应用监控CAT之cat-consumer源码阅读
Posted yougewe
tags:
篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了应用监控CAT之cat-consumer源码阅读相关的知识,希望对你有一定的参考价值。
之前讲了 cat-client 进行cat埋点上报,那么上报给谁呢?以及后续故事如何?让我们来看看 cat-consumer 是如何接收处理的?
由cat-client发送数据,cat-consumer进行接收请求处理,开始了处理问题之旅!
首先,让我们来回顾一下 TcpSocketSender 是如何发送数据的:
// TcpSocketSender 往channel中写入数据,此处有兴趣的同学可以延伸下 netty 的源码! private void sendInternal(MessageTree tree) { ChannelFuture future = m_manager.channel(); ByteBuf buf = PooledByteBufAllocator.DEFAULT.buffer(10 * 1024); // 10K m_codec.encode(tree, buf); int size = buf.readableBytes(); Channel channel = future.channel(); // 以 ByteBuf 形式发送数据 channel.writeAndFlush(buf); // 更新统计数据 if (m_statistics != null) { m_statistics.onBytes(size); } }
// TcpSocketReceiver, 接收发送过来的数据,默认端口 2280, 注册服务,线上为分布式部署,即为接口调用式。
public void init() { try { startServer(m_port); } catch (Throwable e) { m_logger.error(e.getMessage(), e); } } public synchronized void startServer(int port) throws InterruptedException { boolean linux = getOSMatches("Linux") || getOSMatches("LINUX"); int threads = 24; ServerBootstrap bootstrap = new ServerBootstrap(); m_bossGroup = linux ? new EpollEventLoopGroup(threads) : new NioEventLoopGroup(threads); m_workerGroup = linux ? new EpollEventLoopGroup(threads) : new NioEventLoopGroup(threads); bootstrap.group(m_bossGroup, m_workerGroup); bootstrap.channel(linux ? EpollServerSocketChannel.class : NioserverSocketChannel.class); // 添加处理handler, 进行请求逻辑处理 bootstrap.childHandler(new ChannelInitializer<SocketChannel>() { @Override protected void initChannel(SocketChannel ch) throws Exception { ChannelPipeline pipeline = ch.pipeline(); // 此处仅为一个解码器,实际功能在该解码器中完成 pipeline.addLast("decode", new MessageDecoder()); } }); bootstrap.childOption(ChannelOption.SO_REUSEADDR, true); bootstrap.childOption(ChannelOption.TCP_NODELAY, true); bootstrap.childOption(ChannelOption.SO_KEEPALIVE, true); bootstrap.childOption(ChannelOption.ALLOCATOR, PooledByteBufAllocator.DEFAULT); try { m_future = bootstrap.bind(port).sync(); m_logger.info("start netty server!"); } catch (Exception e) { m_logger.error("Started Netty Server Failed:" + port, e); } } // 消息解码器,并处理具体业务逻辑,先确认数据已上传完成,再进行逻辑处理 public class MessageDecoder extends ByteToMessageDecoder { @Override protected void decode(ChannelHandlerContext ctx, ByteBuf buffer, List<Object> out) throws Exception { if (buffer.readableBytes() < 4) { return; } buffer.markReaderIndex(); int length = buffer.readInt(); buffer.resetReaderIndex(); if (buffer.readableBytes() < length + 4) { return; } try { if (length > 0) { ByteBuf readBytes = buffer.readBytes(length + 4); readBytes.markReaderIndex(); readBytes.readInt(); DefaultMessageTree tree = (DefaultMessageTree) m_codec.decode(readBytes); readBytes.resetReaderIndex(); tree.setBuffer(readBytes); // 交由handler处理实际逻辑 m_handler.handle(tree); m_processCount++; long flag = m_processCount % CatConstants.SUCCESS_COUNT; if (flag == 0) { m_serverStateManager.addMessageTotal(CatConstants.SUCCESS_COUNT); } } else { // client message is error buffer.readBytes(length); } } catch (Exception e) { m_serverStateManager.addMessageTotalLoss(1); m_logger.error(e.getMessage(), e); } } }
// handler 处理流程,由DefaultMessageHandler接手,安排后续工作。
// DefaultMessageHandler, 接过处理器的第一棒, 交由另一实际的consumer(RealtimeConsumer) handler处理 @Override public void handle(MessageTree tree) { if (m_consumer == null) { m_consumer = lookup(MessageConsumer.class); } try { m_consumer.consume(tree); } catch (Throwable e) { m_logger.error("Error when consuming message in " + m_consumer + "! tree: " + tree, e); } } // RealtimeConsumer, 进行消费数据 @Override public void consume(MessageTree tree) { String domain = tree.getDomain(); String ip = tree.getIpAddress(); // 进行权限检测,ip,domain if (!m_blackListManager.isBlack(domain, ip)) { long timestamp = tree.getMessage().getTimestamp(); Period period = m_periodManager.findPeriod(timestamp); // 找到period, 再将消息分配过去,否则算作网络异常 if (period != null) { period.distribute(tree); } else { m_serverStateManager.addNetworkTimeError(1); } } else { m_black++; if (m_black % CatConstants.SUCCESS_COUNT == 0) { Cat.logEvent("Discard", domain); } } }
// Period.distribute, 将消息依次取出,进行分发到队列
public void distribute(MessageTree tree) { // 统计进行数进行加1 m_serverStateManager.addMessageTotal(tree.getDomain(), 1); boolean success = true; String domain = tree.getDomain(); // 将各种类型的监控数据分别取出进行处理 for (Entry<String, List<PeriodTask>> entry : m_tasks.entrySet()) { List<PeriodTask> tasks = entry.getValue(); int length = tasks.size(); int index = 0; boolean manyTasks = length > 1; if (manyTasks) { index = Math.abs(domain.hashCode()) % length; } PeriodTask task = tasks.get(index); // 如果有金条消息,将task重新入队 boolean enqueue = task.enqueue(tree); if (enqueue == false) { if (manyTasks) { task = tasks.get((index + 1) % length); enqueue = task.enqueue(tree); if (enqueue == false) { success = false; } } else { success = false; } } } if (!success) { m_serverStateManager.addMessageTotalLoss(tree.getDomain(), 1); } } // PeriodTask.enqueue, 重新入队消息,让消费线程自行消费 LinkedBlockingQueue.offer(..) public boolean enqueue(MessageTree tree) { boolean result = m_queue.offer(tree); if (!result) { // trace queue overflow, 记录入队失败日志 m_queueOverflow++; if (m_queueOverflow % (10 * CatConstants.ERROR_COUNT) == 0) { m_logger.warn(m_analyzer.getClass().getSimpleName() + " queue overflow number " + m_queueOverflow); } } return result; }
到此,一条消费线路就完成了。
// PeriodTask 线程,作为第二个消费线路
@Override public void run() { try { // 分析各消息数据,做后台消费处理 m_analyzer.analyze(m_queue); } catch (Exception e) { Cat.logError(e); } } // 调用统一的抽象类的模板方法,由各类进行具体的 process 处理 @Override public void analyze(MessageQueue queue) { while (!isTimeout() && isActive()) { MessageTree tree = queue.poll(); if (tree != null) { try { // 调用具体类的process process(tree); } catch (Throwable e) { m_errors++; if (m_errors == 1 || m_errors % 10000 == 0) { Cat.logError(e); } } } } // 如果出现超时或者停止动作,则把剩余队列处理完成再退出线程 while (true) { MessageTree tree = queue.poll(); if (tree != null) { try { process(tree); } catch (Throwable e) { m_errors++; if (m_errors == 1 || m_errors % 10000 == 0) { Cat.logError(e); } } } else { break; } } } // 超时规则,当前时间 > 开始时间+1小时+设置额外超时时间 protected boolean isTimeout() { long currentTime = System.currentTimeMillis(); long endTime = m_startTime + m_duration + m_extraTime; return currentTime > endTime; }
// 具体的 Anlalyzer示例: DumpAnlalyzer.process
// 具体的 Anlalyzer示例: DumpAnlalyzer.process @Override public void process(MessageTree tree) { String domain = tree.getDomain(); if ("PhoenixAgent".equals(domain)) { return; } else { MessageId messageId = MessageId.parse(tree.getMessageId()); if (messageId.getVersion() == 2) { // 计算出当前时间范围, long time = tree.getMessage().getTimestamp(); long fixedTime = time - time % (TimeHelper.ONE_HOUR); long idTime = messageId.getTimestamp(); long duration = fixedTime - idTime; if (duration == 0 || duration == ONE_HOUR || duration == -ONE_HOUR) { m_bucketManager.storeMessage(tree, messageId); } else { m_serverStateManager.addPigeonTimeError(1); } } } } // 存储log消息到本地文件,并后续上传到hdfs @Override public void storeMessage(final MessageTree tree, final MessageId id) { boolean errorFlag = true; int hash = Math.abs((id.getDomain() + ‘-‘ + id.getIpAddress()).hashCode()); int index = (int) (hash % m_gzipThreads); MessageItem item = new MessageItem(tree, id); LinkedBlockingQueue<MessageItem> queue = m_messageQueues.get(index % (m_gzipThreads - 1)); boolean result = queue.offer(item); if (result) { errorFlag = false; } else { if (m_last.offer(item)) { errorFlag = false; } } if (errorFlag) { m_serverStateManager.addMessageDumpLoss(1); } logStorageState(tree); } // 每1000个消息添加一个messageDump=1000 protected void logStorageState(final MessageTree tree) { String domain = tree.getDomain(); int size = ((DefaultMessageTree) tree).getBuffer().readableBytes(); m_serverStateManager.addMessageSize(domain, size); if ((++m_total) % CatConstants.SUCCESS_COUNT == 0) { m_serverStateManager.addMessageDump(CatConstants.SUCCESS_COUNT); } }
// EventAnalyzer.process 处理event消息
@Override public void process(MessageTree tree) { String domain = tree.getDomain(); if (m_serverFilterConfigManager.validateDomain(domain)) { EventReport report = m_reportManager.getHourlyReport(getStartTime(), domain, true); Message message = tree.getMessage(); String ip = tree.getIpAddress(); if (message instanceof Transaction) { processTransaction(report, tree, (Transaction) message, ip); } else if (message instanceof Event) { processEvent(report, tree, (Event) message, ip); } } } // 循环处理多个transation private void processTransaction(EventReport report, MessageTree tree, Transaction t, String ip) { List<Message> children = t.getChildren(); for (Message child : children) { if (child instanceof Transaction) { processTransaction(report, tree, (Transaction) child, ip); } else if (child instanceof Event) { processEvent(report, tree, (Event) child, ip); } } } // StateAnalyzer.process 对cat的机器作展示 @Override protected void process(MessageTree tree) { String domain = tree.getDomain(); if (m_serverFilterConfigManager.validateDomain(domain)) { StateReport report = m_reportManager.getHourlyReport(getStartTime(), Constants.CAT, true); String ip = tree.getIpAddress(); Machine machine = report.findOrCreateMachine(NetworkInterfaceManager.INSTANCE.getLocalHostAddress()); machine.findOrCreateProcessDomain(domain).addIp(ip); } }
// 所有分析线程,由 Period 进行初始化启动所有的Analyzer备用
public void start() { SimpleDateFormat df = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss"); m_logger.info(String.format("Starting %s tasks in period [%s, %s]", m_tasks.size(), df.format(new Date(m_startTime)), df.format(new Date(m_endTime - 1)))); for (Entry<String, List<PeriodTask>> tasks : m_tasks.entrySet()) { List<PeriodTask> taskList = tasks.getValue(); for (int i = 0; i < taskList.size(); i++) { PeriodTask task = taskList.get(i); task.setIndex(i); Threads.forGroup("Cat-RealtimeConsumer").start(task); } } }
// 为保证高可用,使用 ChannelManager, 专门检查channel通道是否仍然存活,如果出问题,则发起重连。
@Override public void run() { while (m_active) { // make save message id index asyc m_idfactory.saveMark(); checkServerChanged(); ChannelFuture activeFuture = m_activeChannelHolder.getActiveFuture(); List<InetSocketAddress> serverAddresses = m_activeChannelHolder.getServerAddresses(); doubleCheckActiveServer(activeFuture); reconnectDefaultServer(activeFuture, serverAddresses); try { Thread.sleep(10 * 1000L); // check every 10 seconds } catch (InterruptedException e) { // ignore } } }
总结起来就几个东西:
1. 使用netty开启高性能的接收服务;
2. 使用队列进行保存消息;
3. 使用单独线程检测channel有效性,保证高可用;
4. 所有单小时的数据,保存在内存中,速度特别快;
等等,来个图展示下:。。。
task 运行过程
以上是关于应用监控CAT之cat-consumer源码阅读的主要内容,如果未能解决你的问题,请参考以下文章