RocketMQ-broker狀態管理及資料統計

在RocketMQ中，狀態管理有BrokerStatsManager，ConsumerStatsManager，FilterServerStatsManager，其實現的方式都是一樣的。

這邊就拿BrokerStatsManager做介紹

一個StatsItemSet，代表著一項資料統計指標，這個指標定義了各個key的統計單項StatsItem，具體的值和次數使用原子類表示

BrokerStatsManager支援這些統一指標，同時每一個指標對應一個StateItemSet，裡面包含了各個topic和group的資訊

    public static
 
 final String TOPIC_PUT_NUMS = "TOPIC_PUT_NUMS";
    public static final String TOPIC_PUT_SIZE = "TOPIC_PUT_SIZE";
    public static final String GROUP_GET_NUMS = "GROUP_GET_NUMS";
    public static final String GROUP_GET_SIZE = "GROUP_GET_SIZE";
    public static final String SNDBCK_PUT_NUMS = "SNDBCK_PUT_NUMS";
    
 
public static final String BROKER_PUT_NUMS = "BROKER_PUT_NUMS";
    public static final String BROKER_GET_NUMS = "BROKER_GET_NUMS";
    public static final String GROUP_GET_FROM_DISK_NUMS = "GROUP_GET_FROM_DISK_NUMS";
    public static final String GROUP_GET_FROM_DISK_SIZE = "GROUP_GET_FROM_DISK_SIZE";
    
 
public static final String BROKER_GET_FROM_DISK_NUMS = "BROKER_GET_FROM_DISK_NUMS";
    public static final String BROKER_GET_FROM_DISK_SIZE = "BROKER_GET_FROM_DISK_SIZE";
    // For commercial
    public static final String COMMERCIAL_SEND_TIMES = "COMMERCIAL_SEND_TIMES";
    public static final String COMMERCIAL_SNDBCK_TIMES = "COMMERCIAL_SNDBCK_TIMES";
    public static final String COMMERCIAL_RCV_TIMES = "COMMERCIAL_RCV_TIMES";
    public static final String COMMERCIAL_RCV_EPOLLS = "COMMERCIAL_RCV_EPOLLS";
    public static final String COMMERCIAL_SEND_SIZE = "COMMERCIAL_SEND_SIZE";
    public static final String COMMERCIAL_RCV_SIZE = "COMMERCIAL_RCV_SIZE";
    public static final String COMMERCIAL_PERM_FAILURES = "COMMERCIAL_PERM_FAILURES";



this.statsTable.put(TOPIC_PUT_NUMS, new StatsItemSet(TOPIC_PUT_NUMS, this.scheduledExecutorService, log));
this.statsTable.put(TOPIC_PUT_SIZE, new StatsItemSet(TOPIC_PUT_SIZE, this.scheduledExecutorService, log));
this.statsTable.put(GROUP_GET_NUMS, new StatsItemSet(GROUP_GET_NUMS, this.scheduledExecutorService, log));
this.statsTable.put(GROUP_GET_SIZE, new StatsItemSet(GROUP_GET_SIZE, this.scheduledExecutorService, log));
this.statsTable.put(GROUP_GET_LATENCY, new StatsItemSet(GROUP_GET_LATENCY, this.scheduledExecutorService, log));
this.statsTable.put(SNDBCK_PUT_NUMS, new StatsItemSet(SNDBCK_PUT_NUMS, this.scheduledExecutorService, log));
this.statsTable.put(BROKER_PUT_NUMS, new StatsItemSet(BROKER_PUT_NUMS, this.scheduledExecutorService, log));
this.statsTable.put(BROKER_GET_NUMS, new StatsItemSet(BROKER_GET_NUMS, this.scheduledExecutorService, log));
this.statsTable.put(GROUP_GET_FROM_DISK_NUMS, new StatsItemSet(GROUP_GET_FROM_DISK_NUMS, this.scheduledExecutorService, log));
this.statsTable.put(GROUP_GET_FROM_DISK_SIZE, new StatsItemSet(GROUP_GET_FROM_DISK_SIZE, this.scheduledExecutorService, log));
this.statsTable.put(BROKER_GET_FROM_DISK_NUMS, new StatsItemSet(BROKER_GET_FROM_DISK_NUMS, this.scheduledExecutorService, log));
this.statsTable.put(BROKER_GET_FROM_DISK_SIZE, new StatsItemSet(BROKER_GET_FROM_DISK_SIZE, this.scheduledExecutorService, log));

this.statsTable.put(COMMERCIAL_SEND_TIMES, new StatsItemSet(COMMERCIAL_SEND_TIMES, this.commercialExecutor, COMMERCIAL_LOG));
this.statsTable.put(COMMERCIAL_RCV_TIMES, new StatsItemSet(COMMERCIAL_RCV_TIMES, this.commercialExecutor, COMMERCIAL_LOG));
this.statsTable.put(COMMERCIAL_SEND_SIZE, new StatsItemSet(COMMERCIAL_SEND_SIZE, this.commercialExecutor, COMMERCIAL_LOG));
this.statsTable.put(COMMERCIAL_RCV_SIZE, new StatsItemSet(COMMERCIAL_RCV_SIZE, this.commercialExecutor, COMMERCIAL_LOG));
this.statsTable.put(COMMERCIAL_RCV_EPOLLS, new StatsItemSet(COMMERCIAL_RCV_EPOLLS, this.commercialExecutor, COMMERCIAL_LOG));
this.statsTable.put(COMMERCIAL_SNDBCK_TIMES, new StatsItemSet(COMMERCIAL_SNDBCK_TIMES, this.commercialExecutor, COMMERCIAL_LOG));
this.statsTable.put(COMMERCIAL_PERM_FAILURES, new StatsItemSet(COMMERCIAL_PERM_FAILURES, this.commercialExecutor, COMMERCIAL_LOG));

最後這些指標都會儲存在 private final HashMap<String, StatsItemSet> statsTable = new HashMap<String, StatsItemSet>();

一個指標StatsItemSet裡面有如下屬性

private final ConcurrentMap<String/* key */, StatsItem> statsItemTable =
        new ConcurrentHashMap<String, StatsItem>(128);

    private final String statsName;
    private final ScheduledExecutorService scheduledExecutorService;
    private final Logger log;

比如一個TOPIC_PUT_NUMS的指標，在其statsItemTable 包含了各個topic的message數量和存放次數。其中key就是topic。

然後會初始化一些定時任務，比如會在每隔10s把當前的狀態封裝之後放在csListMinute，最多放6個統計記錄，然後在一分鐘內列印結果，格式如下

"TOPIC_PUT_NUMS [topicname] Stats In One Minute, SUM: %d TPS: %.2f AVGPT: %.2

再次之前會進行簡單計算程式碼如下

  private static StatsSnapshot computeStatsData(final LinkedList<CallSnapshot> csList) {
        StatsSnapshot statsSnapshot = new StatsSnapshot();
        synchronized (csList) {
            double tps = 0;
            double avgpt = 0;
            long sum = 0;
            if (!csList.isEmpty()) {
                CallSnapshot first = csList.getFirst();
                CallSnapshot last = csList.getLast();
                // 比如統計前後的值差距
                sum = last.getValue() - first.getValue();
                // last.getTimestamp() - first.getTimestamp()代表的是統計的時候時間差
                tps = (sum * 1000.0d) / (last.getTimestamp() - first.getTimestamp());

                // 次數差距
                long timesDiff = last.getTimes() - first.getTimes();
                if (timesDiff > 0) {
                    // 平均每次的達到的數量級   ，在TOPIC_PUT_NUMS中表示 每次在在該topic中平均寫入avgpt個message
                    avgpt = (sum * 1.0d) / timesDiff;
                }
            }

            statsSnapshot.setSum(sum);
            statsSnapshot.setTps(tps);
            statsSnapshot.setAvgpt(avgpt);
        }

        return statsSnapshot;
    }

每隔一小時，一天的統計結果處理邏輯雷同。