All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.apache.hadoop.hbase.io.FSDataInputStreamWrapper Maven / Gradle / Ivy

There is a newer version: 3.0.0-beta-1
Show newest version
/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
package org.apache.hadoop.hbase.io;

import java.io.Closeable;
import java.io.IOException;
import java.io.InputStream;
import java.util.concurrent.atomic.AtomicInteger;
import org.apache.hadoop.fs.CanUnbuffer;
import org.apache.hadoop.fs.FSDataInputStream;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hbase.fs.HFileSystem;
import org.apache.hadoop.hdfs.client.HdfsDataInputStream;
import org.apache.yetus.audience.InterfaceAudience;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import org.apache.hbase.thirdparty.com.google.common.io.Closeables;

/**
 * Wrapper for input stream(s) that takes care of the interaction of FS and HBase checksums, as well
 * as closing streams. Initialization is not thread-safe, but normal operation is; see method
 * comments.
 */
@InterfaceAudience.Private
public class FSDataInputStreamWrapper implements Closeable {
  private static final Logger LOG = LoggerFactory.getLogger(FSDataInputStreamWrapper.class);
  private static final boolean isLogTraceEnabled = LOG.isTraceEnabled();

  private final HFileSystem hfs;
  private final Path path;
  private final FileLink link;
  private final boolean doCloseStreams;
  private final boolean dropBehind;
  private final long readahead;

  /**
   * Two stream handles, one with and one without FS-level checksum. HDFS checksum setting is on FS
   * level, not single read level, so you have to keep two FS objects and two handles open to
   * interleave different reads freely, which is very sad. This is what we do: 1) First, we need to
   * read the trailer of HFile to determine checksum parameters. We always use FS checksum to do
   * that, so ctor opens {@link #stream}. 2.1) After that, if HBase checksum is not used, we'd just
   * always use {@link #stream}; 2.2) If HBase checksum can be used, we'll open
   * {@link #streamNoFsChecksum}, and close {@link #stream}. User MUST call prepareForBlockReader
   * for that to happen; if they don't, (2.1) will be the default. 3) The users can call
   * {@link #shouldUseHBaseChecksum()}, and pass its result to {@link #getStream(boolean)} to get
   * stream (if Java had out/pointer params we could return both in one call). This stream is
   * guaranteed to be set. 4) The first time HBase checksum fails, one would call
   * {@link #fallbackToFsChecksum(int)}. That will take lock, and open {@link #stream}. While this
   * is going on, others will continue to use the old stream; if they also want to fall back,
   * they'll also call {@link #fallbackToFsChecksum(int)}, and block until {@link #stream} is set.
   * 5) After some number of checksumOk() calls, we will go back to using HBase checksum. We will
   * have 2 handles; however we presume checksums fail so rarely that we don't care.
   */
  private volatile FSDataInputStream stream = null;
  private volatile FSDataInputStream streamNoFsChecksum = null;
  private final Object streamNoFsChecksumFirstCreateLock = new Object();

  // The configuration states that we should validate hbase checksums
  private boolean useHBaseChecksumConfigured;

  // Record the current state of this reader with respect to
  // validating checkums in HBase. This is originally set the same
  // value as useHBaseChecksumConfigured, but can change state as and when
  // we encounter checksum verification failures.
  private volatile boolean useHBaseChecksum;

  // In the case of a checksum failure, do these many succeeding
  // reads without hbase checksum verification.
  private AtomicInteger hbaseChecksumOffCount = new AtomicInteger(-1);

  private final static ReadStatistics readStatistics = new ReadStatistics();

  private static class ReadStatistics {
    long totalBytesRead;
    long totalLocalBytesRead;
    long totalShortCircuitBytesRead;
    long totalZeroCopyBytesRead;
  }

  private Boolean instanceOfCanUnbuffer = null;
  private CanUnbuffer unbuffer = null;

  public FSDataInputStreamWrapper(FileSystem fs, Path path) throws IOException {
    this(fs, path, false, -1L);
  }

  public FSDataInputStreamWrapper(FileSystem fs, Path path, boolean dropBehind, long readahead)
    throws IOException {
    this(fs, null, path, dropBehind, readahead);
  }

  public FSDataInputStreamWrapper(FileSystem fs, FileLink link, boolean dropBehind, long readahead)
    throws IOException {
    this(fs, link, null, dropBehind, readahead);
  }

  private FSDataInputStreamWrapper(FileSystem fs, FileLink link, Path path, boolean dropBehind,
    long readahead) throws IOException {
    assert (path == null) != (link == null);
    this.path = path;
    this.link = link;
    this.doCloseStreams = true;
    this.dropBehind = dropBehind;
    this.readahead = readahead;
    // If the fs is not an instance of HFileSystem, then create an instance of HFileSystem
    // that wraps over the specified fs. In this case, we will not be able to avoid
    // checksumming inside the filesystem.
    this.hfs = (fs instanceof HFileSystem) ? (HFileSystem) fs : new HFileSystem(fs);

    // Initially we are going to read the tail block. Open the reader w/FS checksum.
    this.useHBaseChecksumConfigured = this.useHBaseChecksum = false;
    this.stream = (link != null) ? link.open(hfs) : hfs.open(path);
    setStreamOptions(stream);
  }

  private void setStreamOptions(FSDataInputStream in) {
    try {
      in.setDropBehind(dropBehind);
    } catch (Exception e) {
      // Skipped.
    }
    if (readahead >= 0) {
      try {
        in.setReadahead(readahead);
      } catch (Exception e) {
        // Skipped.
      }
    }
  }

  /**
   * Prepares the streams for block reader. NOT THREAD SAFE. Must be called once, after any reads
   * finish and before any other reads start (what happens in reality is we read the tail, then call
   * this based on what's in the tail, then read blocks).
   * @param forceNoHBaseChecksum Force not using HBase checksum.
   */
  public void prepareForBlockReader(boolean forceNoHBaseChecksum) throws IOException {
    if (hfs == null) return;
    assert this.stream != null && !this.useHBaseChecksumConfigured;
    boolean useHBaseChecksum =
      !forceNoHBaseChecksum && hfs.useHBaseChecksum() && (hfs.getNoChecksumFs() != hfs);

    if (useHBaseChecksum) {
      FileSystem fsNc = hfs.getNoChecksumFs();
      this.streamNoFsChecksum = (link != null) ? link.open(fsNc) : fsNc.open(path);
      setStreamOptions(streamNoFsChecksum);
      this.useHBaseChecksumConfigured = this.useHBaseChecksum = useHBaseChecksum;
      // Close the checksum stream; we will reopen it if we get an HBase checksum failure.
      this.stream.close();
      this.stream = null;
    }
  }

  /** For use in tests. */
  public FSDataInputStreamWrapper(FSDataInputStream fsdis) {
    this(fsdis, fsdis);
  }

  /** For use in tests. */
  public FSDataInputStreamWrapper(FSDataInputStream fsdis, FSDataInputStream noChecksum) {
    doCloseStreams = false;
    stream = fsdis;
    streamNoFsChecksum = noChecksum;
    path = null;
    link = null;
    hfs = null;
    useHBaseChecksumConfigured = useHBaseChecksum = false;
    dropBehind = false;
    readahead = 0;
  }

  /** Returns Whether we are presently using HBase checksum. */
  public boolean shouldUseHBaseChecksum() {
    return this.useHBaseChecksum;
  }

  /**
   * Get the stream to use. Thread-safe.
   * @param useHBaseChecksum must be the value that shouldUseHBaseChecksum has returned at some
   *                         point in the past, otherwise the result is undefined.
   */
  public FSDataInputStream getStream(boolean useHBaseChecksum) {
    return useHBaseChecksum ? this.streamNoFsChecksum : this.stream;
  }

  /**
   * Read from non-checksum stream failed, fall back to FS checksum. Thread-safe.
   * @param offCount For how many checksumOk calls to turn off the HBase checksum.
   */
  public FSDataInputStream fallbackToFsChecksum(int offCount) throws IOException {
    // checksumOffCount is speculative, but let's try to reset it less.
    boolean partOfConvoy = false;
    if (this.stream == null) {
      synchronized (streamNoFsChecksumFirstCreateLock) {
        partOfConvoy = (this.stream != null);
        if (!partOfConvoy) {
          this.stream = (link != null) ? link.open(hfs) : hfs.open(path);
        }
      }
    }
    if (!partOfConvoy) {
      this.useHBaseChecksum = false;
      this.hbaseChecksumOffCount.set(offCount);
    }
    return this.stream;
  }

  /** Report that checksum was ok, so we may ponder going back to HBase checksum. */
  public void checksumOk() {
    if (
      this.useHBaseChecksumConfigured && !this.useHBaseChecksum
        && (this.hbaseChecksumOffCount.getAndDecrement() < 0)
    ) {
      // The stream we need is already open (because we were using HBase checksum in the past).
      assert this.streamNoFsChecksum != null;
      this.useHBaseChecksum = true;
    }
  }

  private void updateInputStreamStatistics(FSDataInputStream stream) {
    // If the underlying file system is HDFS, update read statistics upon close.
    if (stream instanceof HdfsDataInputStream) {
      /**
       * Because HDFS ReadStatistics is calculated per input stream, it is not feasible to update
       * the aggregated number in real time. Instead, the metrics are updated when an input stream
       * is closed.
       */
      HdfsDataInputStream hdfsDataInputStream = (HdfsDataInputStream) stream;
      synchronized (readStatistics) {
        readStatistics.totalBytesRead +=
          hdfsDataInputStream.getReadStatistics().getTotalBytesRead();
        readStatistics.totalLocalBytesRead +=
          hdfsDataInputStream.getReadStatistics().getTotalLocalBytesRead();
        readStatistics.totalShortCircuitBytesRead +=
          hdfsDataInputStream.getReadStatistics().getTotalShortCircuitBytesRead();
        readStatistics.totalZeroCopyBytesRead +=
          hdfsDataInputStream.getReadStatistics().getTotalZeroCopyBytesRead();
      }
    }
  }

  public static long getTotalBytesRead() {
    synchronized (readStatistics) {
      return readStatistics.totalBytesRead;
    }
  }

  public static long getLocalBytesRead() {
    synchronized (readStatistics) {
      return readStatistics.totalLocalBytesRead;
    }
  }

  public static long getShortCircuitBytesRead() {
    synchronized (readStatistics) {
      return readStatistics.totalShortCircuitBytesRead;
    }
  }

  public static long getZeroCopyBytesRead() {
    synchronized (readStatistics) {
      return readStatistics.totalZeroCopyBytesRead;
    }
  }

  /** CloseClose stream(s) if necessary. */
  @Override
  public void close() {
    if (!doCloseStreams) {
      return;
    }
    updateInputStreamStatistics(this.streamNoFsChecksum);
    // we do not care about the close exception as it is for reading, no data loss issue.
    Closeables.closeQuietly(streamNoFsChecksum);

    updateInputStreamStatistics(stream);
    Closeables.closeQuietly(stream);
  }

  public HFileSystem getHfs() {
    return this.hfs;
  }

  /**
   * This will free sockets and file descriptors held by the stream only when the stream implements
   * org.apache.hadoop.fs.CanUnbuffer. NOT THREAD SAFE. Must be called only when all the clients
   * using this stream to read the blocks have finished reading. If by chance the stream is
   * unbuffered and there are clients still holding this stream for read then on next client read
   * request a new socket will be opened by Datanode without client knowing about it and will serve
   * its read request. Note: If this socket is idle for some time then the DataNode will close the
   * socket and the socket will move into CLOSE_WAIT state and on the next client request on this
   * stream, the current socket will be closed and a new socket will be opened to serve the
   * requests.
   */
  @SuppressWarnings({ "rawtypes" })
  public void unbuffer() {
    FSDataInputStream stream = this.getStream(this.shouldUseHBaseChecksum());
    if (stream != null) {
      InputStream wrappedStream = stream.getWrappedStream();
      // CanUnbuffer interface was added as part of HDFS-7694 and the fix is available in Hadoop
      // 2.6.4+ and 2.7.1+ versions only so check whether the stream object implements the
      // CanUnbuffer interface or not and based on that call the unbuffer api.
      final Class streamClass = wrappedStream.getClass();
      if (this.instanceOfCanUnbuffer == null) {
        // To ensure we compute whether the stream is instance of CanUnbuffer only once.
        this.instanceOfCanUnbuffer = false;
        if (wrappedStream instanceof CanUnbuffer) {
          this.unbuffer = (CanUnbuffer) wrappedStream;
          this.instanceOfCanUnbuffer = true;
        }
      }
      if (this.instanceOfCanUnbuffer) {
        try {
          this.unbuffer.unbuffer();
        } catch (UnsupportedOperationException e) {
          if (isLogTraceEnabled) {
            LOG.trace("Failed to invoke 'unbuffer' method in class " + streamClass
              + " . So there may be the stream does not support unbuffering.", e);
          }
        }
      } else {
        if (isLogTraceEnabled) {
          LOG.trace("Failed to find 'unbuffer' method in class " + streamClass);
        }
      }
    }
  }
}




© 2015 - 2024 Weber Informatics LLC | Privacy Policy