auto-sync: 2026-06-02 22:04:41

scripts/gateway-watchdog.sh

@@ -139,6 +139,17 @@ with open('${LOG_FILE}') as f:
 " 2>/dev/null
 }
 
+# 辅助函数：安全计数（从管道输入 grep 后取整数值）
+_safe_count() {
+    # 接受 grep pattern，对 stdin 计数，返回纯整数
+    local count
+    count=$(grep -c "$1" 2>/dev/null || true)
+    # 去掉所有空白和换行，确保是纯数字
+    count=$(echo "$count" | tr -d '[:space:]')
+    [ -z "$count" ] && count=0
+    echo "$count"
+}
+
 # 统计各规则命中次数
 # 用全局变量返回：R1_COUNT, R2_COUNT, R3_COUNT
 count_rules() {
@@ -152,20 +163,14 @@ count_rules() {
         return
     fi
 
-    # R1: 含 "lane task error" 且含 "FailoverError"
-    R1_COUNT=$(echo "$recent_lines" | grep -c "lane task error" 2>/dev/null | tr -d ' ' || echo 0)
-    if [ "$R1_COUNT" -gt 0 ]; then
-        R1_COUNT=$(echo "$recent_lines" | grep "lane task error" | grep -c "FailoverError" 2>/dev/null | tr -d ' ' || echo 0)
-    fi
+    # R1: 含 "lane task error" 且含 "FailoverError"（两步过滤）
+    R1_COUNT=$(echo "$recent_lines" | grep "lane task error" | _safe_count "FailoverError")
 
-    # R2: 含 "stalled session" 且含 "recovery=none"
-    R2_COUNT=$(echo "$recent_lines" | grep -c "stalled session" 2>/dev/null | tr -d ' ' || echo 0)
-    if [ "$R2_COUNT" -gt 0 ]; then
-        R2_COUNT=$(echo "$recent_lines" | grep "stalled session" | grep -c "recovery=none" 2>/dev/null | tr -d ' ' || echo 0)
-    fi
+    # R2: 含 "stalled session" 且含 "recovery=none"（两步过滤）
+    R2_COUNT=$(echo "$recent_lines" | grep "stalled session" | _safe_count "recovery=none")
 
-    # R3: 含 "rate_limit" 或含 "429"
-    R3_COUNT=$(echo "$recent_lines" | grep -c "rate_limit\|\"429\"" 2>/dev/null | tr -d ' ' || echo 0)
+    # R3: 含 "rate_limit" 或含 "429"（匹配 JSON 中 rate_limit 关键字或 429 错误码）
+    R3_COUNT=$(echo "$recent_lines" | _safe_count 'rate_limit|429')
 }
 
 # === 主逻辑 ===