Merge pull request #530 from SimmerV/bug-fix

fix: NaN telemetry writes, docker-compose port exposure, /v1/events docs

Author: SimmerV

API.md

@@ -35,6 +35,12 @@ directly for integrations. Responses are JSON unless noted.
 | GET | `/v1/traceroutes` | All recent traceroutes. |
 | GET | `/v1/stats` | Mesh totals (counts, top nodes, modem preset, etc.). |
 
+### Live events (SSE)
+
+| Method | Path | Notes |
+|---|---|---|
+| GET | `/v1/events` | Server-Sent Events stream of live updates, one multiplexed connection per client. Each frame's `event:` type is `node`, `chat`, `telemetry`, or `coverage`; `: ...` comment frames are keep-alive heartbeats. Served unbuffered through Caddy (`flush_interval -1`); the SPA reaches it as `/api/v1/events`. |
+
 ### Static map / Server
 
 | Method | Path | Notes |

docker-compose.yml

@@ -17,8 +17,10 @@ services:
       - "maintenance_work_mem=${PG_MAINTENANCE_WORK_MEM:-64MB}"
       - "-c"
       - "effective_cache_size=${PG_EFFECTIVE_CACHE_SIZE:-1GB}"
-    ports:
-      - "5432:5432"
+    # No published port: the app reaches Postgres over the compose network,
+    # and docker-published ports bypass host firewalls (ufw). For host access
+    # use `docker compose exec postgres psql -U postgres meshinfo`, or map
+    # "127.0.0.1:5432:5432" in a docker-compose.override.yml.
     volumes:
       # postgres:18+ stores data in a version-specific subdir; mount the
       # parent (/var/lib/postgresql), not /data. See POSTGRES.md.
@@ -32,6 +34,9 @@ services:
   mqtt:
     container_name: mqtt
     image: eclipse-mosquitto:latest
+    # 1883 stays published on all interfaces by design — mesh gateways on the
+    # WAN publish uplinks here. Protect it with broker auth: see the
+    # password_file/acl_file notes in mosquitto/config/mosquitto.conf.sample.
     ports:
       - 1883:1883
     volumes:
@@ -54,8 +59,10 @@ services:
     # dns:
     #   - 1.1.1.1
     #   - 1.0.0.1
+    # Localhost-only: caddy is the public entry for the API (/v1, /api, /tiles).
+    # Docker-published ports bypass host firewalls, so don't expose 9000 directly.
     ports:
-      - 9000:9000
+      - 127.0.0.1:9000:9000
     restart: unless-stopped
     # Wait for Postgres readiness, not just container start.
     depends_on:
@@ -72,8 +79,9 @@ services:
 
   frontend:
     image: ghcr.io/meshaddicts/meshinfo-spa:latest
+    # Localhost-only: caddy is the public entry for the SPA.
     ports:
-      - 8000:80
+      - 127.0.0.1:8000:80
     env_file:
       - path: ./frontend/.env
         required: false

storage/db/postgres.py

@@ -12,6 +12,7 @@
 import datetime
 import json
 import logging
+import math
 from collections import OrderedDict
 from pathlib import Path
 from typing import Any, Dict, Optional, List, Tuple
@@ -41,6 +42,24 @@ def _json_default(obj: Any) -> Any:
     return str(obj)
 
 
+def _finite_or_none(value: Any) -> Any:
+    """Map non-finite telemetry metrics to None for typed numeric columns.
+
+    Protobuf's proto3 JSON mapping serializes non-finite floats as the
+    strings "NaN"/"Infinity"/"-Infinity" (and some decode paths yield real
+    non-finite floats); asyncpg can bind neither into a numeric column, so
+    the whole node write fails. Everything else passes through unchanged."""
+    if isinstance(value, float) and not math.isfinite(value):
+        return None
+    if isinstance(value, str):
+        try:
+            if not math.isfinite(float(value)):
+                return None
+        except (TypeError, ValueError):
+            pass
+    return value
+
+
 def _encode_cursor(created_at: datetime.datetime, row_id: int) -> str:
     """Opaque keyset-pagination cursor for mqtt_messages ordered by (created_at, id) DESC."""
     raw = f"{created_at.isoformat()}|{row_id}".encode()
@@ -830,7 +849,7 @@ async def _write_node_telemetry_current(self, conn, node_id: str, telemetry: Dic
         present: Dict[str, Any] = {}
         for payload_key, col_name in self.TELEMETRY_COLUMNS.items():
             if payload_key in telemetry:
-                present[col_name] = telemetry[payload_key]
+                present[col_name] = _finite_or_none(telemetry[payload_key])
 
         if not present:
             logger.debug("_write_node_telemetry_current: no recognized telemetry fields for node %s; skipping", node_norm)

tests/test_postgres_helpers.py

@@ -12,6 +12,7 @@
     PostgresStorage,
     _decode_cursor,
     _encode_cursor,
+    _finite_or_none,
     _json_default,
     _month_partition_specs,
 )
@@ -52,6 +53,37 @@ def test_plain_dumps_still_raises_without_default(self):
             json.dumps({"ts": datetime.datetime.now()})
 
 
+class TestFiniteOrNone:
+    """Non-finite telemetry values must become NULL, not fail the whole
+    node write. The exact failure shape from production: proto3 JSON
+    renders a sensor's NaN float as the string "NaN", which asyncpg
+    rejects for numeric columns ('must be real number, not str')."""
+
+    def test_nan_string_becomes_none(self):
+        assert _finite_or_none("NaN") is None
+
+    def test_infinity_strings_become_none(self):
+        assert _finite_or_none("Infinity") is None
+        assert _finite_or_none("-Infinity") is None
+
+    def test_nan_float_becomes_none(self):
+        assert _finite_or_none(float("nan")) is None
+
+    def test_infinite_floats_become_none(self):
+        assert _finite_or_none(float("inf")) is None
+        assert _finite_or_none(float("-inf")) is None
+
+    def test_finite_values_pass_through(self):
+        assert _finite_or_none(3.7) == 3.7
+        assert _finite_or_none(87) == 87
+        assert _finite_or_none(0) == 0
+        assert _finite_or_none(None) is None
+
+    def test_non_numeric_values_pass_through(self):
+        # Not this helper's job to coerce or reject other bad input.
+        assert _finite_or_none("not-a-number") == "not-a-number"
+
+
 class TestCursorCodec:
     """Keyset-pagination cursor for mqtt_messages — see _encode_cursor/_decode_cursor."""