Solr4.8.0源码分析(3)之index的线程池管理

Solr4.8.0源码分析(3)之index的线程池管理

Solr建索引时候是有最大的线程数限制的，它由solrconfig.xml的<maxIndexingThreads>8</maxIndexingThreads>控制的，该值等于8就是说Solr最多只能用8个线程来进行updatedocument。

那么Solr建索引时候是怎么管理线程池的呢，主要是通过ThreadAffinityDocumentsWriterThreadPool来进行管理的，它继承了DocumentsWriterPerThreadPool类。ThreadAffinityDocumentsWriterThreadPool的结构并不复杂，主要的一个函数是getAndLock()。

在建索引时候即updatedocuments时候，Solr先要调用getAndLock去获取ThreadState这个锁。而ThreadState这个锁就是存放在ThreadAffinityDocumentsWriterThreadPool的threadBings这个线程池里面。

首先先看下什么是ThreadState锁，源码如下：

ThreadState是DocumentsWriterPerThreadPool的一个内部类。它包含了一个DocumentsWriterPerThread类的实例以及状态控制，DocumentsWriterPerThread是线程池的一个线程，主要作用是索引的建立。该类比较简单就不详细介绍了。

 /**
   * {@link ThreadState} references and guards a
   * {@link DocumentsWriterPerThread} instance that is used during indexing to
   * build a in-memory index segment. {@link ThreadState} also holds all flush
   * related per-thread data controlled by {@link DocumentsWriterFlushControl}.
   * <p>
   * A {@link ThreadState}, its methods and members should only accessed by one
   * thread a time. Users must acquire the lock via {@link ThreadState#lock()}
   * and release the lock in a finally block via {@link ThreadState#unlock()}
   * before accessing the state.
   */
  @SuppressWarnings("serial")
  final static class ThreadState extends ReentrantLock {
    DocumentsWriterPerThread dwpt;
    // TODO this should really be part of DocumentsWriterFlushControl
    // write access guarded by DocumentsWriterFlushControl
    volatile boolean flushPending = false;
    // TODO this should really be part of DocumentsWriterFlushControl
    // write access guarded by DocumentsWriterFlushControl
    long bytesUsed = 0;
    // guarded by Reentrant lock
    private boolean isActive = true;

    ThreadState(DocumentsWriterPerThread dpwt) {
      this.dwpt = dpwt;
    }
    
    /**
     * Resets the internal {@link DocumentsWriterPerThread} with the given one. 
     * if the given DWPT is <code>null</code> this ThreadState is marked as inactive and should not be used
     * for indexing anymore.
     * @see #isActive()  
     */
  
    private void deactivate() {
      assert this.isHeldByCurrentThread();
      isActive = false;
      reset();
    }
    
    private void reset() {
      assert this.isHeldByCurrentThread();
      this.dwpt = null;
      this.bytesUsed = 0;
      this.flushPending = false;
    }
    
    /**
     * Returns <code>true</code> if this ThreadState is still open. This will
     * only return <code>false</code> iff the DW has been closed and this
     * ThreadState is already checked out for flush.
     */
    boolean isActive() {
      assert this.isHeldByCurrentThread();
      return isActive;
    }
    
    boolean isInitialized() {
      assert this.isHeldByCurrentThread();
      return isActive() && dwpt != null;
    }
    
    /**
     * Returns the number of currently active bytes in this ThreadState's
     * {@link DocumentsWriterPerThread}
     */
    public long getBytesUsedPerThread() {
      assert this.isHeldByCurrentThread();
      // public for FlushPolicy
      return bytesUsed;
    }
    
    /**
     * Returns this {@link ThreadState}s {@link DocumentsWriterPerThread}
     */
    public DocumentsWriterPerThread getDocumentsWriterPerThread() {
      assert this.isHeldByCurrentThread();
      // public for FlushPolicy
      return dwpt;
    }
    
    /**
     * Returns <code>true</code> iff this {@link ThreadState} is marked as flush
     * pending otherwise <code>false</code>
     */
    public boolean isFlushPending() {
      return flushPending;
    }
  }

/**
 * A {@link DocumentsWriterPerThreadPool} implementation that tries to assign an
 * indexing thread to the same {@link ThreadState} each time the thread tries to
 * obtain a {@link ThreadState}. Once a new {@link ThreadState} is created it is
 * associated with the creating thread. Subsequently, if the threads associated
 * {@link ThreadState} is not in use it will be associated with the requesting
 * thread. Otherwise, if the {@link ThreadState} is used by another thread
 * {@link ThreadAffinityDocumentsWriterThreadPool} tries to find the currently
 * minimal contended {@link ThreadState}.
 */
class ThreadAffinityDocumentsWriterThreadPool extends DocumentsWriterPerThreadPool {
  private Map<Thread, ThreadState> threadBindings = new ConcurrentHashMap<>();
  
  /**
   * Creates a new {@link ThreadAffinityDocumentsWriterThreadPool} with a given maximum of {@link ThreadState}s.
   */
  public ThreadAffinityDocumentsWriterThreadPool(int maxNumPerThreads) {
    super(maxNumPerThreads);
    assert getMaxThreadStates() >= 1;
  }

  @Override
  public ThreadState getAndLock(Thread requestingThread, DocumentsWriter documentsWriter) {
    ThreadState threadState = threadBindings.get(requestingThread);
    if (threadState != null && threadState.tryLock()) {
      return threadState;
    }
    ThreadState minThreadState = null;

    
    /* TODO -- another thread could lock the minThreadState we just got while 
     we should somehow prevent this. */
    // Find the state that has minimum number of threads waiting
    minThreadState = minContendedThreadState();
    if (minThreadState == null || minThreadState.hasQueuedThreads()) {
      final ThreadState newState = newThreadState(); // state is already locked if non-null
      if (newState != null) {
        assert newState.isHeldByCurrentThread();
        threadBindings.put(requestingThread, newState);
        return newState;
      } else if (minThreadState == null) {
        /*
         * no new threadState available we just take the minContented one
         * This must return a valid thread state since we accessed the 
         * synced context in newThreadState() above.
         */
        minThreadState = minContendedThreadState();
      }
    }
    assert minThreadState != null: "ThreadState is null";
    
    minThreadState.lock();
    return minThreadState;
  }

  @Override
  public ThreadAffinityDocumentsWriterThreadPool clone() {
    ThreadAffinityDocumentsWriterThreadPool clone = (ThreadAffinityDocumentsWriterThreadPool) super.clone();
    clone.threadBindings = new ConcurrentHashMap<>();
    return clone;
  }
}

再回到ThreadAffinityDocumentWriterThreadPool类。getAndLock的主要流程如下:

1. 请求线程requestingThread需要进行updatedocument操作，它首先会尝试从线程池threadBings获取自身线程的ThreadState锁并尝试去锁它即trylock。如果锁成功了，那么它就能再度获取到自身线程的ThreadState，这是最好的一种情况。

2. 如果自身线程的trylock失败，说明该ThreadState已经被别的requestingThread线程抢去，那么请求线程requestingThread只能去线程池threadBings获取别的线程。获取的规则是minContendedThreadState(),源码如下所示.

minContendedThreadState的规则就是遍历所有活跃的ThreadState，如果ThreadState的队列内元素个数最少(即等待这个ThreadState的线程最少)，那么这个ThreadState就是返回的那个ThreadState，即minThreadState.

  /**
   * Returns the ThreadState with the minimum estimated number of threads
   * waiting to acquire its lock or <code>null</code> if no {@link ThreadState}
   * is yet visible to the calling thread.
   */
  ThreadState minContendedThreadState() {
    ThreadState minThreadState = null;
    final int limit = numThreadStatesActive;
    for (int i = 0; i < limit; i++) {
      final ThreadState state = threadStates[i];
      if (minThreadState == null || state.getQueueLength() < minThreadState.getQueueLength()) {
        minThreadState = state;
      }
    }
    return minThreadState;
  }

3. 如果minThreadState==null(一般是第一个获取ThreadState这种情况)或者minThreadState有其他线程在等待(正常情况下都会有线程在等的)，那么requestingThread会去申请新的ThreadState，即从maxIndexingThreads的线程里申请，源码如下。

threadStates是一个ThreadStates的数组，当需要threadBings的ThreadState个数(也就是活跃的线程)小于threadStates的元素个数（maxIndexingThreads）时就能申请到新的ThreadState。

  /**
   * Returns a new {@link ThreadState} iff any new state is available otherwise
   * <code>null</code>.
   * <p>
   * NOTE: the returned {@link ThreadState} is already locked iff non-
   * <code>null</code>.
   * 
   * @return a new {@link ThreadState} iff any new state is available otherwise
   *         <code>null</code>
   */
  synchronized ThreadState newThreadState() {
    if (numThreadStatesActive < threadStates.length) {
      final ThreadState threadState = threadStates[numThreadStatesActive];
      threadState.lock(); // lock so nobody else will get this ThreadState
      boolean unlock = true;
      try {
        if (threadState.isActive()) {
          // unreleased thread states are deactivated during DW#close()
          numThreadStatesActive++; // increment will publish the ThreadState
          assert threadState.dwpt == null;
          unlock = false;
          return threadState;
        }
        // unlock since the threadstate is not active anymore - we are closed!
        assert assertUnreleasedThreadStatesInactive();
        return null;
      } finally {
        if (unlock) {
          // in any case make sure we unlock if we fail 
          threadState.unlock();
        }
      }
    }
    return null;
  }

4. 如果minContentedThreadState获取成功，那么threadBings的线程池就会得到更新。如果minContentedThreadState获取失败，那么说明threadStates数组以及分配完全，那么请求线程会再去取获取minContentedThreadState。

5. 最后请求线程会去lock minThreadState，如果lock失败就进入休眠，一直等到lock成功。这是最不好的一种结果。

最后在源码说道，请求线程在获取minThreadState时候别的线程也有可能获取到该minThreadState，目前来说这是一种缺陷。

<maxIndexingThreads>8</maxIndexingThreads>这个配置对建索引的性能有较大影响，如果太小那么建索引时候等待情况就会较多。如果太大又增加服务器的负荷，所以要综合选择。