RocketMQ设计之故障规避机制

NameServer为了简化和客户端通信，发现Broker故障时并不会立即通知客户端。故障规避机制就是用来解决当Broker出现故障，Producer不能及时感知而导致消息发送失败的问题。默认不开启，如果开启，消息发送失败的时候会将失败的Broker暂时排除在队列选择列表外

MQFaultStrategy类的：

public class MQFaultStrategy {
    private final static InternalLogger log = ClientLogger.getLog();
    private final LatencyFaultTolerance<String> latencyFaultTolerance = new LatencyFaultToleranceImpl();

    private boolean sendLatencyFaultEnable = false;

    private long[] latencyMax = {50L, 100L, 550L, 1000L, 2000L, 3000L, 15000L};
    private long[] notAvailableDuration = {0L, 0L, 30000L, 60000L, 120000L, 180000L, 600000L};

    public long[] getNotAvailableDuration() {
        return notAvailableDuration;
    }

    public void setNotAvailableDuration(final long[] notAvailableDuration) {
        this.notAvailableDuration = notAvailableDuration;
    }

    public long[] getLatencyMax() {
        return latencyMax;
    }

    public void setLatencyMax(final long[] latencyMax) {
        this.latencyMax = latencyMax;
    }

    public boolean isSendLatencyFaultEnable() {
        return sendLatencyFaultEnable;
    }

    public void setSendLatencyFaultEnable(final boolean sendLatencyFaultEnable) {
        this.sendLatencyFaultEnable = sendLatencyFaultEnable;
    }

    public MessageQueue selectOneMessageQueue(final TopicPublishInfo tpInfo, final String lastBrokerName) {
        //是否开启故障延迟机制
        if (this.sendLatencyFaultEnable) {
            try {
                int index = tpInfo.getSendWhichQueue().getAndIncrement();
                for (int i = 0; i < tpInfo.getMessageQueueList().size(); i++) {
                    int pos = Math.abs(index++) % tpInfo.getMessageQueueList().size();
                    if (pos < 0)
                        pos = 0;
                    MessageQueue mq = tpInfo.getMessageQueueList().get(pos);
                    //判断Queue是否可用
                    if (latencyFaultTolerance.isAvailable(mq.getBrokerName())) {
                        if (null == lastBrokerName || mq.getBrokerName().equals(lastBrokerName))
                            return mq;
                    }
                }

                final String notBestBroker = latencyFaultTolerance.pickOneAtLeast();
                int writeQueueNums = tpInfo.getQueueIdByBroker(notBestBroker);
                if (writeQueueNums > 0) {
                    final MessageQueue mq = tpInfo.selectOneMessageQueue();
                    if (notBestBroker != null) {
                        mq.setBrokerName(notBestBroker);
                        mq.setQueueId(tpInfo.getSendWhichQueue().getAndIncrement() % writeQueueNums);
                    }
                    return mq;
                } else {
                    latencyFaultTolerance.remove(notBestBroker);
                }
            } catch (Exception e) {
                log.error("Error occurred when selecting message queue", e);
            }

            return tpInfo.selectOneMessageQueue();
        }

        //默认轮询
        return tpInfo.selectOneMessageQueue(lastBrokerName);
    }

    public void updateFaultItem(final String brokerName, final long currentLatency, boolean isolation) {
        if (this.sendLatencyFaultEnable) {
            long duration = computeNotAvailableDuration(isolation ? 30000 : currentLatency);
            this.latencyFaultTolerance.updateFaultItem(brokerName, currentLatency, duration);
        }
    }

    private long computeNotAvailableDuration(final long currentLatency) {
        for (int i = latencyMax.length - 1; i >= 0; i--) {
            if (currentLatency >= latencyMax[i])
                return this.notAvailableDuration[i];
        }

        return 0;
    }
}

在选择查找路由时，选择消息队列的关键步骤：

先按轮询算法选择一个消息队列
从故障列表判断该消息队列是否可用

LatencyFaultToleranceImpl中判断是否可用：

@Override
public boolean isAvailable(final String name) {
    final FaultItem faultItem = this.faultItemTable.get(name);
    if (faultItem != null) {
        return faultItem.isAvailable();
    }
    return true;
}

public boolean isAvailable() {
            return (System.currentTimeMillis() - startTimestamp) >= 0;
        }

判断是否在故障列表中，不在故障列表中代表可用。
在故障列表中判断当前时间是否大于等于故障规避的开始时间startTimestamp

在消息发送结束后和发送出现异常时调用updateFaultItem()方法来更新故障列表，computeNotAvailableDuration()根据响应时间来计算故障周期时长，响应时间越长故障周期越长。网络异常、Broker异常、客户端异常都是固定响应时长30s，它们故障周期时长为10分钟。消息发送成功或线程中断异常响应时间在100毫秒以内，故障周期时长为0。

LatencyFaultToleranceImpl类的updateFaultItem方法：

@Override
public void updateFaultItem(final String name, final long currentLatency, final long notAvailableDuration) {
    FaultItem old = this.faultItemTable.get(name);
    if (null == old) {
        final FaultItem faultItem = new FaultItem(name);
        faultItem.setCurrentLatency(currentLatency);
        faultItem.setStartTimestamp(System.currentTimeMillis() + notAvailableDuration);

        //加入故障列表
        old = this.faultItemTable.putIfAbsent(name, faultItem);
        if (old != null) {
            old.setCurrentLatency(currentLatency);
            old.setStartTimestamp(System.currentTimeMillis() + notAvailableDuration);
        }
    } else {
        old.setCurrentLatency(currentLatency);
        old.setStartTimestamp(System.currentTimeMillis() + notAvailableDuration);
    }
}

FaultItem存储Broker名称、响应时长、故障规避开始时间，最重要的是故障规避开始时间，用来判断Queue是否可用

频道导航

RocketMQ设计之故障规避机制

您可能感兴趣的文章: