dubbo监控 – 小飞侠

1、原理

dubbo架构如下:

通过阅读dubbo源码，所有的rpc方法调用都会经过MonitorFilter进行拦截，

MonitorFilter.invoke()

public Result invoke(Invoker<?> invoker, Invocation invocation) throws RpcException {
        if (invoker.getUrl().hasParameter("monitor")) {
            RpcContext context = RpcContext.getContext();
            long start = System.currentTimeMillis();
            this.getConcurrent(invoker, invocation).incrementAndGet();

            Result var7;
            try {
                Result result = invoker.invoke(invocation);
            this.collect(invoker, invocation, result, context, start, false);
                var7 = result;
            } catch (RpcException var11) {
                this.collect(invoker, invocation, (Result)null, context, start, true);
                throw var11;
            } finally {
                this.getConcurrent(invoker, invocation).decrementAndGet();
            }

            return var7;
        } else {
            return invoker.invoke(invocation);
        }
    }
复制代码

对于配置了监控的服务，会收集一些方法的基本统计信息。

MonitorFilter.collect()

private void collect(Invoker<?> invoker, Invocation invocation, Result result, RpcContext context, long start, boolean error) {
        try {
            long elapsed = System.currentTimeMillis() - start;
            int concurrent = this.getConcurrent(invoker, invocation).get();
            String application = invoker.getUrl().getParameter("application");
            String service = invoker.getInterface().getName();
            String method = RpcUtils.getMethodName(invocation);
            URL url = invoker.getUrl().getUrlParameter("monitor");
            Monitor monitor = this.monitorFactory.getMonitor(url);
            int localPort;
            String remoteKey;
            String remoteValue;
            if ("consumer".equals(invoker.getUrl().getParameter("side"))) {
                context = RpcContext.getContext();
                localPort = 0;
                remoteKey = "provider";
                remoteValue = invoker.getUrl().getAddress();
            } else {
                localPort = invoker.getUrl().getPort();
                remoteKey = "consumer";
                remoteValue = context.getRemoteHost();
            }

            String input = "";
            String output = "";
            if (invocation.getAttachment("input") != null) {
                input = invocation.getAttachment("input");
            }

            if (result != null && result.getAttachment("output") != null) {
                output = result.getAttachment("output");
            }

            monitor.collect(new URL("count", NetUtils.getLocalHost(), localPort, service + "/" + method, new String[]{"application", application, "interface", service, "method", method, remoteKey, remoteValue, error ? "failure" : "success", "1", "elapsed", String.valueOf(elapsed), "concurrent", String.valueOf(concurrent), "input", input, "output", output}));
        } catch (Throwable var21) {
            logger.error("Failed to monitor count service " + invoker.getUrl() + ", cause: " + var21.getMessage(), var21);
        }
    }
复制代码

DubboMonitor对收集到的数据进行简单统计，诸如成功次数，失败次数，调用时间等，统计完后存储数据到本地。

DubboMonitor.collect()

public void collect(URL url) {
        int success = url.getParameter("success", 0);
        int failure = url.getParameter("failure", 0);
        int input = url.getParameter("input", 0);
        int output = url.getParameter("output", 0);
        int elapsed = url.getParameter("elapsed", 0);
        int concurrent = url.getParameter("concurrent", 0);
        Statistics statistics = new Statistics(url);
        AtomicReference<long[]> reference = (AtomicReference)this.statisticsMap.get(statistics);
        if (reference == null) {
            this.statisticsMap.putIfAbsent(statistics, new AtomicReference());
            reference = (AtomicReference)this.statisticsMap.get(statistics);
        }

        long[] update = new long[10];

        long[] current;
        do {
            current = (long[])reference.get();
            if (current == null) {
                update[0] = (long)success;
                update[1] = (long)failure;
                update[2] = (long)input;
                update[3] = (long)output;
                update[4] = (long)elapsed;
                update[5] = (long)concurrent;
                update[6] = (long)input;
                update[7] = (long)output;
                update[8] = (long)elapsed;
                update[9] = (long)concurrent;
            } else {
                update[0] = current[0] + (long)success;
                update[1] = current[1] + (long)failure;
                update[2] = current[2] + (long)input;
                update[3] = current[3] + (long)output;
                update[4] = current[4] + (long)elapsed;
                update[5] = (current[5] + (long)concurrent) / 2L;
                update[6] = current[6] > (long)input ? current[6] : (long)input;
                update[7] = current[7] > (long)output ? current[7] : (long)output;
                update[8] = current[8] > (long)elapsed ? current[8] : (long)elapsed;
                update[9] = current[9] > (long)concurrent ? current[9] : (long)concurrent;
            }
        } while(!reference.compareAndSet(current, update));

    }
复制代码

DubboMonitor有异步线程定时(默认每分钟)将收集到数据发送到远端监控服务。

 public DubboMonitor(Invoker<MonitorService> monitorInvoker, MonitorService monitorService) {
        this.monitorInvoker = monitorInvoker;
        this.monitorService = monitorService;
        this.monitorInterval = (long)monitorInvoker.getUrl().getPositiveParameter("interval", 60000);
        this.sendFuture = this.scheduledExecutorService.scheduleWithFixedDelay(new Runnable() {
            public void run() {
                try {
                    DubboMonitor.this.send();
                } catch (Throwable var2) {
                    DubboMonitor.logger.error("Unexpected error occur at send statistic, cause: " + var2.getMessage(), var2);
                }

            }
        }, this.monitorInterval, this.monitorInterval, TimeUnit.MILLISECONDS);
    }
复制代码

调用远端的MonitorService.collect方法，然后将本地缓存数据置置零。

DubboMonitor.send()

 public void send() {
        if (logger.isInfoEnabled()) {
            logger.info("Send statistics to monitor " + this.getUrl());
        }

        String timestamp = String.valueOf(System.currentTimeMillis());
        Iterator i$ = this.statisticsMap.entrySet().iterator();

        while(i$.hasNext()) {
            Entry<Statistics, AtomicReference<long[]>> entry = (Entry)i$.next();
            Statistics statistics = (Statistics)entry.getKey();
            AtomicReference<long[]> reference = (AtomicReference)entry.getValue();
            long[] numbers = (long[])reference.get();
            long success = numbers[0];
            long failure = numbers[1];
            long input = numbers[2];
            long output = numbers[3];
            long elapsed = numbers[4];
            long concurrent = numbers[5];
            long maxInput = numbers[6];
            long maxOutput = numbers[7];
            long maxElapsed = numbers[8];
            long maxConcurrent = numbers[9];
            URL url = statistics.getUrl().addParameters(new String[]{"timestamp", timestamp, "success", String.valueOf(success), "failure", String.valueOf(failure), "input", String.valueOf(input), "output", String.valueOf(output), "elapsed", String.valueOf(elapsed), "concurrent", String.valueOf(concurrent), "max.input", String.valueOf(maxInput), "max.output", String.valueOf(maxOutput), "max.elapsed", String.valueOf(maxElapsed), "max.concurrent", String.valueOf(maxConcurrent)});
            this.monitorService.collect(url);
            long[] update = new long[10];

            while(true) {
                long[] current = (long[])reference.get();
                if (current == null) {
                    update[0] = 0L;
                    update[1] = 0L;
                    update[2] = 0L;
                    update[3] = 0L;
                    update[4] = 0L;
                    update[5] = 0L;
                } else {
                    update[0] = current[0] - success;
                    update[1] = current[1] - failure;
                    update[2] = current[2] - input;
                    update[3] = current[3] - output;
                    update[4] = current[4] - elapsed;
                    update[5] = current[5] - concurrent;
                }

                if (reference.compareAndSet(current, update)) {
                    break;
                }
            }
        }

    }

复制代码

dubbo监控的主流开源项目，都是实现了MonitorService接口来实现监控，区别无非就是数据存储，报表统计逻辑的差异，基本原理都大同小异。

public interface MonitorService {
    String APPLICATION = "application";
    String INTERFACE = "interface";
    String METHOD = "method";
    String GROUP = "group";
    String VERSION = "version";
    String CONSUMER = "consumer";
    String PROVIDER = "provider";
    String TIMESTAMP = "timestamp";
    String SUCCESS = "success";
    String FAILURE = "failure";
    String INPUT = "input";
    String OUTPUT = "output";
    String ELAPSED = "elapsed";
    String CONCURRENT = "concurrent";
    String MAX_INPUT = "max.input";
    String MAX_OUTPUT = "max.output";
    String MAX_ELAPSED = "max.elapsed";
    String MAX_CONCURRENT = "max.concurrent";

    void collect(URL var1);

    List<URL> lookup(URL var1);
}
复制代码

2、监控选型

主流dubbo监控主要有：

dubbo-monitor
dubbo-d-monitor
dubbokeeper
dubbo-monitor-simple

下面进行简单的对比：

方案	支持版本	基础功能	开源作者	社区活跃度	数据存储	维护成本
dubbo-monitor	基于dubbox，理论上也支持dubbo	一般,QPS、RT、服务状态等，缺乏报表功能	韩都衣舍	513星，两年前有提交	mysql、mongodb	无侵入、需要定期清理历史数据
dubbo-d-monitor	dubbo	一般，只有一些基础数据	个人	189星，一年前有提交	mysql、redis(后续不再维护)	无侵入、需要定期清理历史数据
dubbokeeper	dubbo	丰富，除了基础指标数据，有top200数据报表，还提供了类似dubbo-admin功能(限流、超时时间设置、消费客户端设置、容错等)，同时支持zk节点可视化	个人组织	989星，一个月内有提交	mysql、mongodb、lucene	无侵入、需要定期清理历史记录
dubbo-monitor-simple	dubbo	简陋	dubbo官方	330星，一个月内有提交	文件存储	无侵入、但目前线上使用发现数据量大了经常挂

对比以上几种，dubbokeeper>dubbo-monitor>dubbo-d-monitor，所以选取dubbokeeper最为dubbo服务监控方案。

3、部署

我们采用mongodb存储方案，采用单机部署。

环境：jdk1.8及以上（低版未测试），安装tomcat，安装zookeeper并启动，安装启动mongodb

1、获取源码

github.com/dubboclub/d…

2、解压下载下来的zip包dubbokeeper-master到任意目录，修改解压后的项目中dubbo及数据库的配置\dubbokeeper-master\conf\dubbo-mongodb.properties。

执行\dubbokeeper-master\

install-mongodb.sh

执行完上一步后会生成一个target目录，目录下会存在以下三个文件夹及一个压缩包

 archive-tmp
 mongodb-dubbokeeper-server
 mongodb-dubbokeeper-ui
 mongodb-dubbokeeper-server.tar.gz
复制代码

3、执行mongodb-dubbokeeper-server/bin/start-mongodb.sh启动存储端（数据存储和web端是分开独立部署的）

4、将mongodb-dubbokeeper-ui下的war包拷贝到tomcat的webapps目录下，启动tomcat。

5、最后，打开浏览器，输入http://localhost:8080/dubbokeeper-ui-1.0.1即可。

在业务代码中，只需要配置dubbo监控连接到注册中心，就能完成监控数据采集。

<dubbo:monitor protocol="registry"/>
复制代码

主要的配置信息：

dubbo.application.name=mongodb-monitor
dubbo.application.owner=bieber
dubbo.registry.address=zookeeper://*.*.*.*:2181?backup=*.*.*.*:2181,*.*.*.*:2181
dubbo.protocol.name=dubbo
dubbo.protocol.port=20884
dubbo.protocol.dubbo.payload=20971520

#dubbo数据采集周期 单位毫秒
monitor.collect.interval=60000

#use netty4
dubbo.provider.transporter=netty4

#dubbokeeper写入mongodb周期 单位秒
monitor.write.interval=60

#mongdb配置
dubbo.monitor.mongodb.url=localhost
dubbo.monitor.mongodb.port=27017
dubbo.monitor.mongodb.dbname=dubbokeeper
dubbo.monitor.mongodb.username=
dubbo.monitor.mongodb.password=
dubbo.monitor.mongodb.storage.timeout=60000
复制代码

4、主要功能介绍

首页能看到应用总体信息（区分应用提供者和消费者），服务数量信息，节点部署信息及依赖关系图等。

Admin提供了所有原生dubbo-admin的绝大部分功能。

ZooPeeper可以查看zookeeper节点信息

Monitor可以查看dubbo监控相关信息

应用总览信息，可根据时间筛选：

应用详细信息，有接口耗时、并发、失败、成功等数据：

方法级别总览及详细信息：

5、遇到的坑

1、官方默认monitor.write.interval(存储周期)配置的是6000，阅读源码发现单位是秒，也就是默认配置100分钟才会写入mongodb，要把它改成60。

2、dubbokeeper默认没有对collections加索引，数据量大了之后打开会异常慢，所以需要自己通过脚本对collection加索引。

import pymongo
from pymongo import MongoClient
import time
import datetime
import sys
import os


client = MongoClient('127.0.0.1', 27017)
db = client['dubbokeeper']

collectionlist = db.collection_names()

for collection in collectionlist:
    if collection!='application':
        db[collection].ensure_index([("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("serviceInterface",pymongo.DESCENDING)])
        db[collection].ensure_index([("method",pymongo.DESCENDING)])
        db[collection].ensure_index([("serviceInterface",pymongo.DESCENDING),("method",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("serviceInterface",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("concurrent",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("elapsed",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("failureCount",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("successCount",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("serviceInterface",pymongo.DESCENDING),("elapsed",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("serviceInterface",pymongo.DESCENDING),("concurrent",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("serviceInterface",pymongo.DESCENDING),("failureCount",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        db[collection].ensure_index([("serviceInterface",pymongo.DESCENDING),("successCount",pymongo.DESCENDING),("timestamp",pymongo.DESCENDING)])
        
print 'success'
复制代码

3、一般历史数据基本不用保存太久，目前我们线上保留2周数据，提供了以下脚本定期删除数据。

import pymongo
from pymongo import MongoClient
import time
import datetime
import sys
import os

day=int(sys.argv[1])
print day
timestamp = time.time()*1000-1000*24*3600*day

print timestamp

client = MongoClient('127.0.0.1', 27017)
db = client['dubbokeeper']

collectionlist = db.collection_names()

for collection in collectionlist:
    if collection!='application':
        db[collection].remove({"timestamp": {"$lt": timestamp}})

print 'clean mongodb data success'
复制代码

每天定时清理15天的数据

0 3 * * * python /home/monitor/shell/clean-mongodb.py 15
复制代码

4、mongodb缓存比较吃内存，最好配置8G以上的服务器，或者量大可以考虑集群部署

5、dubbokeeper-ui原生交互有点坑，有些页面会遍历展示所有应用的数据，效率比较低下。如果应用过多可能会超时打不开，服务端团队对交互进行了简单优化，每次只能查看一个应用或一个接口，如果大家有需求可以留言，我们后续会开源出来。

1、原理

2、监控选型

3、部署

4、主要功能介绍

5、遇到的坑

你可能也喜欢