drb-c2-core/app/internal/weather.py

@@ -48,6 +48,11 @@ ZONE_CONCURRENCY = 6
 # the next refresh (the zone cache fills across refreshes); meanwhile the alert
 # shows in zone_only or with the zones resolved so far.
 MAX_ZONE_FETCHES_PER_REFRESH = 80
+# Wall-clock cap on one refresh's zone fetches. The refresh runs under the feed
+# lock, so without it a slow zone endpoint (20 s timeout x 80 fetches / 6 wide)
+# could hold every /weather/alerts request for minutes. Unfinished zones are
+# simply not cached and are retried next refresh.
+ZONE_BUDGET_S = 15.0
 
 WIND_TTL_S = 5 * 60
 WIND_CELL_DEG = 0.5      # wind cache cell; one upstream call per cell per TTL
@@ -61,7 +66,7 @@ _warned_ua = False
 def _headers(accept: str) -> Dict[str, str]:
     global _warned_ua
     ua = settings.nws_user_agent or DEFAULT_USER_AGENT
-    if ua == DEFAULT_USER_AGENT and not _warned_ua:
+    if (ua == DEFAULT_USER_AGENT or "example.com" in ua) and not _warned_ua:
         _warned_ua = True
         logger.warning("NWS_USER_AGENT is the placeholder -- set it to 'app (contact email)'; NWS may block us")
     return {"User-Agent": ua, "Accept": accept}
@@ -193,7 +198,12 @@ async def _resolve_zones(client: "httpx.AsyncClient", alerts: List[Dict[str, Any
                     wanted.append(z)
     if wanted:
         sem = asyncio.Semaphore(ZONE_CONCURRENCY)
-        await asyncio.gather(*(_fetch_zone(client, z, sem) for z in wanted[:MAX_ZONE_FETCHES_PER_REFRESH]))
+        try:
+            await asyncio.wait_for(
+                asyncio.gather(*(_fetch_zone(client, z, sem) for z in wanted[:MAX_ZONE_FETCHES_PER_REFRESH])),
+                timeout=ZONE_BUDGET_S)
+        except asyncio.TimeoutError:
+            logger.warning(f"NWS zone fetches over {ZONE_BUDGET_S:.0f}s budget; the rest resolve next refresh")
     for a in alerts:
         if a["geometry"] is not None:
             continue

drb-c2-core/tests/test_weather.py

@@ -216,3 +216,24 @@ def test_wind_unavailable_reasons():
     with patch.object(weather.httpx, "AsyncClient", FakeHttp({weather.METAR_URL: Resp(status=503)})):
         body = client.get("/weather/wind", params={"lat": 41.03, "lon": -73.76}).json()
     assert body["available"] is False and "unavailable" in body["reason"] and "503" in body["error"]
+
+
+def test_slow_zone_endpoint_cannot_hold_the_feed_lock():
+    """Zone fetches run under the feed lock; a hung zone endpoint must not stall
+    every /weather/alerts request -- the alert lands in zone_only instead."""
+    import time
+
+    class SlowZones(FakeHttp):
+        async def get(self, url, params=None, headers=None):
+            if url.startswith(weather.NWS_ZONE_PREFIX):
+                await asyncio.sleep(5)
+            return await super().get(url, params, headers)
+
+    http = SlowZones({weather.NWS_ALERTS_URL: Resp({"features": [_alert("z1", "Flood Watch", None, [ZONE_A])]}),
+                      ZONE_A: Resp({"geometry": ZONE_POLY})})
+    with patch.object(weather, "ZONE_BUDGET_S", 0.2), patch.object(weather.httpx, "AsyncClient", http):
+        t0 = time.time()
+        feed = asyncio.run(weather.alerts.get())
+    assert time.time() - t0 < 2
+    assert feed.error is None and [z["id"] for z in feed.zone_only] == ["z1"]
+    assert weather.zones.get(ZONE_A) == (False, None)   # not cached as a failure; retried next refresh

infra/ansible/roles/deploy/templates/c2-core.env.j2

@@ -44,3 +44,8 @@ AGENT_SERVICE_KEY={{ vault_agent_service_key | default('') }}
 # frontend call to the API would have failed CORS. If the frontend ever moves
 # to app.{{ domain }}, change this at the same time.
 CORS_ORIGINS=["https://{{ domain }}"]
+
+# NWS weather alerts / METAR wind (server-26#237). api.weather.gov requires a
+# User-Agent naming the app and a contact; default to the site URL so a vault
+# that predates this key still sends an identifiable agent, not the placeholder.
+NWS_USER_AGENT={{ vault_nws_user_agent | default('DRB scanner map (https://' ~ domain ~ ')', true) }}

infra/ansible/vault.yml.example

@@ -60,3 +60,5 @@ vault_firebase_app_id: ""
 
 # No GCP key needed — the VM uses Application Default Credentials via the
 # GCE metadata server. Terraform grants the required IAM roles at apply time.
+
+vault_nws_user_agent: ""        # optional; e.g. "DRB scanner map (ops@yourdomain)" -- blank/absent uses the site URL