关于微信的IM聊天消息序列号生成实践的讨论

Mike12138 · 5 年前

前一阵子阅读了本站的这篇文章

IM消息ID技术专题(一)：微信的海量IM聊天消息序列号生成实践（算法原理篇）

当下觉得蛮有趣的，最近就用Java自己实现看看

完成的结果如下

@Service
@Scope(value="singleton")
public class SequenceGenerator {
    private static final Logger LOGGER = LoggerFactory.getLogger(SequenceGenerator.class);

    private final float REDUNDANCY = 0.85F;

    @Autowired
    private JdbcTemplate jdbcTemplate;
    private int machineID;
    private long seq = 0;
    private long maxSeq = 100000;
    private int stepSeq = 100000;
    private long deathLine = (long)(maxSeq * REDUNDANCY);
    private boolean initFlag = false;
    private boolean updateFlag = false;
    private Map<String, Long> container = new HashMap();

    public void init(int machineID) throws DataAccessException {
        this.machineID = machineID;

        List<Map<String, Object>> rows = jdbcTemplate.queryForList("SELECT * FROM chat WHERE machine_id = ?", new Object[] { machineID });

        if (rows.size() == 0) {
            initFlag = true;
            String dateString = DateUtil.getNowString();

            jdbcTemplate.update(
                    "INSERT INTO chat (machine_id, max_value, step_value, created_at, updated_at) VALUES (?, ?, ?, ?, ?)",
                    new Object[] {machineID, maxSeq, stepSeq, dateString, dateString}
            );
            LOGGER.info(toString());

            return;
        }

        seq = (long) rows.get(0).get("max_value");
        maxSeq = (long) rows.get(0).get("max_value");
        stepSeq = (int) rows.get(0).get("step_value");

        updateSeq();

        initFlag = true;
    }

    private void updateSeq() throws DataAccessException {
        LOGGER.info("Update sequence starting...");

        String dateString = DateUtil.getNowString();

        long _maxSeq = maxSeq + stepSeq;
        long _deathLine = (long)(_maxSeq * REDUNDANCY);

        jdbcTemplate.update(
                "UPDATE chat SET max_value = ?, updated_at = ? WHERE machine_id = ?",
                new Object[] {_maxSeq, dateString, machineID}
        );

        maxSeq = _maxSeq;
        deathLine = _deathLine;

        LOGGER.info(toString());
        LOGGER.info("Update sequence finish");
    }

    public synchronized long getSeq(String groupID) {
        if (!initFlag) {
            throw  new RuntimeException("The generator not initialized yet");
        }

        long _seq = container.getOrDefault(groupID, seq);

        _seq++;

        if (_seq >= deathLine) {
            startUpdateSeq();
        }

        if (_seq >= maxSeq) {
            throw  new RuntimeException("The sequence has reached its limit");
        }

        container.put(groupID, _seq);

        return _seq;
    }

    private synchronized void startUpdateSeq() {
        if (updateFlag) {
            return;
        }

        updateFlag = true;

        new Thread(() -> {
            try {
                updateSeq();
            } catch (DataAccessException dae) {
                LOGGER.error(dae.getMessage());
            } finally {
                updateFlag = false;
            }
        }).start();
    }

    @Override
    public String toString() {
        return "SequenceGenerator{" +
                "machineID=" + machineID +
                ", seq=" + seq +
                ", maxSeq=" + maxSeq +
                ", stepSeq=" + stepSeq +
                ", deathLine=" + deathLine +
                '}';
    }
}

我不确定我实现的跟那篇文章的想法有没有对上

不过我不是以 uid 当区分，我用的是 gid(两人私聊也可以当作是一种群聊)

但我对以下这部分有点存疑

public synchronized long getSeq(String groupID) {
        if (!initFlag) {
            throw  new RuntimeException("The generator not initialized yet");
        }

        long _seq = container.getOrDefault(groupID, seq);

        _seq++;

        if (_seq >= deathLine) {
            startUpdateSeq();
        }

        if (_seq >= maxSeq) {
            throw  new RuntimeException("The sequence has reached its limit");
        }

        container.put(groupID, _seq);

        return _seq;
    }

为了确保不会有两个人拿到一样的 seq ，所以加上了 synchronized

但是这种方式对吞吐量来说好像反而是种拖累，不管外层开放给用户的方式是走 HTTP 还是 TCP

到了 getSeq 还是得要排队，不知道大神们有没有什么建议可以改善这部分的

又或是说我做错了?