fix: track alert backend errors independently

Per-backend error counts with exponential backoff: after 5 consecutive
failures a backend is skipped every 2^(n-5) cycles (capped at 32).
Working backends are no longer penalized by one flaky backend doubling
the entire poll interval.

Migrates last_error (string) to last_errors (dict per backend).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

plugins/alert.py

@@ -80,7 +80,8 @@ _HUGGINGFACE_API = "https://huggingface.co/api/models"
 
 _pollers: dict[str, asyncio.Task] = {}
 _subscriptions: dict[str, dict] = {}
-_errors: dict[str, int] = {}
+_errors: dict[str, dict[str, int]] = {}
+_poll_count: dict[str, int] = {}
 
 # -- History database --------------------------------------------------------
 
@@ -1711,17 +1712,27 @@ async def _poll_once(bot, key: str, announce: bool = True) -> None:
     now = datetime.now(timezone.utc).isoformat()
     data["last_poll"] = now
 
-    had_error = False
+    cycle = _poll_count[key] = _poll_count.get(key, 0) + 1
+    tag_errors = _errors.setdefault(key, {})
     loop = asyncio.get_running_loop()
 
     for tag, backend in _BACKENDS.items():
+        errs = tag_errors.get(tag, 0)
+        if errs >= 5:
+            skip = min(2 ** (errs - 5), 32)
+            if cycle % skip != 0:
+                continue
+
         try:
             items = await loop.run_in_executor(None, backend, keyword)
         except Exception as exc:
-            data["last_error"] = f"{tag}: {exc}"
-            had_error = True
+            tag_errors[tag] = errs + 1
+            data.setdefault("last_errors", {})[tag] = str(exc)
             continue
 
+        tag_errors[tag] = 0
+        data.setdefault("last_errors", {}).pop(tag, None)
+
         seen_set = set(data.get("seen", {}).get(tag, []))
         seen_list = list(data.get("seen", {}).get(tag, []))
         new_items = [item for item in items if item["id"] not in seen_set]
@@ -1795,12 +1806,6 @@ async def _poll_once(bot, key: str, announce: bool = True) -> None:
             seen_list = seen_list[-_MAX_SEEN:]
         data.setdefault("seen", {})[tag] = seen_list
 
-    if had_error:
-        _errors[key] = _errors.get(key, 0) + 1
-    else:
-        data["last_error"] = ""
-        _errors[key] = 0
-
     _subscriptions[key] = data
     _save(bot, key, data)
 
@@ -1813,9 +1818,6 @@ async def _poll_loop(bot, key: str) -> None:
             if data is None:
                 return
             interval = data.get("interval", _DEFAULT_INTERVAL)
-            errs = _errors.get(key, 0)
-            if errs >= 5:
-                interval = min(interval * 2, _MAX_INTERVAL)
             await asyncio.sleep(interval)
             await _poll_once(bot, key, announce=True)
     except asyncio.CancelledError:
@@ -1837,7 +1839,8 @@ def _stop_poller(key: str) -> None:
     if task and not task.done():
         task.cancel()
     _subscriptions.pop(key, None)
-    _errors.pop(key, 0)
+    _errors.pop(key, None)
+    _poll_count.pop(key, None)
 
 
 # -- Restore on connect -----------------------------------------------------
@@ -1895,9 +1898,9 @@ async def cmd_alert(bot, message):
                 data = _load(bot, key)
                 if data:
                     name = data["name"]
-                    err = data.get("last_error", "")
-                    if err:
-                        subs.append(f"{name} (error)")
+                    errs = data.get("last_errors", {})
+                    if errs:
+                        subs.append(f"{name} ({len(errs)} backend errors)")
                     else:
                         subs.append(name)
         if not subs:
@@ -1924,8 +1927,10 @@ async def cmd_alert(bot, message):
         _subscriptions[key] = data
         await _poll_once(bot, key, announce=True)
         data = _subscriptions.get(key, data)
-        if data.get("last_error"):
-            await bot.reply(message, f"{name}: error -- {data['last_error']}")
+        errs = data.get("last_errors", {})
+        if errs:
+            tags = ", ".join(sorted(errs))
+            await bot.reply(message, f"{name}: errors on {tags}")
         else:
             await bot.reply(message, f"{name}: checked")
         return
@@ -2063,7 +2068,7 @@ async def cmd_alert(bot, message):
             "added_by": message.nick,
             "added_at": now,
             "last_poll": now,
-            "last_error": "",
+            "last_errors": {},
             "seen": {},
         }
         _save(bot, key, data)