Feat : Notifications system, State log

vps-monitor/backend/main.py

@@ -64,6 +64,10 @@ def _get_expected_version() -> str:
 # ─── Ring buffer de stats (en mémoire) ───────────────────────────────────────
 _STATS_MAX_POINTS = 120   # 10 min à 5 s d'intervalle
 _stats_history: dict[str, deque] = {}
+
+# État courant des conteneurs : vps_id → {container_name: effective_state}
+_container_states: dict[str, dict[str, str]] = {}
+
 SECRET_FILE    = Path(os.getenv("SECRET_FILE", "data/.jwt_secret"))
 AGENT_TIMEOUT  = int(os.getenv("AGENT_TIMEOUT", "5"))
 JWT_ALGORITHM  = "HS256"
@@ -252,6 +256,31 @@ def init_db() -> None:
         conn.execute("""
             INSERT OR IGNORE INTO settings (key, value) VALUES ('passkey_enabled', 'true')
         """)
+        conn.execute("""
+            INSERT OR IGNORE INTO settings (key, value) VALUES ('pushover_enabled', 'false')
+        """)
+        conn.execute("""
+            INSERT OR IGNORE INTO settings (key, value) VALUES ('pushover_app_token', '')
+        """)
+        conn.execute("""
+            INSERT OR IGNORE INTO settings (key, value) VALUES ('pushover_user_key', '')
+        """)
+        conn.execute("""
+            CREATE TABLE IF NOT EXISTS container_events (
+                id         INTEGER PRIMARY KEY AUTOINCREMENT,
+                ts         INTEGER NOT NULL,
+                vps_id     TEXT    NOT NULL,
+                vps_name   TEXT    NOT NULL,
+                container  TEXT    NOT NULL,
+                image      TEXT    NOT NULL DEFAULT '',
+                old_state  TEXT    NOT NULL DEFAULT '',
+                new_state  TEXT    NOT NULL
+            )
+        """)
+        conn.execute("""
+            CREATE INDEX IF NOT EXISTS idx_container_events_ts
+            ON container_events(ts DESC)
+        """)
         conn.execute("""
             CREATE TABLE IF NOT EXISTS passkeys (
                 id            INTEGER PRIMARY KEY AUTOINCREMENT,
@@ -327,6 +356,32 @@ def _get_setting(key: str) -> str:
         return row["value"] if row else ""
 
 
+# ─── Pushover notifications ───────────────────────────────────────────────────
+
+def _effective_state(status: str, health: str) -> str:
+    if health and health not in ("none", ""):
+        return f"{status}:{health}"
+    return status
+
+
+async def _send_pushover(title: str, message: str) -> bool:
+    token = _get_setting("pushover_app_token")
+    user  = _get_setting("pushover_user_key")
+    if not token or not user:
+        return False
+    try:
+        timeout = aiohttp.ClientTimeout(total=10)
+        async with aiohttp.ClientSession() as session:
+            async with session.post(
+                "https://api.pushover.net/1/messages.json",
+                data={"token": token, "user": user, "title": title, "message": message},
+                timeout=timeout,
+            ) as r:
+                return r.status == 200
+    except Exception:
+        return False
+
+
 # ─── WebAuthn / Passkey helpers ───────────────────────────────────────────────
 
 def _b64url_encode(b: bytes) -> str:
@@ -626,6 +681,76 @@ async def _stats_collector() -> None:
         await asyncio.sleep(5)
 
 
+# ─── Collecteur d'états des conteneurs ───────────────────────────────────────
+
+async def _check_container_states(vps: dict) -> None:
+    try:
+        containers = await agent_get(vps, "/containers")
+    except Exception:
+        return
+
+    vps_id   = vps["id"]
+    vps_name = vps["name"]
+    current  = {
+        c["name"]: _effective_state(c["status"], c.get("health", "none"))
+        for c in containers
+    }
+
+    prev = _container_states.get(vps_id)
+    if prev is None:
+        _container_states[vps_id] = current
+        return
+
+    now    = int(time.time())
+    notify = _get_setting("pushover_enabled") == "true"
+    events: list[dict] = []
+
+    for name, state in current.items():
+        prev_state = prev.get(name, "")
+        if state != prev_state:
+            image = next((c["image"] for c in containers if c["name"] == name), "")
+            events.append({
+                "ts": now, "vps_id": vps_id, "vps_name": vps_name,
+                "container": name, "image": image,
+                "old_state": prev_state, "new_state": state,
+            })
+
+    for name, state in prev.items():
+        if name not in current:
+            events.append({
+                "ts": now, "vps_id": vps_id, "vps_name": vps_name,
+                "container": name, "image": "",
+                "old_state": state, "new_state": "removed",
+            })
+
+    if events:
+        with get_db() as conn:
+            conn.executemany("""
+                INSERT INTO container_events
+                    (ts, vps_id, vps_name, container, image, old_state, new_state)
+                VALUES (:ts, :vps_id, :vps_name, :container, :image, :old_state, :new_state)
+            """, events)
+        if notify:
+            for ev in events:
+                title = f"[{ev['vps_name']}] {ev['container']}"
+                msg   = f"Etat : {ev['old_state'] or 'nouveau'} -> {ev['new_state']}"
+                asyncio.create_task(_send_pushover(title, msg))
+
+    _container_states[vps_id] = current
+
+
+async def _container_state_collector() -> None:
+    """Tache de fond : verifie les etats des conteneurs toutes les 15 secondes."""
+    await asyncio.sleep(5)
+    while True:
+        vps_list = load_vps()
+        await asyncio.gather(
+            *[_check_container_states(v) for v in vps_list],
+            return_exceptions=True,
+        )
+        await asyncio.sleep(15)
+
+
 async def _refresh_latest_agent_version() -> None:
     """Récupère AGENT_VERSION depuis le dépôt Gitea toutes les heures.
     Si EXPECTED_AGENT_VERSION est défini en env var, cette tâche n'écrase pas la valeur forcée.
@@ -655,17 +780,19 @@ async def _refresh_latest_agent_version() -> None:
 @app.on_event("startup")
 async def startup_event() -> None:
     asyncio.create_task(_stats_collector())
+    asyncio.create_task(_container_state_collector())
     asyncio.create_task(_cleanup_old_stats())
     asyncio.create_task(_refresh_latest_agent_version())
 
 
 async def _cleanup_old_stats() -> None:
-    """Supprime les statistiques de plus de 31 jours (s'exécute toutes les heures)."""
+    """Supprime les statistiques et evenements de plus de 31 jours (toutes les heures)."""
     while True:
         await asyncio.sleep(3600)
         cutoff = int(time.time()) - 31 * 24 * 3600
         with get_db() as conn:
             conn.execute("DELETE FROM vps_stats WHERE ts < ?", (cutoff,))
+            conn.execute("DELETE FROM container_events WHERE ts < ?", (cutoff,))
 
 
 # ─── Routes Auth ──────────────────────────────────────────────────────────────
@@ -760,7 +887,7 @@ def admin_update_setting(
     _: Annotated[dict, Depends(require_admin)],
 ):
     """Met à jour un paramètre d'administration."""
-    allowed_keys = {"registration_open", "passkey_enabled"}
+    allowed_keys = {"registration_open", "passkey_enabled", "pushover_enabled", "pushover_app_token", "pushover_user_key"}
     if key not in allowed_keys:
         raise HTTPException(status_code=400, detail="Clé de paramètre inconnue")
     with get_db() as conn:
@@ -825,12 +952,13 @@ def admin_db_info(_: Annotated[dict, Depends(require_admin)]):
                 "newest_ts": row["newest"],
             }
         return {
-            "vps_stats":  _table_info("vps_stats"),
-            "login_logs": _table_info("login_logs"),
+            "vps_stats":        _table_info("vps_stats"),
+            "login_logs":       _table_info("login_logs"),
+            "container_events": _table_info("container_events"),
         }
 
 
-_ALLOWED_TABLES = frozenset({"vps_stats", "login_logs"})
+_ALLOWED_TABLES  = frozenset({"vps_stats", "login_logs", "container_events"})
 _ALLOWED_PERIODS = frozenset({"last_24h", "last_7d", "last_30d", "all", "custom"})
 
 
@@ -879,6 +1007,52 @@ def admin_db_purge(body: PurgeRequest, _: Annotated[dict, Depends(require_admin)
     return {"deleted": deleted, "status": "ok"}
 
 
+@app.get("/api/admin/container-events")
+def admin_container_events(
+    limit:  int = 100,
+    offset: int = 0,
+    _: Annotated[dict, Depends(require_admin)] = None,
+):
+    """Retourne l'historique des changements d'etat des conteneurs."""
+    limit  = max(1, min(limit, 500))
+    offset = max(0, offset)
+    with get_db() as conn:
+        rows = conn.execute(
+            "SELECT * FROM container_events ORDER BY ts DESC LIMIT ? OFFSET ?",
+            (limit, offset),
+        ).fetchall()
+        total = conn.execute("SELECT COUNT(*) FROM container_events").fetchone()[0]
+    return {
+        "total": total,
+        "events": [
+            {
+                "id":        row["id"],
+                "ts":        datetime.fromtimestamp(row["ts"], tz=timezone.utc).isoformat(),
+                "vps_name":  row["vps_name"],
+                "container": row["container"],
+                "image":     row["image"],
+                "old_state": row["old_state"],
+                "new_state": row["new_state"],
+            }
+            for row in rows
+        ],
+    }
+
+
+@app.post("/api/admin/notifications/test")
+async def admin_test_notification(_: Annotated[dict, Depends(require_admin)]):
+    """Envoie une notification Pushover de test."""
+    if _get_setting("pushover_enabled") != "true":
+        raise HTTPException(status_code=400, detail="Les notifications Pushover sont desactivees")
+    ok = await _send_pushover("VPS Monitor - Test", "La notification de test a bien ete recue.")
+    if not ok:
+        raise HTTPException(
+            status_code=502,
+            detail="Echec de l'envoi - verifiez le token et la cle utilisateur",
+        )
+    return {"status": "ok"}
+
+
 # ─── Routes Passkeys ──────────────────────────────────────────────────────────
 
 @app.post("/api/auth/passkey/register/begin")