feat(syslog): 优化大日志通道并增加异常连续消费兜底

- 128 KiB 及以上日志默认进入大日志 Topic - InfluxDB 原文及达梦大字段超过 256 KiB 时截断保存 - 截断内容追加 HB_XDR_TRUNCATED 标记 - InfluxDB 写入失败时继续执行规则解析和达梦写入 - 未知或系统性异常优先写入普通/大日志 DLT - DLT 写入失败时记录特殊告警并提交 offset，避免阻塞后续消息 - DLT 失败等待时间默认限制为 1 秒 - 增加连续性放行、DLT 成功及失败监控指标 - 更新 Docker Compose、v3 升级文档及相关测试
2026-09-12 19:22:52 +08:00
parent 3b6c9e4714
commit 0e9e6656db
64 changed files with 3189 additions and 63 deletions
@@ -0,0 +1,96 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# v3 统计投影 Topic。新建时使用 1 个分区；如果现场已经建成多个分区则保留现状，
+# 应用使用固定 Kafka key，投影消息仍只会落入其中一个稳定分区。
+BOOTSTRAP="${BOOTSTRAP:-10.141.121.35:9092,10.141.121.50:9092,10.141.121.59:9092}"
+RAW_TOPIC="${RAW_TOPIC:-agent-01-syslog-topic}"
+PROJECTION_TOPIC="${PROJECTION_TOPIC:-${RAW_TOPIC}.STATS.PROJECTION}"
+RETENTION_MS="${RETENTION_MS:-604800000}"
+
+if [[ "${PROJECTION_TOPIC}" == "${RAW_TOPIC}.STATS.MATERIALIZE" ]]; then
+  echo "ERROR: STATS.PROJECTION 不能与 STATS.MATERIALIZE 使用同一个 Topic" >&2
+  exit 1
+fi
+
+if [[ -z "${KAFKA_BIN:-}" ]]; then
+  for candidate in \
+    "$(pwd)" \
+    "$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)" \
+    "/opt/kafka/kafka_2.13-3.6.1/bin" \
+    "/opt/kafka/bin" \
+    "/usr/local/kafka/bin"; do
+    if [[ -x "${candidate}/kafka-topics.sh" ]] \
+        && [[ -x "${candidate}/kafka-configs.sh" ]]; then
+      KAFKA_BIN="${candidate}"
+      break
+    fi
+  done
+fi
+
+if [[ -z "${KAFKA_BIN:-}" ]] && command -v kafka-topics.sh >/dev/null 2>&1; then
+  KAFKA_BIN=$(dirname -- "$(command -v kafka-topics.sh)")
+fi
+if [[ -z "${KAFKA_BIN:-}" ]]; then
+  echo "ERROR: 未找到 kafka-topics.sh；请在 Kafka bin 目录执行，或设置 KAFKA_BIN" >&2
+  exit 1
+fi
+
+TOPICS="${KAFKA_BIN}/kafka-topics.sh"
+CONFIGS="${KAFKA_BIN}/kafka-configs.sh"
+COMMON=(--bootstrap-server "${BOOTSTRAP}")
+if [[ -n "${COMMAND_CONFIG:-}" ]]; then
+  [[ -r "${COMMAND_CONFIG}" ]] || { echo "ERROR: 无法读取 ${COMMAND_CONFIG}" >&2; exit 1; }
+  COMMON+=(--command-config "${COMMAND_CONFIG}")
+fi
+
+echo "Kafka=${BOOTSTRAP}"
+echo "原始 Topic=${RAW_TOPIC}"
+echo "统计投影 Topic=${PROJECTION_TOPIC}"
+
+RAW_INFO=$("${TOPICS}" "${COMMON[@]}" --describe --topic "${RAW_TOPIC}")
+REPLICAS=$(printf '%s\n' "${RAW_INFO}" | awk '
+  /ReplicationFactor:/ {for(i=1;i<=NF;i++) if($i=="ReplicationFactor:"){print $(i+1);exit}}
+')
+if [[ ! "${REPLICAS}" =~ ^[1-9][0-9]*$ ]]; then
+  echo "ERROR: 无法从 ${RAW_TOPIC} 读取副本数" >&2
+  exit 1
+fi
+
+echo "[1/3] 创建 Topic（已经存在时不重复创建）"
+"${TOPICS}" "${COMMON[@]}" --create --if-not-exists \
+  --topic "${PROJECTION_TOPIC}" \
+  --partitions 1 \
+  --replication-factor "${REPLICAS}" \
+  --config cleanup.policy=delete \
+  --config "retention.ms=${RETENTION_MS}"
+
+echo "[2/3] 更新保留策略"
+if ! "${CONFIGS}" "${COMMON[@]}" --alter \
+  --entity-type topics --entity-name "${PROJECTION_TOPIC}" \
+  --add-config "cleanup.policy=delete,retention.ms=${RETENTION_MS}"; then
+  echo "WARN: Kafka 配置更新超时；新建命令已携带相同策略，继续验证 Topic" >&2
+fi
+
+echo "[3/3] 验证 Topic"
+INFO=$("${TOPICS}" "${COMMON[@]}" --describe --topic "${PROJECTION_TOPIC}")
+printf '%s\n' "${INFO}"
+PARTITIONS=$(printf '%s\n' "${INFO}" | awk '
+  /PartitionCount:/ {for(i=1;i<=NF;i++) if($i=="PartitionCount:"){print $(i+1);exit}}
+')
+ACTUAL_REPLICAS=$(printf '%s\n' "${INFO}" | awk '
+  /ReplicationFactor:/ {for(i=1;i<=NF;i++) if($i=="ReplicationFactor:"){print $(i+1);exit}}
+')
+if [[ ! "${PARTITIONS}" =~ ^[1-9][0-9]*$ ]]; then
+  echo "ERROR: ${PROJECTION_TOPIC} 没有有效分区" >&2
+  exit 1
+fi
+if [[ "${PARTITIONS}" != "1" ]]; then
+  echo "WARN: Topic 已有 ${PARTITIONS} 个分区；不会修改分区数，固定 key 仍只使用其中一个分区" >&2
+fi
+if [[ "${ACTUAL_REPLICAS}" != "${REPLICAS}" ]]; then
+  echo "WARN: Topic 副本数=${ACTUAL_REPLICAS}，原始 Topic 副本数=${REPLICAS}" >&2
+fi
+
+echo "PASS: ${PROJECTION_TOPIC} 已创建/确认。分区数=${PARTITIONS}，保留时间=${RETENTION_MS}ms。"
+echo "脚本未删除消息，也未修改任何消费组 offset。"