可用性问题

[lipengming]

RedisSessionDao完全依赖于redis，如果redis挂掉了，或者本来redis就没有启动，那么该组件会导致项目全部不可用。所以考虑一下可用性。

1、确保单机一定可用：

private static Logger logger = LoggerFactory.getLogger(SessionCache.class);
private String keyPrefix = "shiro_redis_session:";

private final CacheClusterClient cacheClusterClient;
private final int timeOut ;

public SessionCache(CacheClusterClient cacheClusterClient, int timeOut) {
    this.cacheClusterClient = cacheClusterClient;
    this.timeOut = timeOut;
}

/**
 * 读取session
 *
 * @param sessionId
 * @return
 */
public Session readSession(Serializable sessionId) {
    try {
        return sessionCache.get(sessionId);
    } catch (Exception e) {
        return null;
    }
}

/**
 * 保存session
 *
 * @param session
 */
public void saveSession(Session session) {
    if(session == null || session.getId() == null){
        logger.error("session or session id is null");
        return;
    }
    byte[] key = getKey(session.getId());
    byte[] value = SerializeUtil.serialize(session);
    session.setTimeout(timeOut * 1000);
    try{
        cacheClusterClient.set(key, value);
        cacheClusterClient.expire(key,timeOut);
    } catch (Exception e) {
        logger.error("###########Redis Connect Failed###########", e);
        //save to local
        sessionCache.put(session.getId(),session);
    }
}

/**
 * 作废session
 *
 * @param sessionId
 */
public void invalid(Serializable sessionId) {
    sessionCache.invalidate(sessionId);
    try{
        cacheClusterClient.del(getKey(sessionId));
    } catch (Exception e) {
        logger.error("###########Redis Connect Failed###########", e);
    }
}

/**
 * 所有活动的session
 *
 * @return
 */
public Collection<Session> getActive() {
    return sessionCache.asMap().values();
}

/////////////////////////////////\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
/////////////////////////////////\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
/////////////////////////////////\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\

private LoadingCache<Serializable,Session> sessionCache = CacheBuilder.newBuilder()
        .maximumSize(100)//设置缓存最大容量为100
        .expireAfterWrite(10, TimeUnit.MINUTES)//设置写缓存后30分钟过期，过期后会触发removalListener
        .concurrencyLevel(4)//并发级别为4，并发级别是指可以同时写缓存的线程数
        .removalListener(new RemovalListener<Object, Object>() {//这段代码可以移除
            @Override
            public void onRemoval(RemovalNotification<Object, Object> notification) {
                logger.debug("session was removed with key" + notification.getKey() + " cause is :" + notification.getCause());
            }
        })
        .build(new CacheLoader<Serializable,Session>() {//如果缓存中根据Key没有取到值，那么执行这里面的操作了
            @Override
            public Session load(Serializable key) throws Exception {
                return readRedis(key);//当返回null的时候会抛异常，CacheLoader.InvalidCacheLoadException
            }
        });


private byte[] getKey(Serializable sessionId){
    String preKey = this.keyPrefix + sessionId;
    return SerializeUtil.serialize(preKey);
}

private Session readRedis(Serializable sessionId) {
    if(sessionId == null){
        logger.error("session id is null");
        return null;
    }
    try {
        byte[] key = getKey(sessionId);
        byte[] value = cacheClusterClient.get(key);
        if(value == null) return null;
        Session s = (Session) SerializeUtil.deserialize(value);
        return s;
    } catch (Exception e) {
        logger.error("Redis Connect Failed! Read Err!",e);
    }
    return null;
}

2、提升redis组件本身的可用性：

这个可以 redis的一些cluster的方案，这里不做衍生。

[alexxiyang]

建议使用Redis sentinel或者Redis cluster来解决可用性问题，保持Shiro-redis的简单，简单的代码比较robust. 也容易阅读