fix: usage collection error no longer kills keys; 401/403 treated as key failure; key3 (renewal cancelled) skipped via accounts.json

gateway/scripts/ocg_router.py

@@ -95,6 +95,8 @@ def load_accounts():
                 "label": a.get("label", kid),
                 "api_key": a.get("key", ""),
                 "workspace_id": a.get("workspace_id", ""),
+                # 续订取消/显式标记 → 视为不可用（采集失败时也能正确跳过）
+                "subscribed": a.get("subscribed", a.get("renewal") != "cancelled"),
             })
         log.info("loaded %d OCG keys from accounts.json", len(keys))
     except Exception as e:
@@ -137,6 +139,12 @@ def refresh_state():
     global _keys, _key_usage
     _keys = load_accounts()
     _key_usage = load_usage()
+    # 合并 accounts.json 的订阅标记：采集失败时 usage 缺失该信息，
+    # 以 accounts.json 的 renewal/subscribed 为准
+    for k in _keys:
+        kid = k["key_id"]
+        if not k.get("subscribed", True):
+            _key_usage.setdefault(kid, {})["subscribed"] = False
     log.info("state refreshed: %d keys, %d with usage data", len(_keys), len(_key_usage))
 
 
@@ -147,16 +155,21 @@ def refresh_state():
 def _key_health_score(kid):
     """
     计算 key 健康度（越低越好）：
-      - 有 error 或 unsubscribe 的 key 排最末（score = 999）
+      - 已取消订阅的 key 排最末（score = 999）
       - 在冷却期内的 key 排倒数第二（score = 888）
+      - 用量采集失败（error）仅降权（score = 500），key 仍可用
       - 否则按 rolling% → weekly% → monthly% 加权
     """
     usage = _key_usage.get(kid, {})
 
-    # 订阅取消或采集错误 → 不可用
-    if not usage.get("subscribed", True) or usage.get("error"):
+    # 订阅取消 → 不可用
+    if not usage.get("subscribed", True):
         return 999
 
+    # 用量采集失败 → 降权但可用（不因采集问题误杀 key）
+    if usage.get("error"):
+        return 500
+
     # 故障冷却期内 → 低优先级
     now = time.time()
     fail_until = _key_failures.get(kid, 0)
@@ -264,7 +277,7 @@ class RouterHandler(BaseHTTPRequestHandler):
                     "rolling_pct": usage.get("rolling_pct", None),
                     "weekly_pct": usage.get("weekly_pct", None),
                     "monthly_pct": usage.get("monthly_pct", None),
-                    "healthy": usage.get("subscribed", True) and not usage.get("error") and (time.time() >= fail_until),
+                    "healthy": usage.get("subscribed", True) and (time.time() >= fail_until),
                     "in_cooldown": time.time() < fail_until,
                     "error": usage.get("error", False),
                     "hits": _route_stats.get(kid, 0),
@@ -295,9 +308,9 @@ class RouterHandler(BaseHTTPRequestHandler):
             api_key = k.get("api_key", "")
             if not api_key:
                 continue
-            # 跳过不可用 key
+            # 跳过不可用 key（采集 error 不跳过，仅降权）
             usage = _key_usage.get(kid, {})
-            if not usage.get("subscribed", True) or usage.get("error"):
+            if not usage.get("subscribed", True):
                 continue
             if time.time() < _key_failures.get(kid, 0):
                 continue
@@ -370,9 +383,9 @@ class RouterHandler(BaseHTTPRequestHandler):
             if not api_key:
                 continue
 
-            # 检查 key 是否可用
+            # 检查 key 是否可用（采集 error 不跳过，仅降权）
             usage = usage_snapshot.get(kid, {})
-            if not usage.get("subscribed", True) or usage.get("error"):
+            if not usage.get("subscribed", True):
                 continue
             if time.time() < failures_snapshot.get(kid, 0):
                 continue
@@ -390,9 +403,9 @@ class RouterHandler(BaseHTTPRequestHandler):
                 else:
                     last_error = result.get("error", "unknown")
                     log.warning("⚠ %s via %s FAILED: %s", model_name, kid, last_error)
-                    # 402/429/5xx → 标记 key 故障
+                    # 401(未授权/CreditsError)/402/403/429/5xx → 标记 key 故障
                     status = result.get("status", 0)
-                    if status in (402, 429) or status >= 500:
+                    if status in (401, 402, 403, 429) or status >= 500:
                         mark_key_failed(kid)
             except Exception as e:
                 last_error = str(e)

gateway/scripts/usage_collector.py

@@ -77,7 +77,7 @@ DEFAULT_POLL_INTERVAL_SEC = 300   # 5 minutes
 CDP_PROXY_URL = os.environ.get('CDP_PROXY_URL', 'http://localhost:3456')
 CDP_EVAL_TIMEOUT = 20
 CDP_TARGETS_TIMEOUT = 5
-HTTP_TIMEOUT = 15
+HTTP_TIMEOUT = 45  # opencode.ai 响应慢（2026-08 实测 10-12s），放宽到 45s
 OPENCODE_BASE_URL = 'https://opencode.ai'
 
 # ── SSR HTML parsing ──