🚧 Add broker status page

backend/app/metrics.py

@@ -0,0 +1,123 @@
+"""$SYS broker telemetry (spec sections 33, 55).
+
+The dynsec channel's MQTT connection subscribes to $SYS/# and forwards
+every message here (see DynsecClient.set_event_sink). Values are parsed
+ONCE at ingest; snapshot() returns the latest known state. The broker
+publishes $SYS on the configured sys_interval (10 s in the bootstrap
+config), so a fresh backend needs a few seconds before the first
+snapshot is populated — every field is therefore Optional and the
+`available` flag says whether ANY $SYS message has been seen.
+"""
+
+from __future__ import annotations
+
+import threading
+from datetime import UTC, datetime
+from typing import Any, Callable
+
+SYS_PREFIX = "$SYS/broker/"
+
+_lock = threading.Lock()
+_values: dict[str, Any] = {}
+_last_update: datetime | None = None
+
+
+def _as_int(payload: str) -> object:
+    try:
+        return int(payload)
+    except ValueError:
+        return payload
+
+
+def _as_float(payload: str) -> object:
+    try:
+        return float(payload)
+    except ValueError:
+        return payload
+
+
+# $SYS topic -> (snapshot field, parser). Field names mirror
+# BrokerStatusOut (schemas.py). NOTE: "retained messages/count"
+# genuinely contains a space in the Mosquitto topic name.
+_TOPIC_FIELDS: dict[str, tuple[str, Callable[[str], object]]] = {
+    f"{SYS_PREFIX}version": ("version", str),
+    f"{SYS_PREFIX}uptime": ("uptime", str),
+    f"{SYS_PREFIX}timestamp": ("build_timestamp", str),
+    # clients/disconnected = persistent clients currently offline,
+    # i.e. the "stored sessions" of the dashboard (spec section 33).
+    f"{SYS_PREFIX}clients/connected": ("clients_connected", _as_int),
+    f"{SYS_PREFIX}clients/disconnected": ("clients_disconnected", _as_int),
+    f"{SYS_PREFIX}clients/maximum": ("clients_maximum", _as_int),
+    f"{SYS_PREFIX}clients/total": ("clients_total", _as_int),
+    f"{SYS_PREFIX}clients/expired": ("clients_expired", _as_int),
+    f"{SYS_PREFIX}messages/received": ("messages_received", _as_int),
+    f"{SYS_PREFIX}messages/sent": ("messages_sent", _as_int),
+    f"{SYS_PREFIX}messages/stored": ("messages_stored", _as_int),
+    f"{SYS_PREFIX}messages/dropped": ("messages_dropped", _as_int),
+    f"{SYS_PREFIX}publish/messages/received": ("publish_received", _as_int),
+    f"{SYS_PREFIX}publish/messages/sent": ("publish_sent", _as_int),
+    f"{SYS_PREFIX}publish/messages/dropped": ("publish_dropped", _as_int),
+    f"{SYS_PREFIX}bytes/received": ("bytes_received", _as_int),
+    f"{SYS_PREFIX}bytes/sent": ("bytes_sent", _as_int),
+    f"{SYS_PREFIX}subscriptions/count": ("subscriptions_count", _as_int),
+    f"{SYS_PREFIX}retained messages/count": ("retained_count", _as_int),
+    # load/* are per-second averages over the last minute ("x.xx").
+    f"{SYS_PREFIX}load/messages/received/1min": (
+        "load_messages_received_1min",
+        _as_float,
+    ),
+    f"{SYS_PREFIX}load/messages/sent/1min": (
+        "load_messages_sent_1min",
+        _as_float,
+    ),
+    f"{SYS_PREFIX}load/publish/received/1min": (
+        "load_publish_received_1min",
+        _as_float,
+    ),
+    f"{SYS_PREFIX}load/publish/sent/1min": (
+        "load_publish_sent_1min",
+        _as_float,
+    ),
+    f"{SYS_PREFIX}load/bytes/received/1min": (
+        "load_bytes_received_1min",
+        _as_float,
+    ),
+    f"{SYS_PREFIX}load/bytes/sent/1min": (
+        "load_bytes_sent_1min",
+        _as_float,
+    ),
+}
+
+_FIELDS = [field for field, _ in _TOPIC_FIELDS.values()]
+
+
+def handle_sys_message(topic: str, payload: bytes) -> None:
+    """Ingest one $SYS message (wired as the dynsec client's event sink)."""
+    entry = _TOPIC_FIELDS.get(topic)
+    if entry is None:
+        return  # unknown $SYS topic — deliberately ignored
+    field, parser = entry
+    text = payload.decode("utf-8", errors="replace").strip()
+    global _last_update
+    with _lock:
+        _values[field] = parser(text)
+        _last_update = datetime.now(UTC)
+
+
+def reset() -> None:
+    """Clear all telemetry (used by tests)."""
+    global _last_update
+    with _lock:
+        _values.clear()
+        _last_update = None
+
+
+def snapshot() -> dict[str, Any]:
+    """Latest known $SYS state; every field None until first publish."""
+    with _lock:
+        values = dict(_values)
+        last_update = _last_update
+    snap: dict[str, Any] = {field: values.get(field) for field in _FIELDS}
+    snap["available"] = last_update is not None
+    snap["last_update"] = last_update
+    return snap