mtr监控策略初版

2025-11-19 18:42:49 +08:00
parent ae084bfa52
commit 7e69fddafd
11 changed files with 415 additions and 7 deletions
@@ -9,9 +9,9 @@
        <relativePath/> <!-- lookup parent from repository -->
    </parent>
    <groupId>com.tongran.agent</groupId>
-    <artifactId>tr-agent-client</artifactId>
+    <artifactId>tr-mtragent-client</artifactId>
    <version>0.0.1-SNAPSHOT</version>
-    <name>tr-agent-client</name>
+    <name>tr-mtragent-client</name>
    <description>tr-agent-client</description>

    <properties>
@@ -56,6 +56,7 @@ public class GlobalConfig {
    public static long SCRIPT_TIME = 0L;
    public static long VERSION_TIME = 0L;
    public static long ROUTE_TIME = 0L;
+    public static String MTRPOLICYMSG = "";
    /**
     * 采集标识
     */
@@ -67,6 +67,8 @@ public enum MsgEnum {

    Agent版本更新应答("AGENT_VERSION_UPDATE_RSP"),

+    MTR探测上报("MTR_DETECT"),
+
    多网IP探测上报("NETWORK_DETECT");

    private String value;
@@ -0,0 +1,45 @@
+package com.tongran.agent.client.core.eo;
+
+import com.fasterxml.jackson.annotation.JsonFormat;
+import lombok.Data;
+
+import java.util.Date;
+
+/**
+ * mtr策略信息
+ */
+@Data
+public class MtrPolicyConfigEO {
+    private static final long serialVersionUID = 1L;
+
+    /** 主键ID */
+    private Long id;
+
+    /** 策略名称 */
+    private String policyName;
+
+    /** 优先级 */
+    private Long priority;
+
+    /** MTR客户端ID */
+    private String mtrClientId;
+
+    /** 服务器集合(换行符分割) */
+    private String serverGroup;
+    /** 探测目标ip集合 */
+    private String serveripGroup;
+
+    /** 是否探测(0-否,1-是) */
+    private Long probeFlag;
+
+    /** 开始时间 */
+    @JsonFormat(pattern = "yyyy-MM-dd HH:mm:ss")
+    private Date startTime;
+
+    /** 结束时间 */
+    @JsonFormat(pattern = "yyyy-MM-dd HH:mm:ss")
+    private Date endTime;
+
+    /** 探测频率(秒) */
+    private Long probeFrequency;
+}
@@ -0,0 +1,16 @@
+package com.tongran.agent.client.core.vo;
+
+import lombok.Data;
+
+import java.util.Date;
+
+@Data
+public class MtrVO {
+    private String targetIp;        // 目标IP
+    private Long policyId;          // 策略ID
+    private double firstLossPercent; // 第一次探测丢包率
+    private double finalLossPercent; // 最终丢包率
+    private long timestamp;         // 探测时间戳
+    private boolean hasRetry;       // 是否重试
+    private String errorMsg;        // 错误信息
+}
@@ -7,6 +7,7 @@ import com.alibaba.fastjson2.JSONObject;
 import com.tongran.agent.client.core.config.ApplicationProperties;
 import com.tongran.agent.client.core.config.GlobalConfig;
 import com.tongran.agent.client.core.enums.MsgEnum;
+import com.tongran.agent.client.core.eo.MtrPolicyConfigEO;
 import com.tongran.agent.client.core.session.SessionManager;
 import com.tongran.agent.client.core.vo.*;
 import com.tongran.agent.client.netty.MultiTargetNettyClient;
@@ -59,6 +60,7 @@ public class BusinessTasks {
    private final AtomicInteger registerTask = new AtomicInteger(0);
    private final AtomicInteger connectionTask = new AtomicInteger(0);
    private final AtomicInteger networkDetectTask = new AtomicInteger(0);
+    private final AtomicInteger mtrTask = new AtomicInteger(0);



@@ -108,6 +110,9 @@ public class BusinessTasks {
    @Resource
    private AgentService agentService;

+    @Resource
+    private MtrService mtrService;
+
    /**
     * 任务1：心跳上报任务
     */
@@ -900,6 +905,55 @@ public class BusinessTasks {
        AssertLog.info("发送多网IP探测定时任务执行 - task #{} completed", count);
    }

+    /**
+     * 任务30：MTR网络探测任务
+     */
+    @Async("taskExecutor")
+    public void mtrTask() {
+        long timestamp = AgentUtil.roundMinutes(); // 使用5分钟整点时间
+        int count = mtrTask.incrementAndGet();
+        AssertLog.info("MTR网络探测定时任务执行 - 时间: {}，task #{}", LocalDateTime.now(), count);
+
+        try {
+            // 检查当前是否有有效的MTR策略
+            if (StringUtils.isBlank(GlobalConfig.MTRPOLICYMSG)) {
+                AssertLog.info("没有MTR策略配置，跳过本次执行");
+                return;
+            }
+
+            // 解析策略
+            List<MtrPolicyConfigEO> mtrPolicyConfigEOList = JSON.parseArray(GlobalConfig.MTRPOLICYMSG, MtrPolicyConfigEO.class);
+            if (CollectionUtil.isEmpty(mtrPolicyConfigEOList)) {
+                AssertLog.info("MTR策略列表为空，跳过本次执行");
+                return;
+            }
+
+            // 判定客户端与服务端是否连接
+            if (Objects.nonNull(sessionManager.getSessionById(GlobalConfig.CLIENT_ID))) {
+                // 执行MTR探测并获取结果
+                String data = "";
+                List<MtrVO> list = mtrService.mtrList(timestamp, mtrPolicyConfigEOList);
+                if (CollectionUtil.isNotEmpty(list)) {
+                    data = JSONArray.toJSONString(list);
+
+                    // 发送MTR探测结果
+                    Message message = Message.builder()
+                            .clientId(GlobalConfig.CLIENT_ID)
+                            .dataType(MsgEnum.MTR探测上报.getValue())
+                            .data(data)
+                            .build();
+                    sessionManager.writeAndFlush(sessionManager.getSessionById(GlobalConfig.CLIENT_ID).getChannel(), message);
+                    AssertLog.info("发送MTR探测信息包，结果数量: {}", list.size());
+                } else {
+                    AssertLog.info("本次MTR探测无有效结果");
+                }
+            }
+        } catch (Exception e) {
+            AssertLog.error("MTR定时任务执行异常", e);
+        }
+
+        AssertLog.info("MTR网络探测定时任务执行 - task #{} completed", count);
+    }



@@ -0,0 +1,10 @@
+package com.tongran.agent.client.service;
+
+import com.tongran.agent.client.core.eo.MtrPolicyConfigEO;
+import com.tongran.agent.client.core.vo.MtrVO;
+
+import java.util.List;
+
+public interface MtrService {
+    List<MtrVO> mtrList(long timestamp, List<MtrPolicyConfigEO> mtrPolicyConfigEOList);
+}
@@ -9,8 +9,10 @@ import com.tongran.agent.client.core.config.GlobalConfig;
 import com.tongran.agent.client.core.enums.MsgEnum;
 import com.tongran.agent.client.core.eo.AgentVersionUpdateEO;
 import com.tongran.agent.client.core.eo.CollectEO;
+import com.tongran.agent.client.core.eo.MtrPolicyConfigEO;
 import com.tongran.agent.client.core.eo.ScriptPolicyEO;
 import com.tongran.agent.client.core.session.SessionManager;
+import com.tongran.agent.client.core.vo.MtrVO;
 import com.tongran.agent.client.netty.MultiTargetNettyClient;
 import com.tongran.agent.client.netty.config.AgentNettyConfig;
 import com.tongran.agent.client.netty.model.Message;
@@ -35,6 +37,9 @@ import java.time.LocalDateTime;
 import java.util.*;
 import java.util.concurrent.CompletableFuture;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.TimeoutException;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
 import java.util.stream.Collectors;

@Service
@@ -825,11 +830,73 @@ public class AgentServiceImpl implements AgentService {
                    }
                }
            }
-
        }
+        if(jsonObject.containsKey("mtrPolicys")){
+            String mtrPolicys = jsonObject.getString("mtrPolicys");
+            if(!mtrPolicys.equals(GlobalConfig.MTRPOLICYMSG)){
+                GlobalConfig.MTRPOLICYMSG = mtrPolicys;
+                List<MtrPolicyConfigEO> mtrPolicyConfigEOList = JSON.parseArray(mtrPolicys, MtrPolicyConfigEO.class);
+                try {
+                    if (CollectionUtil.isEmpty(mtrPolicyConfigEOList)) {
+                        // 如果没有策略，取消定时任务
+                        dynamicTaskService.cancelTask("mtrTask");
+                        AssertLog.info("MTR策略为空，取消MTR定时任务");
+                        return;
+                    }

+                    // 检查是否有有效的策略（在时间范围内的策略）
+                    boolean hasValidPolicy = false;
+                    long currentTimestamp = System.currentTimeMillis() / 1000;
+
+                    for (MtrPolicyConfigEO policy : mtrPolicyConfigEOList) {
+                        if (isPolicyTimeValid(policy, currentTimestamp)) {
+                            hasValidPolicy = true;
+                            break;
+                        }
+                    }
+
+                    if (!hasValidPolicy) {
+                        // 没有有效策略，取消定时任务
+                        dynamicTaskService.cancelTask("mtrTask");
+                        AssertLog.info("没有有效的MTR策略，取消MTR定时任务");
+                        return;
+                    }
+
+                    // 计算下一个5分钟整点的延迟时间
+                    long milli = AgentUtil.millisecondsToNext5Minute();
+
+                    // 启动或更新定时任务
+                    dynamicTaskService.scheduleTask("mtrTask",
+                            businessTasks::mtrTask, milli, 300000L); // 5分钟间隔
+
+                    AssertLog.info("MTR定时任务已启动/更新，延迟: {}ms, 间隔: 300000ms", milli);
+
+                } catch (Exception e) {
+                    AssertLog.error("启动MTR定时任务异常", e);
+                }
+            }
+        }
    }
+    /**
+     * 检查策略时间是否有效
+     */
+    private boolean isPolicyTimeValid(MtrPolicyConfigEO policy, long currentTimestamp) {
+        try {
+            Date currentTime = new Date(currentTimestamp * 1000);
+            Date startTime = policy.getStartTime();
+            Date endTime = policy.getEndTime();

+            // 开始时间为空表示立即开始
+            boolean startValid = startTime == null || !currentTime.before(startTime);
+            // 结束时间为空表示永不过期
+            boolean endValid = endTime == null || !currentTime.after(endTime);
+
+            return startValid && endValid;
+        } catch (Exception e) {
+            AssertLog.error("检查策略时间有效性异常", e);
+            return false;
+        }
+    }
    @Override
    public void checkTrAgent() {
        String CRON_FILE_PATH = properties.getScriptPath()+"/tr_live.cron";
@@ -904,7 +971,7 @@ public class AgentServiceImpl implements AgentService {

    @Override
    public boolean connection() {
-        String clientId = MachineFingerprint.getHardwareFingerprint();
+        String clientId = "MTR" + MachineFingerprint.getHardwareFingerprint();
        //初始化SN信息
        GlobalConfig.DEVICE_SN = AgentUtil.getDeviceSN();
        //检查外置目录是否存在
@@ -0,0 +1,213 @@
+package com.tongran.agent.client.service.impl;
+
+import cn.hutool.core.collection.CollectionUtil;
+import com.alibaba.fastjson2.JSON;
+import com.alibaba.fastjson2.JSONObject;
+import com.tongran.agent.client.core.eo.MtrPolicyConfigEO;
+import com.tongran.agent.client.core.vo.MtrVO;
+import com.tongran.agent.client.core.vo.NetVO;
+import com.tongran.agent.client.service.MtrService;
+import com.tongran.agent.client.utils.AgentUtil;
+import com.tongran.agent.client.utils.AssertLog;
+import org.springframework.stereotype.Service;
+import oshi.SystemInfo;
+import oshi.hardware.HardwareAbstractionLayer;
+import oshi.hardware.NetworkIF;
+import oshi.util.FormatUtil;
+
+import java.io.BufferedReader;
+import java.io.InputStreamReader;
+import java.util.*;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.TimeUnit;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+
+@Service
+public class MtrServiceImpl implements MtrService {
+    /**
+     * 执行MTR探测列表 - 异步版本
+     */
+    public List<MtrVO> mtrList(long timestamp, List<MtrPolicyConfigEO> mtrPolicyConfigEOList) {
+        List<CompletableFuture<MtrVO>> futures = new ArrayList<>();
+        List<MtrVO> mtrResults = new ArrayList<>();
+
+        try {
+            if (CollectionUtil.isEmpty(mtrPolicyConfigEOList)) {
+                return mtrResults;
+            }
+
+            // 为每个IP创建异步任务
+            for (MtrPolicyConfigEO policy : mtrPolicyConfigEOList) {
+                if (!isPolicyTimeValid(policy, timestamp)) {
+                    continue;
+                }
+
+                String[] targetIps = policy.getServeripGroup().split(";");
+                for (String targetIp : targetIps) {
+                    String ip = targetIp.trim();
+                    if (!ip.isEmpty()) {
+                        // 异步执行每个IP的探测
+                        CompletableFuture<MtrVO> future = CompletableFuture.supplyAsync(() ->
+                                executeSingleMtrProbe(ip, policy, timestamp)
+                        );
+                        futures.add(future);
+                    }
+                }
+            }
+
+            // 等待所有任务完成
+            CompletableFuture<Void> allFutures = CompletableFuture.allOf(
+                    futures.toArray(new CompletableFuture[0])
+            );
+
+            // 获取所有结果（设置超时时间，比如2分钟）
+            allFutures.get(2, TimeUnit.MINUTES);
+
+            for (CompletableFuture<MtrVO> future : futures) {
+                MtrVO result = future.get();
+                if (result != null) {
+                    mtrResults.add(result);
+                }
+            }
+
+        } catch (Exception e) {
+            AssertLog.error("MTR异步探测异常", e);
+        }
+
+        return mtrResults;
+    }
+
+
+    /**
+     * 检查策略时间是否有效
+     */
+    private boolean isPolicyTimeValid(MtrPolicyConfigEO policy, long currentTimestamp) {
+        try {
+            // 将时间戳转换为Date对象进行比较
+            Date currentTime = new Date(currentTimestamp * 1000);
+            Date startTime = policy.getStartTime();
+            Date endTime = policy.getEndTime();
+
+            // 开始时间为空表示立即开始
+            boolean startValid = startTime == null || !currentTime.before(startTime);
+            // 结束时间为空表示永不过期
+            boolean endValid = endTime == null || !currentTime.after(endTime);
+
+            return startValid && endValid;
+        } catch (Exception e) {
+            AssertLog.error("检查策略时间有效性异常", e);
+            return false;
+        }
+    }
+
+    /**
+     * 执行单个IP的MTR探测
+     */
+    private MtrVO executeSingleMtrProbe(String targetIp, MtrPolicyConfigEO policy, long timestamp) {
+        MtrVO mtrVO = new MtrVO();
+        mtrVO.setTargetIp(targetIp);
+        mtrVO.setPolicyId(policy.getId());
+        mtrVO.setTimestamp(timestamp);
+
+        try {
+            // 第一次MTR探测
+            double firstLossPercent = executeMtrCommand(targetIp);
+            mtrVO.setFirstLossPercent(firstLossPercent);
+
+            // 如果丢包率超过5%，30秒后重试
+            if (firstLossPercent > 5.0) {
+                AssertLog.info("目标IP {} 丢包率 {}% > 5%，30秒后重试探测", targetIp, firstLossPercent);
+
+                // 等待30秒
+                Thread.sleep(30000);
+
+                // 第二次MTR探测
+                double secondLossPercent = executeMtrCommand(targetIp);
+                mtrVO.setFinalLossPercent(secondLossPercent);
+                mtrVO.setHasRetry(true);
+
+                AssertLog.info("目标IP {} 重试探测完成，最终丢包率: {}%", targetIp, secondLossPercent);
+            } else {
+                mtrVO.setFinalLossPercent(firstLossPercent);
+                mtrVO.setHasRetry(false);
+                AssertLog.info("目标IP {} 丢包率 {}%，无需重试", targetIp, firstLossPercent);
+            }
+
+            return mtrVO;
+
+        } catch (Exception e) {
+            AssertLog.error("MTR探测异常 - IP: {}", targetIp, e);
+            mtrVO.setFinalLossPercent(-1.0); // -1表示探测失败
+            mtrVO.setErrorMsg(e.getMessage());
+            return mtrVO;
+        }
+    }
+
+    /**
+     * 执行mtr -r命令并解析结果
+     */
+    private double executeMtrCommand(String targetIp) {
+        Process process = null;
+        try {
+            // 执行mtr -r -c 10命令
+            ProcessBuilder processBuilder = new ProcessBuilder("mtr", "-r", "-c", "10", targetIp);
+            process = processBuilder.start();
+
+            // 读取命令输出
+            BufferedReader reader = new BufferedReader(new InputStreamReader(process.getInputStream()));
+            String line;
+            String lastValidLine = null;
+
+            while ((line = reader.readLine()) != null) {
+                // 跳过标题行，记录最后一条有效数据行
+                if (!line.trim().isEmpty() &&
+                        !line.startsWith("Start:") &&
+                        !line.startsWith("HOST:") &&
+                        Character.isDigit(line.trim().charAt(0))) {
+                    lastValidLine = line;
+                }
+            }
+
+            // 等待命令执行完成
+            boolean finished = process.waitFor(2, TimeUnit.MINUTES);
+            if (!finished) {
+                process.destroy();
+                throw new RuntimeException("MTR命令执行超时");
+            }
+
+            if (lastValidLine == null) {
+                throw new RuntimeException("MTR命令无有效输出");
+            }
+
+            // 解析丢包率
+            return parseLossPercentFromMtrOutput(lastValidLine);
+
+        } catch (Exception e) {
+            throw new RuntimeException("MTR命令执行异常: " + e.getMessage());
+        } finally {
+            if (process != null) {
+                process.destroy();
+            }
+        }
+    }
+
+    /**
+     * 从MTR输出中解析丢包率
+     */
+    private double parseLossPercentFromMtrOutput(String mtrOutputLine) {
+        try {
+            // 使用正则表达式匹配Loss%列
+            Pattern pattern = Pattern.compile("(\\d+\\.?\\d*)%");
+            Matcher matcher = pattern.matcher(mtrOutputLine);
+
+            if (matcher.find()) {
+                return Double.parseDouble(matcher.group(1));
+            }
+
+            throw new RuntimeException("未找到Loss%数据: " + mtrOutputLine);
+        } catch (Exception e) {
+            throw new RuntimeException("解析Loss%失败: " + e.getMessage());
+        }
+    }
+}
@@ -17,7 +17,7 @@ logging:
 netty:
  server:
    host: 120.211.95.173
-    port: 6620
+    port: 6610
  client:
    client-id: client-001
    reconnect-interval: 5
@@ -5,8 +5,8 @@ spring:
    pathmatch:
      matching-strategy: ant_path_matcher
  application:
-    name: tr-agent-client
-    version: 1.1.4
+    name: tr-mtragent-client
+    version: 1.0.0
    conf-path: /usr/local/tongran/conf
    script-path: /usr/local/tongran/sbin
    tmp-path: /usr/local/tongran/tmp