Merge pull request MyCATApache#20 from MyCATApache/master

合拼
zwyqz · Oct 19, 2017 · af5590d · af5590d
2 parents e5b04d3 + 8e74c6f
commit af5590d
Show file tree

Hide file tree

Showing 43 changed files with 2,572 additions and 264 deletions.
diff --git a/doc/.project b/doc/.project
@@ -0,0 +1,11 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<projectDescription>
+	<name>doc</name>
+	<comment></comment>
+	<projects>
+	</projects>
+	<buildSpec>
+	</buildSpec>
+	<natures>
+	</natures>
+</projectDescription>
diff --git a/doc/articles/mycat2.0-启动流程.md b/doc/articles/mycat2.0-启动流程.md
@@ -0,0 +1,351 @@
+1. 程序的入口是io.mycat.mycat2.MycatCore.
+在main 方法中 首选取得ProxyRuntime的实例,该类是一个单例模式.
+2. 
+
+```
+runtime.setConfig(new MycatConfig());
+```
+该方法为runtime设置了一个MycatConfig ,MycatConfig 是负责读取配置文件的.
+
+3. 
+```
+ConfigLoader.INSTANCE.loadCore();
+```
+
+调用ConfigLoader 进行配置文件的加载.
+
+ 3.1. 加载mycat.yml,heartbeat.yml,cluster.yml,balancer.yml,user.yml 的配置文件.代码如下:
+
+```
+public void loadCore() throws IOException {
+    loadConfig(ConfigEnum.PROXY, GlobalBean.INIT_VERSION);
+    loadConfig(ConfigEnum.HEARTBEAT, GlobalBean.INIT_VERSION);
+    loadConfig(ConfigEnum.CLUSTER, GlobalBean.INIT_VERSION);
+    loadConfig(ConfigEnum.BALANCER, GlobalBean.INIT_VERSION);
+    loadConfig(ConfigEnum.USER, GlobalBean.INIT_VERSION);
+}
+```
+ 3.2 进行配置文件的加载,调用了YamlUtil进行加载,并将配置信息赋值给MycatConfig,关键代码如下 :
+
+```
+  conf.putConfig(configEnum, (Configurable) YamlUtil.load(fileName, configEnum.getClazz()), version);
+```
+ 3.3 在io.mycat.util.YamlUtil.load(String, Class<T>) 中,处理也很简单,通过Yaml进行文件的加载,然后通过反射机制将属性进行赋值.代码如下:
+
+```
+/**
+ * 从指定的文件中加载配置
+ * @param fileName 需要加载的文件名
+ * @param clazz 加载后需要转换成的类对象
+ * @return
+ * @throws FileNotFoundException
+ */
+public static <T> T load(String fileName, Class<T> clazz) throws FileNotFoundException {
+    InputStreamReader fis = null;
+    try {
+        URL url = YamlUtil.class.getClassLoader().getResource(fileName);
+        if (url != null) {
+            Yaml yaml = new Yaml();
+            fis = new InputStreamReader(new FileInputStream(url.getFile()), StandardCharsets.UTF_8);
+            T obj = yaml.loadAs(fis, clazz);
+            return obj;
+        }
+        return null;
+    } finally {
+        if (fis != null) {
+            try {
+                fis.close();
+            } catch (IOException ignored) {
+            }
+        }
+    }
+}
+```
+4. 进行通过命令行传入参数的解析.如:可以传入
+    -  -mycat.proxy.port 8067
+    -  -mycat.cluster.enable true
+    -  -mycat.cluster.port 9067
+    -  -mycat.cluster.myNodeId leader-2
+
+参数等 
+支持的参数可在 io.mycat.mycat2.beans.ArgsBean 中进行查看.
+
+ 获得参数后通过遍历进行赋值操作即可,代码如下:
+
+```
+private static void solveArgs(String[] args) {
+int lenght = args.length;
+
+MycatConfig conf = ProxyRuntime.INSTANCE.getConfig();
+ProxyConfig proxyConfig = conf.getConfig(ConfigEnum.PROXY);
+ClusterConfig clusterConfig = conf.getConfig(ConfigEnum.CLUSTER);
+BalancerConfig balancerConfig= conf.getConfig(ConfigEnum.BALANCER);
+
+for (int i = 0; i < lenght; i++) {
+	switch(args[i]) {
+		case ArgsBean.PROXY_PORT:
+			proxyConfig.getProxy().setPort(Integer.parseInt(args[++i]));
+			break;
+		case ArgsBean.CLUSTER_ENABLE:
+			clusterConfig.getCluster().setEnable(Boolean.parseBoolean(args[++i]));
+			break;
+		case ArgsBean.CLUSTER_PORT:
+			clusterConfig.getCluster().setPort(Integer.parseInt(args[++i]));
+			break;
+		case ArgsBean.CLUSTER_MY_NODE_ID:
+			clusterConfig.getCluster().setMyNodeId(args[++i]);
+			break;
+		case ArgsBean.BALANCER_ENABLE:
+			balancerConfig.getBalancer().setEnable(Boolean.parseBoolean(args[++i]));
+			break;
+		case ArgsBean.BALANCER_PORT:
+			balancerConfig.getBalancer().setPort(Integer.parseInt(args[++i]));
+			break;
+		case ArgsBean.BALANCER_STRATEGY:
+			BalancerBean.BalancerStrategyEnum strategy = BalancerBean.BalancerStrategyEnum.getEnum(args[++i]);
+			if (strategy == null) {
+				throw new IllegalArgumentException("no such balancer strategy");
+			}
+			balancerConfig.getBalancer().setStrategy(strategy);
+			break;
+		default:
+			break;
+	}
+}
+}
+```
+
+
+
+5. 设置NioReactorThreads,线程数目按照cpu的数目而定.
+
+```
+int cpus = Runtime.getRuntime().availableProcessors();
+runtime.setNioReactorThreads(cpus);
+runtime.setReactorThreads(new MycatReactorThread[cpus]);
+```
+6.设置MycatSessionManager
+
+7. 然后调用io.mycat.proxy.ProxyRuntime 的init方法,进行资源的初始化.代码如下:
+
+```
+public void init() {
+//心跳调度独立出来，避免被其他任务影响
+heartbeatScheduler = Executors.newSingleThreadScheduledExecutor();
+HeartbeatConfig heartbeatConfig = config.getConfig(ConfigEnum.HEARTBEAT);
+timerExecutor = ExecutorUtil.create("Timer", heartbeatConfig.getHeartbeat().getTimerExecutor());
+businessExecutor = ExecutorUtil.create("BusinessExecutor",Runtime.getRuntime().availableProcessors());
+listeningExecutorService = MoreExecutors.listeningDecorator(businessExecutor);
+MatchMethodGenerator.initShrinkCharTbl();
+}
+```
+7.1. 建立心跳线程, 原因是避免被其他任务影响.
+
+7.2  建立timerExecutor,线程数目默认为2.
+
+7.3  建立businessExecutor,线程数目为cpu数目.
+
+7.4  建立listeningExecutorService, 该ExecutorService将会进行任务的提交给businessExecutor
+
+7.5 调用io.mycat.mycat2.sqlparser.MatchMethodGenerator#initShrinkCharTbl方法.该方法只是对0-9a-zA-Z的字符进行映射.代码如下:
+
+
+```
+static final byte[] shrinkCharTbl = new byte[96];//为了压缩hash字符映射空间，再次进行转义
+public static void initShrinkCharTbl () {
+    shrinkCharTbl[0] = 1;//从 $ 开始计算
+    IntStream.rangeClosed('0', '9').forEach(c -> shrinkCharTbl[c-'$'] = (byte)(c-'0'+2));
+    IntStream.rangeClosed('A', 'Z').forEach(c -> shrinkCharTbl[c-'$'] = (byte)(c-'A'+12));
+    IntStream.rangeClosed('a', 'z').forEach(c -> shrinkCharTbl[c-'$'] = (byte)(c-'a'+12));
+    shrinkCharTbl['_'-'$'] = (byte)38;
+}
+```
+8. 调用ProxyStarter的start,首先 是创建了NIOAcceptor,负责对请求进行处理.然后根据ClusterConfig的配置进行相应的处理.
+
+
+```
+ClusterConfig clusterConfig = conf.getConfig(ConfigEnum.CLUSTER);
+ClusterBean clusterBean = clusterConfig.getCluster();
+if (clusterBean.isEnable()) {
+	// 启动集群
+	startCluster(runtime, clusterBean, acceptor);
+} else {
+	// 未配置集群，直接启动
+	startProxy(true);
+}
+```
+
+9.接下来分析未配置集群的启动方式.
+
+9.1 首先通过ProxyRuntime获取到ProxyConfig(该对象是mycat.yml的封装)，mycat.yml的配置文件如下:
+
+```
+proxy:
+  ip: 0.0.0.0
+  port: 8066
+```
+因此可以通过该类拿到启动端口和ip.因此传入NIOAcceptor进行监听.
+
+9.2 在io.mycat.proxy.NIOAcceptor#startServerChannel中,做了如下处理:
+
+9.2.1 首先检查ServerSocketChannel是否已经启动,如果已经启动,不进行后续处理.
+
+9.2.2 根据传入的ServerType做不同的处理,当前我们传入的值为MYCAT. 因此不进行处理.
+
+```
+if (serverType == ServerType.CLUSTER) {
+	adminSessionMan = new DefaultAdminSessionManager();
+	ProxyRuntime.INSTANCE.setAdminSessionManager(adminSessionMan);
+	logger.info("opend cluster conmunite port on {}:{}", ip, port);
+} else if (serverType == ServerType.LOAD_BALANCER){
+	logger.info("opend load balance conmunite port on {}:{}", ip, port);
+	ProxyRuntime.INSTANCE.setProxySessionSessionManager(new ProxySessionManager());
+	ProxyRuntime.INSTANCE.setLbSessionSessionManager(new LBSessionManager());
+}
+```
+
+9.2.3 调用io.mycat.proxy.NIOAcceptor#getServerSocketChannel,进行启动.代码如下:
+
+
+```
+private void openServerChannel(Selector selector, String bindIp, int bindPort, ServerType serverType)
+	throws IOException {
+final ServerSocketChannel serverChannel = ServerSocketChannel.open();
+final InetSocketAddress isa = new InetSocketAddress(bindIp, bindPort);
+serverChannel.bind(isa);
+serverChannel.configureBlocking(false);
+serverChannel.setOption(StandardSocketOptions.SO_REUSEADDR, true);
+serverChannel.register(selector, SelectionKey.OP_ACCEPT, serverType);
+if (serverType == ServerType.CLUSTER) {
+	logger.info("open cluster server port on {}:{}", bindIp, bindPort);
+	clusterServerSocketChannel = serverChannel;
+} else if (serverType == ServerType.LOAD_BALANCER) {
+	logger.info("open load balance server port on {}:{}", bindIp, bindPort);
+	loadBalanceServerSocketChannel = serverChannel;
+} else {
+	logger.info("open proxy server port on {}:{}", bindIp, bindPort);
+	proxyServerSocketChannel = serverChannel;
+}
+}
+```
+
+9.4 调用io.mycat.mycat2.ProxyStarter#startReactor,进行MycatReactorThread的启动.该线程负责对请求进行处理.
+
+9.5 加载配置文件信息,该方法在ROOT_PATH目录下创建了prepare和archive两个文件夹,并加载了replica-index.yml,datasource.yml,schema.yml,最后调用了AnnotationProcessor#getInstance  进行初始化.
+
+9.5.1 创建DynamicAnnotationManager.
+
+9.5.2 对AnnotationProcessor.class.getClassLoader().getResource("") 的路径进行监听.当文件目录有变化时会回调io.mycat.mycat2.sqlannotations.AnnotationProcessor#listen方法.代码如下:
+
+
+```
+public static void listen() {
+try {
+while (true) {
+    WatchKey key = watcher.take();//todo 线程复用,用 poll比较好?
+    boolean flag = false;
+    for (WatchEvent<?> event: key.pollEvents()) {
+        String str = event.context().toString();
+        if ("actions.yml".equals(str)|| "annotations.yml".equals(str)) {
+            flag=true;
+            break;
+        }
+    }
+    if (flag){
+        System.out.println("动态注解更新次数" + count.incrementAndGet());
+        init();
+    }
+    boolean valid = key.reset();
+    if (!valid) {
+        break;
+    }
+
+}
+} catch (Exception e) {
+e.printStackTrace();
+}
+}
+```
+
+==可以看到,当actions.yml或者annotations.yml变化时,就会重新调用init方法,从而对DynamicAnnotationManager进行修改.==
+
+
+9.6 分别对datasource.yml 和 schema.yml 进行加载.
+
+9.6.1 对MySQLRepBean进行初始化.该bean是对datasource.yml的封装.datasource.yml如下所示:
+
+
+```
+replicas:
+  - name: test                      # 复制组 名称   必须唯一
+    repType: MASTER_SLAVE           # 复制类型
+    switchType: SWITCH              # 切换类型
+    balanceType: BALANCE_ALL_READ   # 读写分离类型
+    mysqls:            
+      - hostName: mysql-01              # mysql 主机名
+        ip: 127.0.0.1               # ip
+        port: 3306                  # port
+        user: root                  # 用户名
+        password: root            # 密码
+        minCon: 1                   # 最小连接
+        maxCon: 10                  # 最大连接
+        maxRetryCount: 3            # 连接重试次数
+
+```
+
+
+```
+conf.getMysqlRepMap().forEach((repName, repBean) -> {
+	repBean.initMaster();
+	repBean.getMetaBeans().forEach(metaBean -> metaBean.prepareHeartBeat(repBean, repBean.getDataSourceInitStatus()));
+});
+```
+
+9.6.2 调用io.mycat.mycat2.beans.MySQLRepBean#initMaster, 做了如下处理:
+
+首先加载replica-index.yml,获取replica name 对应的index，并对MySQLRepBean中的metaBeans进行赋值,完成对mysqls 的封装.
+
+之后调用MySQLMetaBean的prepareHeartBeat方法.
+完成
+
+
+9.7 ==调用io.mycat.proxy.ProxyRuntime#startHeartBeatScheduler,启动heartbeatScheduler线程.该线程会每10000 ms 调用io.mycat.proxy.ProxyRuntime#replicaHeartbeat.== 心跳的配置在heartbeat.yml中,默认配置如下:
+
+
+```
+heartbeat:
+  timerExecutor: 2
+  replicaHeartbeatPeriod: 10000
+  replicaIdleCheckPeriod: 2000
+  idleTimeout: 2000
+  processorCheckPeriod: 2000
+  minSwitchtimeInterval: 120000
+```
+
+9.7.1 在io.mycat.proxy.ProxyRuntime#replicaHeartbeat,方法中代码如下:
+
+
+```
+private Runnable replicaHeartbeat() {
+return ()->{
+ProxyReactorThread<?> reactor  = getReactorThreads()[ThreadLocalRandom.current().nextInt(getReactorThreads().length)];
+reactor.addNIOJob(()-> config.getMysqlRepMap().values().stream().forEach(f -> f.doHeartbeat()));
+};
+}
+```
+
+最终会调用io.mycat.mycat2.beans.heartbeat.MySQLHeartbeat#heartbeat.
+
+
+9.8 对cluster的配置进行处理.代码如下:
+
+
+```
+BalancerConfig balancerConfig = conf.getConfig(ConfigEnum.BALANCER);
+BalancerBean balancerBean = balancerConfig.getBalancer();
+// 集群模式下才开启负载均衡服务
+if (clusterBean.isEnable() && balancerBean.isEnable()) {
+	runtime.getAcceptor().startServerChannel(balancerBean.getIp(), balancerBean.getPort(), ServerType.LOAD_BALANCER);
+}
+```
+