本文共 7530 字，大约阅读时间需要 25 分钟。

最近项目需要，要求做一个远程服务中心，为各客户端提供远程存取接口，压力基本上是每个客户端平均每秒300次请求，估计大约有10个客户端。 我考虑实现的方案是: 客户端写数据时直接写到本地ehcache缓存，然后同时发送数据到memcached缓存，最后通过远程服务接口定时同步数据到后台数据库。 客户端读数据首先从本地ehcache查找，然后再从memcached获取，最后通过远程数据接口查找数据 那么，此方案的一些需要考虑的问题就是： 1.memcached是否可以支撑这么大的压力 2.memcahced繁忙或down掉是否会影响client端线程阻塞或堆积到下一秒，造成压力 之前也通过测过Java client的效率，开10个线程，基本可以达到 2000~3000每秒。 首先分析一下Java client 启动时的部分代码 Memcached 支持直接设置多个servers属性 来实现多个memcahced均衡，对应还有一个属性是weights，字面意思就是权重，分析了一下代码，和我想的是一样的 启动memcached的代码通常是这样的

Java代码 SockIOPool pool = SockIOPool.getInstance(poolname);    pool.setServers(servers);    pool.setWeights(weights);    pool.setInitConn(initConn);    pool.setMinConn(minConn);    pool.setMaxConn(maxConn);    pool.setMaxIdle(maxIdle);    pool.setMaxBusyTime(maxBusyTime);    pool.setMaintSleep(maintSleep);    pool.setSocketTO(socketTO);    pool.setSocketConnectTO(socketConnectTO);    pool.setNagle(nagle);    pool.setHashingAlg(SockIOPool.NEW_COMPAT_HASH);    pool.initialize();    MemCachedClient client = new MemCachedClient(poolname);   SockIOPool pool = SockIOPool.getInstance(poolname);pool.setServers(servers);pool.setWeights(weights);pool.setInitConn(initConn);pool.setMinConn(minConn);pool.setMaxConn(maxConn);pool.setMaxIdle(maxIdle);pool.setMaxBusyTime(maxBusyTime);pool.setMaintSleep(maintSleep);pool.setSocketTO(socketTO);pool.setSocketConnectTO(socketConnectTO);pool.setNagle(nagle);pool.setHashingAlg(SockIOPool.NEW_COMPAT_HASH);pool.initialize();MemCachedClient client = new MemCachedClient(poolname);servers 和 weights 都是一个数组，就是说可以同时设置多个server 然后看一下 pool.initialize() 做了什么 Java代码 availPool   = new HashMap<String,Map<SockIO,Long>>( servers.length * initConn );    busyPool    = new HashMap<String,Map<SockIO,Long>>( servers.length * initConn );    deadPool    = new IdentityHashMap<SockIO,Integer>();    hostDeadDur = new HashMap<String,Long>();    hostDead    = new HashMap<String,Date>();    maxCreate   = (poolMultiplier > minConn) ? minConn : minConn / poolMultiplier;   // only create up to maxCreate connections at once    if ( log.isDebugEnabled() ) {        log.debug( "++++ initializing pool with following settings:" );        log.debug( "++++ initial size: " + initConn );        log.debug( "++++ min spare   : " + minConn );        log.debug( "++++ max spare   : " + maxConn );    }    // if servers is not set, or it empty, then    // throw a runtime exception    if ( servers == null || servers.length <= 0 ) {        log.error( "++++ trying to initialize with no servers" );        throw new IllegalStateException( "++++ trying to initialize with no servers" );    }        // initalize our internal hashing structures            if ( this.hashingAlg == CONSISTENT_HASH )            populateConsistentBuckets();        else               populateBuckets();   availPool   = new HashMap<String,Map<SockIO,Long>>( servers.length * initConn );busyPool    = new HashMap<String,Map<SockIO,Long>>( servers.length * initConn );deadPool    = new IdentityHashMap<SockIO,Integer>();hostDeadDur = new HashMap<String,Long>();hostDead    = new HashMap<String,Date>();maxCreate   = (poolMultiplier > minConn) ? minConn : minConn / poolMultiplier; // only create up to maxCreate connections at onceif ( log.isDebugEnabled() ) {  log.debug( "++++ initializing pool with following settings:" ); log.debug( "++++ initial size: " + initConn ); log.debug( "++++ min spare   : " + minConn ); log.debug( "++++ max spare   : " + maxConn );}// if servers is not set, or it empty, then// throw a runtime exceptionif ( servers == null || servers.length <= 0 ) {  log.error( "++++ trying to initialize with no servers" ); throw new IllegalStateException( "++++ trying to initialize with no servers" );} // initalize our internal hashing structures        if ( this.hashingAlg == CONSISTENT_HASH )  populateConsistentBuckets(); else         populateBuckets();看到这里就是开辟一些连接池的空间，然后调用了根据我们选择的hash 算法 执行populateBuckets();或者populateConsistentBuckets(); hash算法共有4种 Java代码 // native String.hashCode();    public static final int NATIVE_HASH     = 0;        // original compatibility hashing algorithm (works with other clients)    public static final int OLD_COMPAT_HASH = 1;    // new CRC32 based compatibility hashing algorithm (works with other clients)       public static final int NEW_COMPAT_HASH = 2;    // MD5 Based -- Stops thrashing when a server added or removed    public static final int CONSISTENT_HASH = 3;           // native String.hashCode();public static final int NATIVE_HASH     = 0; // original compatibility hashing algorithm (works with other clients)public static final int OLD_COMPAT_HASH = 1;// new CRC32 based compatibility hashing algorithm (works with other clients) public static final int NEW_COMPAT_HASH = 2;// MD5 Based -- Stops thrashing when a server added or removedpublic static final int CONSISTENT_HASH = 3;  我们通常用的是 NEW_COMPAT_HASH，这个保证可以wokrs with other clients 所以看一下populateBuckets()做了什么 Java代码 this.buckets = new ArrayList<String>();    for ( int i = 0; i < servers.length; i++ ) {      if ( this.weights != null && this.weights.length > i ) {         for ( int k = 0; k < this.weights[i].intValue(); k++ ) {         this.buckets.add( servers[i] );         if ( log.isDebugEnabled() )            log.debug( "++++ added " + servers[i] + " to server bucket" );         }      }      else {         this.buckets.add( servers[i] );      }         // create initial connections      for ( int j = 0; j < initConn; j++ ) {              SockIO socket = createSocket( servers[i] );          if ( socket == null ) {            break;              }              addSocketToPool( availPool, servers[i], socket );      }    }   this.buckets = new ArrayList<String>();for ( int i = 0; i < servers.length; i++ ) {   if ( this.weights != null && this.weights.length > i ) {      for ( int k = 0; k < this.weights[i].intValue(); k++ ) {   this.buckets.add( servers[i] );  if ( log.isDebugEnabled() )     log.debug( "++++ added " + servers[i] + " to server bucket" );     }  }  else {      this.buckets.add( servers[i] );  }     // create initial connections  for ( int j = 0; j < initConn; j++ ) {           SockIO socket = createSocket( servers[i] );   if ( socket == null ) {   break;          }          addSocketToPool( availPool, servers[i], socket );  }}假如我们设置的servers是 192.168.0.1:44444和192.168.0.2:22222 然后我们设置了weights是 5和3 那么 buckets list的值最终会是 [ 192.168.0.1:44444, 192.168.0.1:44444, 192.168.0.1:44444, 192.168.0.1:44444, 192.168.0.1:44444, 192.168.0.2:22222, 192.168.0.2:22222. 192.168.0.2:22222. ] 然后就开始根据initCon初始连接数按servers分别创建socket 那么究竟这个buckets做什么用呢？ 在我们使用set存放对象时会调用 Java代码 SockIOPool.SockIO sock = pool.getSock( key, hashCode );   SockIOPool.SockIO sock = pool.getSock( key, hashCode );看一看pool.getSock的代码 Java代码 // get initial bucket    long bucket = getBucket( key, hashCode );    String server = ( this.hashingAlg == CONSISTENT_HASH )            ? consistentBuckets.get( bucket )            : buckets.get( (int)bucket );   // get initial bucketlong bucket = getBucket( key, hashCode );String server = ( this.hashingAlg == CONSISTENT_HASH )  ? consistentBuckets.get( bucket )  : buckets.get( (int)bucket );其中有段代码是这样的，看看getBucket Java代码 private long getBucket( String key, Integer hashCode ) {        long hc = getHash( key, hashCode );        if ( this.hashingAlg == CONSISTENT_HASH ) {            return findPointFor( hc );        }        else {             long bucket = hc % buckets.size();            if ( bucket < 0 ) bucket *= -1;                return bucket;        }    }   private long getBucket( String key, Integer hashCode ) {  long hc = getHash( key, hashCode ); if ( this.hashingAlg == CONSISTENT_HASH ) {   return findPointFor( hc ); } else {       long bucket = hc % buckets.size();  if ( bucket < 0 ) bucket *= -1;   return bucket; }}先不管key和hashCode，我们看到首先算出一个hc值后会直接做hc%buckets.size()实际上就是根据buckets的数量散列，最终值一定是buckets.size()范围里的一个值 然后最终server值就根据buckets.get( (int)bucket )得到，那么假如我们得到bucket是3，则参照上面buckets 里的值，得到 list.get(3)=192.168.0.1:44444 所以会根据weight设置的值的不同得到不同的server ，如果 weights设置10:1 那buckets里就是10个相同的server和另一个不同的，将来散列得到的server很大可能性是servers里设置的第一个server。

本文来自CSDN博客，转载请标明出处：

转载地址：https://blog.csdn.net/yuezu1026/article/details/4540597 如侵犯您的版权，请留言回复原文章的地址，我们会给您删除此文章，给您带来不便请您谅解！