细说HBase怎么完成一个Get操作 (client side)

 

 

源码解析基于HBase-0.20.6。

先看HTable类get()方法的code：

 

   HTable.java

  /**
   * Extracts certain cells from a given row.
   * @param get The object that specifies what data to fetch and from which row.
   * @return The data coming from the specified row, if it exists.  If the row
   * specified doesn't exist, the {@link Result} instance returned won't
   * contain any {@link KeyValue}, as indicated by {@link Result#isEmpty()}.
   * @throws IOException if a remote or network exception occurs.
   * @since 0.20.0
   */
  public Result get(final Get get) throws IOException {
    return connection.getRegionServerWithRetries(
        new ServerCallable<Result>(connection, tableName, get.getRow()) {
          public Result call() throws IOException {
            return server.get(location.getRegionInfo().getRegionName(), get);
          }
        }
    );
  }

 这段code 比较绕，但至少我们知道可以去查connection的getRegionServerWithRetries方法。那么connection是个什么东西呢？

这个玩意是定义在HTable里面的：

private final HConnection connection;

何时实例化的呢？在HTable的构造函数里面：

this.connection = HConnectionManager.getConnection(conf);

这个conf是一个HBaseConfiguration对象，是HTable构造函数的参数。OK，继续道HConnectionManager里面看看这个connection怎么来的吧：

HConnectionManager.java

  /**
   * Get the connection object for the instance specified by the configuration
   * If no current connection exists, create a new connection for that instance
   * @param conf
   * @return HConnection object for the instance specified by the configuration
   */
  public static HConnection getConnection(HBaseConfiguration conf) {
    TableServers connection;
    synchronized (HBASE_INSTANCES) {
      connection = HBASE_INSTANCES.get(conf);
      if (connection == null) {
        connection = new TableServers(conf);
        HBASE_INSTANCES.put(conf, connection);
      }
    }
    return connection;
  }

 

现在我们知道每一个conf对应一个connection，具体来说是TableServers类对象（实现了HConnection接口），所有的connections放在一个pool里。那么connection到底干嘛用呢？我们要看看HConnection这个接口的定义。

 

HConnection.java

/**
 * Cluster connection.
 * {@link HConnectionManager} manages instances of this class.
 */
public interface HConnection {
  /**
   * Retrieve ZooKeeperWrapper used by the connection.
   * @return ZooKeeperWrapper handle being used by the connection.
   * @throws IOException
   */
  public ZooKeeperWrapper getZooKeeperWrapper() throws IOException;

  /**
   * @return proxy connection to master server for this instance
   * @throws MasterNotRunningException
   */
  public HMasterInterface getMaster() throws MasterNotRunningException;

  /** @return - true if the master server is running */
  public boolean isMasterRunning();
  
  /**
   * Checks if <code>tableName</code> exists.
   * @param tableName Table to check.
   * @return True if table exists already.
   * @throws MasterNotRunningException
   */
  public boolean tableExists(final byte [] tableName)
  throws MasterNotRunningException;

  /**
   * A table that isTableEnabled == false and isTableDisabled == false
   * is possible. This happens when a table has a lot of regions
   * that must be processed.
   * @param tableName
   * @return true if the table is enabled, false otherwise
   * @throws IOException
   */
  public boolean isTableEnabled(byte[] tableName) throws IOException;
  
  /**
   * @param tableName
   * @return true if the table is disabled, false otherwise
   * @throws IOException
   */
  public boolean isTableDisabled(byte[] tableName) throws IOException;

  /**
   * @param tableName
   * @return true if all regions of the table are available, false otherwise
   * @throws IOException
   */
  public boolean isTableAvailable(byte[] tableName) throws IOException;

  /**
   * List all the userspace tables.  In other words, scan the META table.
   *
   * If we wanted this to be really fast, we could implement a special
   * catalog table that just contains table names and their descriptors.
   * Right now, it only exists as part of the META table's region info.
   *
   * @return - returns an array of HTableDescriptors 
   * @throws IOException
   */
  public HTableDescriptor[] listTables() throws IOException;
  
  /**
   * @param tableName
   * @return table metadata 
   * @throws IOException
   */
  public HTableDescriptor getHTableDescriptor(byte[] tableName)
  throws IOException;
  
  /**
   * Find the location of the region of <i>tableName</i> that <i>row</i>
   * lives in.
   * @param tableName name of the table <i>row</i> is in
   * @param row row key you're trying to find the region of
   * @return HRegionLocation that describes where to find the reigon in 
   * question
   * @throws IOException
   */
  public HRegionLocation locateRegion(final byte [] tableName,
      final byte [] row)
  throws IOException;
  
  /**
   * Find the location of the region of <i>tableName</i> that <i>row</i>
   * lives in, ignoring any value that might be in the cache.
   * @param tableName name of the table <i>row</i> is in
   * @param row row key you're trying to find the region of
   * @return HRegionLocation that describes where to find the reigon in 
   * question
   * @throws IOException
   */
  public HRegionLocation relocateRegion(final byte [] tableName,
      final byte [] row)
  throws IOException;  
  
  /** 
   * Establishes a connection to the region server at the specified address.
   * @param regionServer - the server to connect to
   * @return proxy for HRegionServer
   * @throws IOException
   */
  public HRegionInterface getHRegionConnection(HServerAddress regionServer)
  throws IOException;
  
  /** 
   * Establishes a connection to the region server at the specified address.
   * @param regionServer - the server to connect to
   * @param getMaster - do we check if master is alive
   * @return proxy for HRegionServer
   * @throws IOException
   */
  public HRegionInterface getHRegionConnection(
      HServerAddress regionServer, boolean getMaster)
  throws IOException;
  
  /**
   * Find region location hosting passed row
   * @param tableName
   * @param row Row to find.
   * @param reload If true do not use cache, otherwise bypass.
   * @return Location of row.
   * @throws IOException
   */
  HRegionLocation getRegionLocation(byte [] tableName, byte [] row,
    boolean reload)
  throws IOException;

  /**
   * Pass in a ServerCallable with your particular bit of logic defined and 
   * this method will manage the process of doing retries with timed waits 
   * and refinds of missing regions.
   *
   * @param <T> the type of the return value
   * @param callable
   * @return an object of type T
   * @throws IOException
   * @throws RuntimeException
   */
  public <T> T getRegionServerWithRetries(ServerCallable<T> callable) 
  throws IOException, RuntimeException;
  
  /**
   * Pass in a ServerCallable with your particular bit of logic defined and
   * this method will pass it to the defined region server.
   * @param <T> the type of the return value
   * @param callable
   * @return an object of type T
   * @throws IOException
   * @throws RuntimeException
   */
  public <T> T getRegionServerForWithoutRetries(ServerCallable<T> callable) 
  throws IOException, RuntimeException;
  
    
  /**
   * Process a batch of Puts. Does the retries.
   * @param list A batch of Puts to process.
   * @param tableName The name of the table
   * @return Count of committed Puts.  On fault, < list.size().
   * @throws IOException
   */
  public int processBatchOfRows(ArrayList<Put> list, byte[] tableName)
  throws IOException;

  /**
   * Process a batch of Deletes. Does the retries.
   * @param list A batch of Deletes to process.
   * @return Count of committed Deletes. On fault, < list.size().
   * @param tableName The name of the table
   * @throws IOException
   */
  public int processBatchOfDeletes(ArrayList<Delete> list, byte[] tableName)
  throws IOException;
}

 

上面的code是整个接口的定义，我们现在知道这玩意是封装了一些客户端查询处理请求，像put、delete这些封装在方法

public <T> T getRegionServerWithRetries(ServerCallable<T> callable) 里执行，put、delete等被封装在callable里面。这也就是为我们刚才在HTable.get()里看到的。

到这里要看TableServers.getRegionServerWithRetries(ServerCallable<T> callable)了，继续看code

public <T> T getRegionServerWithRetries(ServerCallable<T> callable) 
    throws IOException, RuntimeException {
      List<Throwable> exceptions = new ArrayList<Throwable>();
      for(int tries = 0; tries < numRetries; tries++) {
        try { 
	callable.instantiateServer(tries!=0); return callable.call();
              } catch (Throwable t) {
          t = translateException(t);
          exceptions.add(t);
          if (tries == numRetries - 1) {
            throw new RetriesExhaustedException(callable.getServerName(),
                callable.getRegionName(), callable.getRow(), tries, exceptions);
          }
        }
        try {
          Thread.sleep(getPauseTime(tries));
        } catch (InterruptedException e) {
          // continue
        }
      }
      return null;    
    }

 

比较核心的code就那两句，首先根据callable对象来完成一些定位ReginServer的工作，然后执行call来进行请求，这里要注意这个call方法是在最最最最开始的HTable.get里面的内部类里重写的。看ServerCallable类的一部分code：

public abstract class ServerCallable<T> implements Callable<T> {
  protected final HConnection connection;
  protected final byte [] tableName;
  protected final byte [] row;
  protected HRegionLocation location;
  protected HRegionInterface server;

  /**
   * @param connection
   * @param tableName
   * @param row
   */
  public ServerCallable(HConnection connection, byte [] tableName, byte [] row) {
    this.connection = connection;
    this.tableName = tableName;
    this.row = row;
  }
  
  /**
   * 
   * @param reload set this to true if connection should re-find the region
   * @throws IOException
   */
  public void instantiateServer(boolean reload) throws IOException {
    this.location = connection.getRegionLocation(tableName, row, reload);
    this.server = connection.getHRegionConnection(location.getServerAddress());
  }

 

所以一个ServerCallable对象包括tableName，row等，并且会通过构造函数传入一个connection引用，并且会调用该connection.getHRegionConnection方法来获取跟RegionServer打交道的一个handle（其实我也不知道称呼它啥了，不能叫connection吧，那就重复了，所以说HBase代码起的名字让我很ft，会误解）。

 

具体看怎么获得这个新玩意的：

HConnectinManager.java

   public HRegionInterface getHRegionConnection(
        HServerAddress regionServer, boolean getMaster) 
    throws IOException {
      if (getMaster) {
        getMaster();
      }
      HRegionInterface server;
      synchronized (this.servers) {
        // See if we already have a connection
        server = this.servers.get(regionServer.toString());
        if (server == null) { // Get a connection
          try {
            server = (HRegionInterface)HBaseRPC.waitForProxy(
                serverInterfaceClass, HBaseRPCProtocolVersion.versionID,
                regionServer.getInetSocketAddress(), this.conf, 
                this.maxRPCAttempts, this.rpcTimeout);
          } catch (RemoteException e) {
            throw RemoteExceptionHandler.decodeRemoteException(e);
          }
          this.servers.put(regionServer.toString(), server);
        }
      }
      return server;
    }

 

再挖下去看这个server怎么出来的（HBaseRPC类里面）：

  public static VersionedProtocol getProxy(Class<?> protocol,
      long clientVersion, InetSocketAddress addr, UserGroupInformation ticket,
      Configuration conf, SocketFactory factory)
  throws IOException {    
    VersionedProtocol proxy =
        (VersionedProtocol) Proxy.newProxyInstance(
            protocol.getClassLoader(), new Class[] { protocol },
            new Invoker(addr, ticket, conf, factory));
    long serverVersion = proxy.getProtocolVersion(protocol.getName(), 
                                                  clientVersion);
    if (serverVersion == clientVersion) {
      return proxy;
    }
    throw new VersionMismatch(protocol.getName(), clientVersion, 
                              serverVersion);
  }

 

 

这两部分code看出用到了java的动态代理机制，server是一个动态代理对象，实现了变量serverInterfaceClass指定的接口。在这里也就是HRegionInterface，也就是说server实现了该接口的内容。那么该接口定义哪些方法呢？

 

public interface HRegionInterface extends HBaseRPCProtocolVersion {
  /** 
   * Get metainfo about an HRegion
   * 
   * @param regionName name of the region
   * @return HRegionInfo object for region
   * @throws NotServingRegionException
   */
  public HRegionInfo getRegionInfo(final byte [] regionName)
  throws NotServingRegionException;
  

  /**
   * Return all the data for the row that matches <i>row</i> exactly, 
   * or the one that immediately preceeds it.
   * 
   * @param regionName region name
   * @param row row key
   * @param family Column family to look for row in.
   * @return map of values
   * @throws IOException
   */
  public Result getClosestRowBefore(final byte [] regionName,
    final byte [] row, final byte [] family)
  throws IOException;

  /**
   * 
   * @return the regions served by this regionserver
   */
  public HRegion [] getOnlineRegionsAsArray();
  
  /**
   * Perform Get operation.
   * @param regionName name of region to get from
   * @param get Get operation
   * @return Result
   * @throws IOException
   */
  public Result get(byte [] regionName, Get get) throws IOException;

  /**
   * Perform exists operation.
   * @param regionName name of region to get from
   * @param get Get operation describing cell to test
   * @return true if exists
   * @throws IOException
   */
  public boolean exists(byte [] regionName, Get get) throws IOException;

  /**
   * Put data into the specified region 
   * @param regionName
   * @param put the data to be put
   * @throws IOException
   */
  public void put(final byte [] regionName, final Put put)
  throws IOException;
  
  /**
   * Put an array of puts into the specified region
   * 
   * @param regionName
   * @param puts
   * @return The number of processed put's.  Returns -1 if all Puts
   * processed successfully.
   * @throws IOException
   */
  public int put(final byte[] regionName, final Put [] puts)
  throws IOException;

  /**
   * Deletes all the KeyValues that match those found in the Delete object, 
   * if their ts <= to the Delete. In case of a delete with a specific ts it
   * only deletes that specific KeyValue.
   * @param regionName
   * @param delete
   * @throws IOException
   */
  public void delete(final byte[] regionName, final Delete delete)
  throws IOException;

  /**
   * Put an array of deletes into the specified region
   * 
   * @param regionName
   * @param deletes
   * @return The number of processed deletes.  Returns -1 if all Deletes
   * processed successfully.
   * @throws IOException
   */
  public int delete(final byte[] regionName, final Delete [] deletes)
  throws IOException;

  /**
   * Atomically checks if a row/family/qualifier value match the expectedValue.
   * If it does, it adds the put.
   * 
   * @param regionName
   * @param row
   * @param family
   * @param qualifier
   * @param value the expected value
   * @param put
   * @throws IOException
   * @return true if the new put was execute, false otherwise
   */
  public boolean checkAndPut(final byte[] regionName, final byte [] row, 
      final byte [] family, final byte [] qualifier, final byte [] value,
      final Put put)
  throws IOException;
  
  /**
   * Atomically increments a column value. If the column value isn't long-like,
   * this could throw an exception.
   * 
   * @param regionName
   * @param row
   * @param family
   * @param qualifier
   * @param amount
   * @param writeToWAL whether to write the increment to the WAL
   * @return new incremented column value
   * @throws IOException
   */
  public long incrementColumnValue(byte [] regionName, byte [] row, 
      byte [] family, byte [] qualifier, long amount, boolean writeToWAL)
  throws IOException;
  
  
  //
  // remote scanner interface
  //

  /**
   * Opens a remote scanner with a RowFilter.
   * 
   * @param regionName name of region to scan
   * @param scan configured scan object
   * @return scannerId scanner identifier used in other calls
   * @throws IOException
   */
  public long openScanner(final byte [] regionName, final Scan scan)
  throws IOException;
  
  /**
   * Get the next set of values
   * @param scannerId clientId passed to openScanner
   * @return map of values; returns null if no results.
   * @throws IOException
   */
  public Result next(long scannerId) throws IOException;
  
  /**
   * Get the next set of values
   * @param scannerId clientId passed to openScanner
   * @param numberOfRows the number of rows to fetch
   * @return Array of Results (map of values); array is empty if done with this
   * region and null if we are NOT to go to the next region (happens when a
   * filter rules that the scan is done).
   * @throws IOException
   */
  public Result [] next(long scannerId, int numberOfRows) throws IOException;
  
  /**
   * Close a scanner
   * 
   * @param scannerId the scanner id returned by openScanner
   * @throws IOException
   */
  public void close(long scannerId) throws IOException;

  /**
   * Opens a remote row lock.
   *
   * @param regionName name of region
   * @param row row to lock
   * @return lockId lock identifier
   * @throws IOException
   */
  public long lockRow(final byte [] regionName, final byte [] row)
  throws IOException;

  /**
   * Releases a remote row lock.
   *
   * @param regionName
   * @param lockId the lock id returned by lockRow
   * @throws IOException
   */
  public void unlockRow(final byte [] regionName, final long lockId)
  throws IOException;
  
  
  /**
   * Method used when a master is taking the place of another failed one.
   * @return All regions assigned on this region server
   * @throws IOException
   */
  public HRegionInfo[] getRegionsAssignment() throws IOException;
  
  /**
   * Method used when a master is taking the place of another failed one.
   * @return The HSI
   * @throws IOException
   */
  public HServerInfo getHServerInfo() throws IOException;
}

 

可以看出HRegionInterface是定义了具体的向RegionServer查询的方法。

 

现在回过头来，当server这个动态代理对象实例化后，经过ServerCallable.call() 最后会调到server.get()。按照java的代理机制，又会传递到我们在构造这个动态代理对象时候传进去的new Invoker(addr, ticket, conf, factory))对象去执行具体的方法。

简单的说，这个Invoker对象使用HBase的RPC客户端跟RegionServer通信完成请求以及结果接收等等。

看看这个RPC客户端长什么样吧：

public Invoker(InetSocketAddress address, UserGroupInformation ticket, 
                   Configuration conf, SocketFactory factory) {
      this.address = address;
      this.ticket = ticket;
      this.client = CLIENTS.getClient(conf, factory); //client就是RPC客户端
    }

 这个client是HBaseClient类的对象，这个HBaseClient类就是HBase中用来做RPC的客户端类。在这里HBaseClient也做了一个pool机制，不理解。。。code里面的注释如下：

      // Construct & cache client.  The configuration is only used for timeout,
      // and Clients have connection pools.  So we can either (a) lose some
      // connection pooling and leak sockets, or (b) use the same timeout for all
      // configurations.  Since the IPC is usually intended globally, not
      // per-job, we choose (a).

 

继续说下去，看这么一个client怎么完成最后的请求：

 public Writable call(Writable param, InetSocketAddress addr, 
                       UserGroupInformation ticket)  
                       throws IOException {
    Call call = new Call(param);
    Connection connection = getConnection(addr, ticket, call);
    connection.sendParam(call);                 // send the parameter
    boolean interrupted = false;
    synchronized (call) {
      while (!call.done) {
        try {
          call.wait();                           // wait for the result
        } catch (InterruptedException ie) {
          // save the fact that we were interrupted
          interrupted = true;
        }
      }

      if (interrupted) {
        // set the interrupt flag now that we are done waiting
        Thread.currentThread().interrupt();
      }

      if (call.error != null) {
        if (call.error instanceof RemoteException) {
          call.error.fillInStackTrace();
          throw call.error;
        }
        // local exception
        throw wrapException(addr, call.error);
      }
      return call.value;
    }
  }

 

又见connection，这次的connection可是用来发送接收数据用的thread了。从getConnection(addr, ticket, call)推断又是一个pool，果不其然：

 

 /** Get a connection from the pool, or create a new one and add it to the
   * pool.  Connections to a given host/port are reused. */
  private Connection getConnection(InetSocketAddress addr, 
                                   UserGroupInformation ticket,
                                   Call call)
                                   throws IOException {
    if (!running.get()) {
      // the client is stopped
      throw new IOException("The client is stopped");
    }
    Connection connection;
    /* we could avoid this allocation for each RPC by having a  
     * connectionsId object and with set() method. We need to manage the
     * refs for keys in HashMap properly. For now its ok.
     */
    ConnectionId remoteId = new ConnectionId(addr, ticket);
    do {
      synchronized (connections) {
        connection = connections.get(remoteId);
        if (connection == null) {
          connection = new Connection(remoteId);
          connections.put(remoteId, connection);
        }
      }
    } while (!connection.addCall(call));
    
    //we don't invoke the method below inside "synchronized (connections)"
    //block above. The reason for that is if the server happens to be slow,
    //it will take longer to establish a connection and that will slow the
    //entire system down.
    connection.setupIOstreams();
    return connection;
  }

 

也就是说，只要所要查询的RegionServer的addr和用户组信息一样，就会共享一个connection。connection拿到后会将当前call放进自己内部的一个队列里（维护着call的id=》call的一个映射），当call完成后会更新call的状态（主要是否完成这么一个标志Call.done以及将请求结果填充在Call.value里）。

 

好了现在的情形是，现在看connection如何发送请求数据。

  /** Initiates a call by sending the parameter to the remote server.
     * Note: this is not called from the Connection thread, but by other
     * threads.
     * @param call
     */
    public void sendParam(Call call) {
      if (shouldCloseConnection.get()) {
        return;
      }

      DataOutputBuffer d=null;
      try {
        synchronized (this.out) {
          if (LOG.isDebugEnabled())
            LOG.debug(getName() + " sending #" + call.id);
          
          //for serializing the
          //data to be written
          d = new DataOutputBuffer();
          d.writeInt(call.id);
          call.param.write(d);
          byte[] data = d.getData();
          int dataLength = d.getLength();
          out.writeInt(dataLength);      //first put the data length
          out.write(data, 0, dataLength);//write the data
          out.flush();
        }
      } catch(IOException e) {
        markClosed(e);
      } finally {
        //the buffer is just an in-memory buffer, but it is still polite to
        // close early
        IOUtils.closeStream(d);
      }
    }  

 

从code里面看出，请求发送是synchronized，所以会有上一篇日志里提到的问题。

HBase客户端的code先看到这里吧。

 下面这个图帮助理解一下上面各种pool

 

评论

1 楼 yintaoxue 2012-02-25