                agent_id=agent_id, node_id=self._node_id,
+                event="matrix.error", agent_id=agent_id, node_id=self._node_id,
                 room_id=room_id, event_id=event_id,
                 status="error", error_code=f"router_http_{exc.response.status_code}",
-                duration_ms=duration_ms,
+                duration_ms=int(invoke_duration_s * 1000),
             )
 
         except (httpx.ConnectError, httpx.TimeoutException) as exc:
             invoke_duration_s = time.monotonic() - t0
-            duration_ms = int(invoke_duration_s * 1000)
-            logger.error(
-                "Router network error agent=%s event=%s: %s duration=%dms",
-                agent_id, event_id, exc, duration_ms,
-            )
+            logger.error("Router network error agent=%s event=%s: %s", agent_id, event_id, exc)
             if self._on_gateway_error:
                 self._on_gateway_error("network_error")
             await _write_audit(
                 http_client, self._console_url, self._internal_token,
-                event="matrix.error",
-                agent_id=agent_id, node_id=self._node_id,
+                event="matrix.error", agent_id=agent_id, node_id=self._node_id,
                 room_id=room_id, event_id=event_id,
                 status="error", error_code="router_network_error",
-                duration_ms=duration_ms,
+                duration_ms=int(invoke_duration_s * 1000),
             )
 
         except Exception as exc:
             invoke_duration_s = time.monotonic() - t0
-            duration_ms = int(invoke_duration_s * 1000)
-            logger.error(
-                "Unexpected router error agent=%s event=%s: %s",
-                agent_id, event_id, exc,
-            )
+            logger.error("Unexpected invoke error agent=%s event=%s: %s", agent_id, event_id, exc)
             if self._on_gateway_error:
                 self._on_gateway_error("unexpected")
             await _write_audit(
                 http_client, self._console_url, self._internal_token,
-                event="matrix.error",
-                agent_id=agent_id, node_id=self._node_id,
+                event="matrix.error", agent_id=agent_id, node_id=self._node_id,
                 room_id=room_id, event_id=event_id,
                 status="error", error_code="router_unexpected",
-                duration_ms=duration_ms,
+                duration_ms=int(invoke_duration_s * 1000),
             )
 
-        if not invoke_ok:
+        if not invoke_ok or not reply_text:
+            if invoke_ok:
+                logger.warning("Empty reply from router agent=%s event=%s", agent_id, event_id)
             return
 
-        if not reply_text:
-            logger.warning("Empty reply from router for agent=%s event=%s", agent_id, event_id)
-            return
-
-        # ── H3: Send with latency measurement ─────────────────────────────────
+        # H3: Send with latency
         send_text = reply_text[:_REPLY_TEXT_MAX]
         txn_id = MatrixClient.make_txn_id(room_id, event_id)
-
         send_t0 = time.monotonic()
+
         try:
             await client.send_text(room_id, send_text, txn_id)
             send_duration_s = time.monotonic() - send_t0
-            send_duration_ms = int(send_duration_s * 1000)
-
             if self._on_send_latency:
                 self._on_send_latency(agent_id, send_duration_s)
             if self._on_message_replied:
                 self._on_message_replied(room_id, agent_id, "ok")
-
             await _write_audit(
                 http_client, self._console_url, self._internal_token,
-                event="matrix.agent.replied",
-                agent_id=agent_id, node_id=self._node_id,
-                room_id=room_id, event_id=event_id,
-                status="ok",
-                duration_ms=send_duration_ms,
+                event="matrix.agent.replied", agent_id=agent_id, node_id=self._node_id,
+                room_id=room_id, event_id=event_id, status="ok",
+                duration_ms=int(send_duration_s * 1000),
                 data={
                     "reply_len": len(send_text),
                     "truncated": len(reply_text) > _REPLY_TEXT_MAX,
                     "router_duration_ms": int(invoke_duration_s * 1000),
+                    "queue_wait_ms": int(wait_s * 1000),
                 },
             )
             logger.info(
                 "Reply sent: agent=%s event=%s reply_len=%d send_ms=%d",
-                agent_id, event_id, len(send_text), send_duration_ms,
+                agent_id, event_id, len(send_text), int(send_duration_s * 1000),
             )
 
         except Exception as exc:
             send_duration_s = time.monotonic() - send_t0
-            send_duration_ms = int(send_duration_s * 1000)
-            logger.error(
-                "Failed to send Matrix reply agent=%s event=%s: %s",
-                agent_id, event_id, exc,
-            )
+            logger.error("Send failed agent=%s event=%s: %s", agent_id, event_id, exc)
             if self._on_message_replied:
                 self._on_message_replied(room_id, agent_id, "error")
             if self._on_gateway_error:
                 self._on_gateway_error("matrix_send_error")
             await _write_audit(
                 http_client, self._console_url, self._internal_token,
-                event="matrix.error",
-                agent_id=agent_id, node_id=self._node_id,
+                event="matrix.error", agent_id=agent_id, node_id=self._node_id,
                 room_id=room_id, event_id=event_id,
                 status="error", error_code="matrix_send_failed",
-                duration_ms=send_duration_ms,
+                duration_ms=int(send_duration_s * 1000),
             )

services/matrix-bridge-dagi/app/main.py

@@ -88,6 +88,22 @@ if _PROM_OK:
         "matrix_bridge_rate_limiter_active_senders",
         "Senders with activity in the current rate-limit window",
     )
+    # H2: Queue metrics
+    _queue_size = Gauge(
+        "matrix_bridge_queue_size",
+        "Current number of pending items in the work queue",
+    )
+    _queue_dropped = Counter(
+        "matrix_bridge_queue_dropped_total",
+        "Messages dropped because queue was full",
+        ["room_id", "agent_id"],
+    )
+    _queue_wait = Histogram(
+        "matrix_bridge_queue_wait_seconds",
+        "Time between enqueue and worker start processing",
+        ["agent_id"],
+        buckets=[0.01, 0.05, 0.1, 0.5, 1.0, 5.0, 30.0],
+    )
 
 # ── Startup state ─────────────────────────────────────────────────────────────
 _START_TIME = time.monotonic()
@@ -97,6 +113,7 @@ _matrix_reachable: Optional[bool] = None
 _gateway_reachable: Optional[bool] = None
 _room_map: Optional[RoomMappingConfig] = None
 _rate_limiter: Optional[InMemoryRateLimiter] = None
+_ingress_loop: Optional["MatrixIngressLoop"] = None   # for /health queue_size
 _ingress_task: Optional[asyncio.Task] = None
 _ingress_stop: Optional[asyncio.Event] = None
 
@@ -116,7 +133,7 @@ async def _probe_url(url: str, timeout: float = 5.0) -> bool:
 @asynccontextmanager
 async def lifespan(app_: Any):
     global _cfg, _config_error, _matrix_reachable, _gateway_reachable
-    global _room_map, _rate_limiter
+    global _room_map, _rate_limiter, _ingress_loop
     try:
         _cfg = load_config()
 
@@ -185,7 +202,6 @@ async def lifespan(app_: Any):
                     _messages_rate_limited.labels(
                         room_id=room_id, agent_id=agent_id, limit_type=limit_type
                     ).inc()
-                    # Update active room/sender gauges from limiter stats
                     if _rate_limiter is not None:
                         stats = _rate_limiter.stats()
                         _rate_limiter_active_rooms.set(stats["active_rooms"])
@@ -199,6 +215,19 @@ async def lifespan(app_: Any):
                 if _PROM_OK:
                     _send_latency.labels(agent_id=agent_id).observe(duration_s)
 
+            # H2 callbacks
+            def _on_queue_dropped(room_id: str, agent_id: str) -> None:
+                if _PROM_OK:
+                    _queue_dropped.labels(room_id=room_id, agent_id=agent_id).inc()
+
+            def _on_queue_size(size: int) -> None:
+                if _PROM_OK:
+                    _queue_size.set(size)
+
+            def _on_queue_wait(agent_id: str, wait_s: float) -> None:
+                if _PROM_OK:
+                    _queue_wait.labels(agent_id=agent_id).observe(wait_s)
+
             ingress = MatrixIngressLoop(
                 matrix_homeserver_url=_cfg.matrix_homeserver_url,
                 matrix_access_token=_cfg.matrix_access_token,
@@ -209,13 +238,24 @@ async def lifespan(app_: Any):
                 sofiia_console_url=_cfg.sofiia_console_url,
                 sofiia_internal_token=_cfg.sofiia_internal_token,
                 rate_limiter=_rate_limiter,
+                queue_max_events=_cfg.queue_max_events,
+                worker_concurrency=_cfg.worker_concurrency,
+                queue_drain_timeout_s=_cfg.queue_drain_timeout_s,
                 on_message_received=_on_msg,
                 on_message_replied=_on_replied,
                 on_gateway_error=_on_gw_error,
                 on_rate_limited=_on_rate_limited,
+                on_queue_dropped=_on_queue_dropped,
+                on_queue_size=_on_queue_size,
                 on_invoke_latency=_on_invoke_latency,
                 on_send_latency=_on_send_latency,
+                on_queue_wait=_on_queue_wait,
             )
+            logger.info(
+                "✅ Backpressure queue: max=%d workers=%d drain_timeout=%.1fs",
+                _cfg.queue_max_events, _cfg.worker_concurrency, _cfg.queue_drain_timeout_s,
+            )
+            _ingress_loop = ingress
             _ingress_task = asyncio.create_task(
                 ingress.run(_ingress_stop),
                 name="matrix_ingress_loop",
@@ -290,6 +330,11 @@ async def health() -> Dict[str, Any]:
         "mappings_count": _room_map.total_mappings if _room_map else 0,
         "config_ok": True,
         "rate_limiter": _rate_limiter.stats() if _rate_limiter else None,
+        "queue": {
+            "size": _ingress_loop.queue_size if _ingress_loop else 0,
+            "max": _cfg.queue_max_events,
+            "workers": _cfg.worker_concurrency,
+        },
     }
 
 

tests/test_matrix_bridge_queue.py

@@ -0,0 +1,519 @@
+"""
+Tests for matrix-bridge-dagi H2: Backpressure queue (reader + workers)
+
+Coverage:
+  1. enqueue_and_process   — single event enqueued → worker processes exactly once
+  2. queue_full_drop       — queue full → dropped + on_queue_dropped + audit matrix.queue_full
+  3. concurrency           — 2 events processed concurrently (worker_concurrency=2)
+  4. graceful_shutdown     — stop_event set → in-flight items completed, queue drained
+  5. queue_wait_metric     — on_queue_wait called with agent_id + float
+  6. rate_limit_before_enqueue — rate-limited event never enters queue
+"""
+
+import asyncio
+import sys
+import time
+from pathlib import Path
+from unittest.mock import AsyncMock, MagicMock, patch
+
+_BRIDGE = Path(__file__).parent.parent / "services" / "matrix-bridge-dagi"
+if str(_BRIDGE) not in sys.path:
+    sys.path.insert(0, str(_BRIDGE))
+
+from app.ingress import MatrixIngressLoop, _QueueEntry  # noqa: E402
+from app.rate_limit import InMemoryRateLimiter          # noqa: E402
+from app.room_mapping import parse_room_map              # noqa: E402
+
+
+def run(coro):
+    return asyncio.run(coro)
+
+
+ALLOWED = frozenset({"sofiia"})
+ROOM_ID = "!QwHczWXgefDHBEVkTH:daarion.space"
+ROOM_MAP_STR = f"sofiia:{ROOM_ID}"
+ROUTER_URL = "http://dagi-router-node1:8000"
+HS_URL = "http://dagi-synapse-node1:8008"
+CONSOLE_URL = "http://dagi-sofiia-console-node1:8002"
+INTERNAL_TOKEN = "test_tok"
+BOT_USER = "@dagi_bridge:daarion.space"
+USER = "@user:daarion.space"
+
+
+def _make_event(event_id: str = "$e1:s", body: str = "Hello") -> dict:
+    return {
+        "type": "m.room.message",
+        "event_id": event_id,
+        "sender": USER,
+        "content": {"msgtype": "m.text", "body": body},
+        "origin_server_ts": 1000,
+    }
+
+
+def _fake_sync(events: list) -> dict:
+    return {
+        "next_batch": "s_next",
+        "rooms": {"join": {ROOM_ID: {"timeline": {"events": events}}}},
+    }
+
+
+def _ok_router_resp(text: str = "Reply!") -> MagicMock:
+    r = MagicMock()
+    r.status_code = 200
+    r.json.return_value = {"response": text}
+    r.raise_for_status = MagicMock()
+    return r
+
+
+def _audit_resp() -> MagicMock:
+    r = MagicMock()
+    r.status_code = 200
+    r.json.return_value = {"ok": True}
+    r.raise_for_status = MagicMock()
+    return r
+
+
+def _make_loop(**kwargs) -> MatrixIngressLoop:
+    room_map = parse_room_map(ROOM_MAP_STR, ALLOWED)
+    defaults = dict(
+        matrix_homeserver_url=HS_URL,
+        matrix_access_token="tok",
+        matrix_user_id=BOT_USER,
+        router_url=ROUTER_URL,
+        node_id="NODA1",
+        room_map=room_map,
+        sofiia_console_url=CONSOLE_URL,
+        sofiia_internal_token=INTERNAL_TOKEN,
+        queue_max_events=10,
+        worker_concurrency=1,
+        queue_drain_timeout_s=2.0,
+    )
+    defaults.update(kwargs)
+    return MatrixIngressLoop(**defaults)
+
+
+def _make_mock_client(events_per_sync: list, stop_after: int = 1) -> tuple:
+    """Returns (mock_client_class, mock_client_instance)."""
+    call_count = [0]
+    seen = set()
+
+    def fake_extract(sync_resp, room_id):
+        evts = (sync_resp.get("rooms", {}).get("join", {})
+                .get(room_id, {}).get("timeline", {}).get("events", []))
+        return [e for e in evts
+                if e.get("type") == "m.room.message"
+                and e.get("sender") != BOT_USER
+                and e.get("event_id") not in seen]
+
+    def fake_mark_seen(eid):
+        seen.add(eid)
+
+    async def fake_sync_poll(stop_event_ref=None, **kwargs):
+        call_count[0] += 1
+        if call_count[0] > stop_after:
+            # Block until cancelled
+            await asyncio.sleep(1000)
+        return _fake_sync(events_per_sync)
+
+    mock_mc = AsyncMock()
+    mock_mc.__aenter__ = AsyncMock(return_value=mock_mc)
+    mock_mc.__aexit__ = AsyncMock(return_value=False)
+    mock_mc.join_room = AsyncMock()
+    mock_mc.mark_seen = MagicMock(side_effect=fake_mark_seen)
+    mock_mc.extract_room_messages = fake_extract
+    mock_mc.send_text = AsyncMock(return_value={"event_id": "$reply"})
+    mock_mc.sync_poll = fake_sync_poll
+
+    return mock_mc
+
+
+# ── Test 1: enqueue and process exactly once ───────────────────────────────────
+
+def test_enqueue_and_process_exactly_once():
+    """One event → router invoked once, send_text called once."""
+    async def _inner():
+        invoke_count = [0]
+        send_calls = []
+        stop = asyncio.Event()
+        loop = _make_loop(worker_concurrency=1)
+        events = [_make_event("$e1:s", "Hello")]
+        mock_mc = _make_mock_client(events, stop_after=1)
+        mock_mc.sync_poll = _stop_after_n_syncs(1, events, stop)
+
+        async def fake_http_post(url, *, json=None, headers=None, timeout=None):
+            if "/infer" in url:
+                invoke_count[0] += 1
+                await asyncio.sleep(0.01)
+                return _ok_router_resp("Hi!")
+            return _audit_resp()
+
+        with patch("app.ingress.MatrixClient") as MC:
+            MC.return_value = mock_mc
+            MC.make_txn_id = lambda r, e: f"txn_{e}"
+            with patch("app.ingress.httpx.AsyncClient") as MH:
+                mh = AsyncMock()
+                mh.__aenter__ = AsyncMock(return_value=mh)
+                mh.__aexit__ = AsyncMock(return_value=False)
+                mh.post = fake_http_post
+                MH.return_value = mh
+                await asyncio.wait_for(loop.run(stop), timeout=5.0)
+
+        assert invoke_count[0] == 1
+        assert mock_mc.send_text.call_count == 1
+
+    run(_inner())
+
+
+def _stop_after_n_syncs(n: int, events: list, stop: asyncio.Event):
+    call_count = [0]
+
+    async def fake_sync_poll(**kwargs):
+        call_count[0] += 1
+        if call_count[0] > n:
+            stop.set()
+            await asyncio.sleep(1000)
+        return _fake_sync(events)
+
+    return fake_sync_poll
+
+
+# ── Test 2: queue full → drop + callback + audit ───────────────────────────────
+
+def test_queue_full_drop_and_audit():
+    """When queue is full, extra events are dropped with audit matrix.queue_full."""
+    async def _inner():
+        dropped = []
+        audit_events = []
+
+        # queue_max=1, worker_concurrency=1 but worker is slow (blocks on barrier)
+        barrier = asyncio.Event()
+        stop = asyncio.Event()
+
+        loop = _make_loop(
+            queue_max_events=1,
+            worker_concurrency=1,
+            on_queue_dropped=lambda r, a: dropped.append((r, a)),
+        )
+
+        # 3 events in one sync — queue holds 1, worker blocked, 2 dropped
+        events = [
+            _make_event("$e1:s", "msg1"),
+            _make_event("$e2:s", "msg2"),
+            _make_event("$e3:s", "msg3"),
+        ]
+        call_count = [0]
+
+        async def fake_sync_poll(**kwargs):
+            call_count[0] += 1
+            if call_count[0] > 1:
+                stop.set()
+                await asyncio.sleep(1000)
+            return _fake_sync(events)
+
+        async def slow_http_post(url, *, json=None, headers=None, timeout=None):
+            if "/infer" in url:
+                await barrier.wait()    # block worker indefinitely
+                return _ok_router_resp()
+            if "/audit/internal" in url:
+                event_name = (json or {}).get("event", "")
+                audit_events.append(event_name)
+            return _audit_resp()
+
+        mock_mc = AsyncMock()
+        mock_mc.__aenter__ = AsyncMock(return_value=mock_mc)
+        mock_mc.__aexit__ = AsyncMock(return_value=False)
+        mock_mc.join_room = AsyncMock()
+        mock_mc.mark_seen = MagicMock()
+        mock_mc.send_text = AsyncMock(return_value={"event_id": "$r"})
+        seen: set = set()
+
+        def fake_extract(sync_resp, room_id):
+            evts = (sync_resp.get("rooms", {}).get("join", {})
+                    .get(room_id, {}).get("timeline", {}).get("events", []))
+            return [e for e in evts
+                    if e.get("type") == "m.room.message"
+                    and e.get("sender") != BOT_USER
+                    and e.get("event_id") not in seen]
+
+        mock_mc.extract_room_messages = fake_extract
+        mock_mc.sync_poll = fake_sync_poll
+
+        with patch("app.ingress.MatrixClient") as MC:
+            MC.return_value = mock_mc
+            MC.make_txn_id = lambda r, e: f"txn_{e}"
+            with patch("app.ingress.httpx.AsyncClient") as MH:
+                mh = AsyncMock()
+                mh.__aenter__ = AsyncMock(return_value=mh)
+                mh.__aexit__ = AsyncMock(return_value=False)
+                mh.post = slow_http_post
+                MH.return_value = mh
+                # Run with short timeout — worker will be stuck but reader finishes
+                try:
+                    await asyncio.wait_for(loop.run(stop), timeout=2.5)
+                except asyncio.TimeoutError:
+                    pass
+
+        # At least 2 events dropped (queue_max=1, 3 incoming)
+        assert len(dropped) >= 2, f"expected >=2 drops, got {dropped}"
+        assert "matrix.queue_full" in audit_events
+
+    run(_inner())
+
+
+# ── Test 3: concurrency — 2 workers process 2 events in parallel ──────────────
+
+def test_two_workers_process_concurrently():
+    """With worker_concurrency=2, two events start processing without waiting for each other."""
+    async def _inner():
+        started_order = []
+        finished_order = []
+        barrier = asyncio.Barrier(2)   # both workers must arrive before either proceeds
+        stop = asyncio.Event()
+
+        loop = _make_loop(worker_concurrency=2, queue_max_events=10)
+        events = [_make_event("$e1:s", "msg1"), _make_event("$e2:s", "msg2")]
+        call_count = [0]
+
+        async def fake_sync_poll(**kwargs):
+            call_count[0] += 1
+            if call_count[0] > 1:
+                stop.set()
+                await asyncio.sleep(1000)
+            return _fake_sync(events)
+
+        async def concurrent_http_post(url, *, json=None, headers=None, timeout=None):
+            if "/infer" in url:
+                prompt = (json or {}).get("prompt", "")
+                started_order.append(prompt)
+                await barrier.wait()   # both workers sync here → proves concurrency
+                finished_order.append(prompt)
+                return _ok_router_resp(f"reply to {prompt}")
+            return _audit_resp()
+
+        seen: set = set()
+        mock_mc = AsyncMock()
+        mock_mc.__aenter__ = AsyncMock(return_value=mock_mc)
+        mock_mc.__aexit__ = AsyncMock(return_value=False)
+        mock_mc.join_room = AsyncMock()
+        mock_mc.mark_seen = MagicMock(side_effect=lambda e: seen.add(e))
+        mock_mc.send_text = AsyncMock(return_value={"event_id": "$r"})
+
+        def fake_extract(sync_resp, room_id):
+            evts = (sync_resp.get("rooms", {}).get("join", {})
+                    .get(room_id, {}).get("timeline", {}).get("events", []))
+            return [e for e in evts
+                    if e.get("type") == "m.room.message"
+                    and e.get("sender") != BOT_USER
+                    and e.get("event_id") not in seen]
+
+        mock_mc.extract_room_messages = fake_extract
+        mock_mc.sync_poll = fake_sync_poll
+
+        with patch("app.ingress.MatrixClient") as MC:
+            MC.return_value = mock_mc
+            MC.make_txn_id = lambda r, e: f"txn_{e}"
+            with patch("app.ingress.httpx.AsyncClient") as MH:
+                mh = AsyncMock()
+                mh.__aenter__ = AsyncMock(return_value=mh)
+                mh.__aexit__ = AsyncMock(return_value=False)
+                mh.post = concurrent_http_post
+                MH.return_value = mh
+                await asyncio.wait_for(loop.run(stop), timeout=5.0)
+
+        # Both events started before either finished → they ran concurrently
+        assert len(started_order) == 2
+        assert len(finished_order) == 2
+        # The barrier ensures both were in-flight at the same time
+        # If sequential, the barrier would deadlock (only 1 worker reaches it)
+
+    run(_inner())
+
+
+# ── Test 4: graceful shutdown — in-flight items completed ─────────────────────
+
+def test_graceful_shutdown_drains_queue():
+    """Stop event set mid-flight → worker finishes current item before exit."""
+    async def _inner():
+        completed = []
+        stop = asyncio.Event()
+
+        loop = _make_loop(worker_concurrency=1, queue_max_events=5, queue_drain_timeout_s=3.0)
+        events = [_make_event("$e1:s", "drain me")]
+        call_count = [0]
+
+        async def fake_sync_poll(**kwargs):
+            call_count[0] += 1
+            if call_count[0] == 1:
+                # Set stop after first sync so reader exits but queue has item
+                asyncio.get_event_loop().call_later(0.05, stop.set)
+            if call_count[0] > 1:
+                await asyncio.sleep(1000)
+            return _fake_sync(events)
+
+        async def fake_http_post(url, *, json=None, headers=None, timeout=None):
+            if "/infer" in url:
+                await asyncio.sleep(0.1)   # simulate slow router
+                completed.append("done")
+                return _ok_router_resp("ok")
+            return _audit_resp()
+
+        seen: set = set()
+        mock_mc = AsyncMock()
+        mock_mc.__aenter__ = AsyncMock(return_value=mock_mc)
+        mock_mc.__aexit__ = AsyncMock(return_value=False)
+        mock_mc.join_room = AsyncMock()
+        mock_mc.mark_seen = MagicMock(side_effect=lambda e: seen.add(e))
+        mock_mc.send_text = AsyncMock(return_value={"event_id": "$r"})
+
+        def fake_extract(sync_resp, room_id):
+            evts = (sync_resp.get("rooms", {}).get("join", {})
+                    .get(room_id, {}).get("timeline", {}).get("events", []))
+            return [e for e in evts
+                    if e.get("type") == "m.room.message"
+                    and e.get("sender") != BOT_USER
+                    and e.get("event_id") not in seen]
+
+        mock_mc.extract_room_messages = fake_extract
+        mock_mc.sync_poll = fake_sync_poll
+
+        with patch("app.ingress.MatrixClient") as MC:
+            MC.return_value = mock_mc
+            MC.make_txn_id = lambda r, e: f"txn_{e}"
+            with patch("app.ingress.httpx.AsyncClient") as MH:
+                mh = AsyncMock()
+                mh.__aenter__ = AsyncMock(return_value=mh)
+                mh.__aexit__ = AsyncMock(return_value=False)
+                mh.post = fake_http_post
+                MH.return_value = mh
+                await asyncio.wait_for(loop.run(stop), timeout=5.0)
+
+        # Worker completed the in-flight invoke before shutdown
+        assert completed == ["done"]
+
+    run(_inner())
+
+
+# ── Test 5: queue_wait metric callback ────────────────────────────────────────
+
+def test_queue_wait_metric_fires():
+    """on_queue_wait must be called with (agent_id, float >= 0)."""
+    async def _inner():
+        wait_calls = []
+        stop = asyncio.Event()
+        loop = _make_loop(
+            on_queue_wait=lambda a, w: wait_calls.append((a, w)),
+        )
+        events = [_make_event()]
+        call_count = [0]
+
+        async def fake_sync_poll(**kwargs):
+            call_count[0] += 1
+            if call_count[0] > 1:
+                stop.set()
+                await asyncio.sleep(1000)
+            return _fake_sync(events)
+
+        async def fake_http_post(url, *, json=None, headers=None, timeout=None):
+            if "/infer" in url:
+                return _ok_router_resp()
+            return _audit_resp()
+
+        seen: set = set()
+        mock_mc = AsyncMock()
+        mock_mc.__aenter__ = AsyncMock(return_value=mock_mc)
+        mock_mc.__aexit__ = AsyncMock(return_value=False)
+        mock_mc.join_room = AsyncMock()
+        mock_mc.mark_seen = MagicMock(side_effect=lambda e: seen.add(e))
+        mock_mc.send_text = AsyncMock(return_value={"event_id": "$r"})
+
+        def fake_extract(sync_resp, room_id):
+            evts = (sync_resp.get("rooms", {}).get("join", {})
+                    .get(room_id, {}).get("timeline", {}).get("events", []))
+            return [e for e in evts
+                    if e.get("type") == "m.room.message"
+                    and e.get("sender") != BOT_USER
+                    and e.get("event_id") not in seen]
+
+        mock_mc.extract_room_messages = fake_extract
+        mock_mc.sync_poll = fake_sync_poll
+
+        with patch("app.ingress.MatrixClient") as MC:
+            MC.return_value = mock_mc
+            MC.make_txn_id = lambda r, e: f"txn_{e}"
+            with patch("app.ingress.httpx.AsyncClient") as MH:
+                mh = AsyncMock()
+                mh.__aenter__ = AsyncMock(return_value=mh)
+                mh.__aexit__ = AsyncMock(return_value=False)
+                mh.post = fake_http_post
+                MH.return_value = mh
+                await asyncio.wait_for(loop.run(stop), timeout=5.0)
+
+        assert len(wait_calls) == 1
+        assert wait_calls[0][0] == "sofiia"
+        assert isinstance(wait_calls[0][1], float)
+        assert wait_calls[0][1] >= 0.0
+
+    run(_inner())
+
+
+# ── Test 6: rate-limited event never enters queue ─────────────────────────────
+
+def test_rate_limited_event_not_enqueued():
+    """Rate-limited event must not enter the queue (invoke never called)."""
+    async def _inner():
+        invoke_count = [0]
+        stop = asyncio.Event()
+
+        rl = InMemoryRateLimiter(room_rpm=1, sender_rpm=100)
+        loop = _make_loop(rate_limiter=rl, worker_concurrency=1)
+
+        # 2 events — first passes, second blocked by rate limiter
+        events = [_make_event("$e1:s", "first"), _make_event("$e2:s", "second")]
+        call_count = [0]
+
+        async def fake_sync_poll(**kwargs):
+            call_count[0] += 1
+            if call_count[0] > 1:
+                stop.set()
+                await asyncio.sleep(1000)
+            return _fake_sync(events)
+
+        async def fake_http_post(url, *, json=None, headers=None, timeout=None):
+            if "/infer" in url:
+                invoke_count[0] += 1
+                return _ok_router_resp()
+            return _audit_resp()
+
+        seen: set = set()
+        mock_mc = AsyncMock()
+        mock_mc.__aenter__ = AsyncMock(return_value=mock_mc)
+        mock_mc.__aexit__ = AsyncMock(return_value=False)
+        mock_mc.join_room = AsyncMock()
+        mock_mc.mark_seen = MagicMock(side_effect=lambda e: seen.add(e))
+        mock_mc.send_text = AsyncMock(return_value={"event_id": "$r"})
+
+        def fake_extract(sync_resp, room_id):
+            evts = (sync_resp.get("rooms", {}).get("join", {})
+                    .get(room_id, {}).get("timeline", {}).get("events", []))
+            return [e for e in evts
+                    if e.get("type") == "m.room.message"
+                    and e.get("sender") != BOT_USER
+                    and e.get("event_id") not in seen]
+
+        mock_mc.extract_room_messages = fake_extract
+        mock_mc.sync_poll = fake_sync_poll
+
+        with patch("app.ingress.MatrixClient") as MC:
+            MC.return_value = mock_mc
+            MC.make_txn_id = lambda r, e: f"txn_{e}"
+            with patch("app.ingress.httpx.AsyncClient") as MH:
+                mh = AsyncMock()
+                mh.__aenter__ = AsyncMock(return_value=mh)
+                mh.__aexit__ = AsyncMock(return_value=False)
+                mh.post = fake_http_post
+                MH.return_value = mh
+                await asyncio.wait_for(loop.run(stop), timeout=5.0)
+
+        assert invoke_count[0] == 1   # only first event processed
+
+    run(_inner())