com.bigdata.io.FileChannelUtility Maven / Gradle / Ivy
/*
Copyright (C) SYSTAP, LLC DBA Blazegraph 2006-2016. All rights reserved.
Contact:
SYSTAP, LLC DBA Blazegraph
2501 Calvert ST NW #106
Washington, DC 20008
[email protected]
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
/*
* Created on May 14, 2008
*/
package com.bigdata.io;
import java.io.IOException;
import java.io.InputStream;
import java.io.RandomAccessFile;
import java.nio.ByteBuffer;
import java.nio.channels.AsynchronousCloseException;
import java.nio.channels.AsynchronousFileChannel;
import java.nio.channels.ClosedByInterruptException;
import java.nio.channels.ClosedChannelException;
import java.nio.channels.FileChannel;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import org.apache.log4j.Logger;
import com.bigdata.journal.TemporaryRawStore;
/**
* A helper class for operations on {@link FileChannel}s.
*
* @author Bryan Thompson
*/
public class FileChannelUtility {
private static final Logger log = Logger.getLogger(FileChannelUtility.class);
private static final boolean INFO = log.isInfoEnabled();
private static final boolean DEBUG = log.isDebugEnabled();
/**
* Reads {@link ByteBuffer#remaining()} bytes into the caller's
* {@link ByteBuffer} from the channel starting at offset pos. The
* position of the {@link ByteBuffer} is advanced to the limit. The offset
* of the channel is NOT modified as a side-effect.
*
* This handles partial reads by reading until nbytes bytes have been
* transferred.
*
* @param channel
* The source {@link FileChannel}.
* @param src
* A {@link ByteBuffer} into which the data will be read.
* @param pos
* The offset from which the data will be read.
*
* @return The #of disk read operations that were required.
*
* @throws IOException
*
* @throws IllegalArgumentException
* if any parameter is null.
* @throws IllegalArgumentException
* if pos is negative.
* @throws IllegalArgumentException
* if dst does not have any bytes remaining.
*
* @todo This might need to use a static direct buffer pool to avoid leaking
* temporary direct buffers since Java may attempt to allocate a
* "temporary" direct buffer if [dst] is not already a direct buffer.
*
* @deprecated by {@link #readAll(IReopenChannel, ByteBuffer, long)} which
* handles transparent re-opening of the store in order to
* complete the read operation.
*/
static public int readAll(final FileChannel channel, final ByteBuffer src,
final long pos) throws IOException {
return readAll(new NOPReopener(channel), src, pos);
}
/**
* Reads {@link ByteBuffer#remaining()} bytes into the caller's
* {@link ByteBuffer} from the channel starting at offset pos. The
* position of the {@link ByteBuffer} is advanced to the limit. The offset
* of the channel is NOT modified as a side-effect.
*
* This handles partial reads by reading until nbytes bytes have been
* transferred.
*
* @param opener
* An object which knows how to re-open the {@link FileChannel}
* and knows when the resource is no longer available.
* @param src
* A {@link ByteBuffer} into which the data will be read.
* @param pos
* The offset from which the data will be read.
*
* @return The #of disk read operations that were required.
*
* @throws IOException
*
* @throws IllegalArgumentException
* if any parameter is null.
* @throws IllegalArgumentException
* if pos is negative.
* @throws IllegalArgumentException
* if dst does not have any bytes remaining.
*
* @todo This might need to use a static direct buffer pool to avoid leaking
* temporary direct buffers since Java may attempt to allocate a
* "temporary" direct buffer if [dst] is not already a direct buffer.
*/
static public int readAll(final IReopenChannel opener,
final ByteBuffer src, final long pos) throws IOException {
if (opener == null)
throw new IllegalArgumentException();
if (src == null)
throw new IllegalArgumentException();
if (pos < 0)
throw new IllegalArgumentException();
final int nbytes = src.remaining();
if (nbytes == 0)
throw new IllegalArgumentException();
if (DEBUG) {
log.debug("pos=" + pos + ", #bytes=" + nbytes);
}
/*
* Read in loop until nbytes == nread
*/
int nreads = 0; // #of read operations.
int count = 0; // #of bytes read.
while (count < nbytes) {
final FileChannel channel;
try {
channel = opener.reopenChannel();
if(channel == null) {
throw new AssertionError("Channel is null?");
}
} catch (IllegalStateException ex) {
// the channel could not be re-opened.
throw ex;
}
final int nread;
try {
// copy the data into the buffer
nread = channel.read(src, pos + count);
} catch (ClosedByInterruptException ex) {
/*
* This indicates that this thread was interrupted. We
* always abort in this case.
*/
throw ex;
} catch (AsynchronousCloseException ex) {
/*
* The channel was closed asynchronously while blocking during
* the read. We will continue to read if the channel can be
* reopened.
*/
continue;
} catch (ClosedChannelException ex) {
/*
* The channel is closed. This could have occurred between the
* moment when we got the FileChannel reference and the moment
* when we tried to read on the FileChannel. We will continue to
* read if the channel can be reopened.
*/
continue;
} catch (IOException ex) {
// anything else.
throw ex;
}
if (nread == -1)
throw new IOException("EOF reading on channel: remaining="
+ (nbytes - count) + ", nread=" + nreads + ", pos="
+ pos + ", bytesRead=" + count);
count += nread; // #bytes read so far.
nreads++;
if (nreads == 100) {
log.warn("reading on channel: remaining=" + (nbytes - count)
+ ", nread=" + nreads + ", pos=" + pos + ", bytesRead="
+ count);
} else if (nreads == 1000) {
log.error("reading on channel: remaining=" + (nbytes - count)
+ ", nread=" + nreads + ", pos=" + pos + ", bytesRead="
+ count);
} else if (nreads > 10000) {
throw new RuntimeException("reading on channel: remaining="
+ (nbytes - count) + ", nread=" + nreads + ", pos="
+ pos + ", bytesRead=" + count);
}
}
if (count != nbytes) {
throw new RuntimeException("Expected to read " + nbytes
+ " bytes but read " + count + " bytes in " + nreads
+ " reads");
}
if (INFO) {
log.info("read " + nbytes + " bytes from offset=" + pos + " in "
+ nreads + " IOs");
}
return nreads;
}
/**
* Define interface to support callback from readAllAsync
*/
public interface IAsyncOpener {
AsynchronousFileChannel getAsyncChannel();
}
/**
* The AsyncTransfer class encapsulates the state required to make
* asynchronous transfer requests. It was written explicitly to support
* asynchronous transfer of blob data that requires multiple reads.
*/
static public class AsyncTransfer {
final long m_addr;
final ByteBuffer m_buffer;
Future m_fut = null;
public AsyncTransfer(final long addr, final ByteBuffer buffer) {
m_addr = addr;
m_buffer = buffer;
m_buffer.mark(); // mark buffer to support reset on any retries
}
protected void read(AsynchronousFileChannel channel) {
if (!isDone()) {
// ensure buffer is ready
m_buffer.reset();
m_fut = channel.read(m_buffer, m_addr);
}
}
public int complete() throws InterruptedException, ExecutionException {
return m_fut.get();
}
public void cancel() {
m_fut.cancel(true);
}
public boolean isDone() {
return m_fut != null && m_fut.isDone();
}
}
/**
* readAllAsync will make repeated attempts to read data into the transfer buffers defined
* in the List of AsyncTransfer instances.
*
* @param opener provides AsynchronouseFileChannel
* @param transfers defines the list of required transfers
* @return the total number of bytes read asynchronously
* @throws IOException
*/
static public int readAllAsync(final IAsyncOpener opener, List transfers) throws IOException {
while (true) {
final AsynchronousFileChannel channel = opener.getAsyncChannel();
try {
for (AsyncTransfer transfer : transfers) {
// the AsyncTransfer object handles retries by checking any existing transfer state
transfer.read(channel);
}
int totalReads = 0;
for (AsyncTransfer transfer : transfers) {
totalReads += transfer.complete();
}
return totalReads;
} catch (Exception e) {
log.info("Error from async IO", e); // just log and retry
} finally {
for (AsyncTransfer transfer : transfers) {
transfer.cancel();
}
}
}
}
/**
* Write bytes in data from the position to the limit on the channel
* starting at pos. The position of the buffer will be advanced to
* the limit. The position of the channel is not changed by this method.
*
* Note: I have seen count != remaining() for a single invocation of
* FileChannel#write(). This occurred 5 hours into a run with the write cache
* disabled (so lots of small record writes). All of a sudden, several
* writes wound up reporting too few bytes written - this persisted until
* the end of the run (Fedora core 6 with Sun JDK 1.6.0_03). I have since
* modified this code to use a loop to ensure that all bytes get written.
*
* @return The #of disk write operations that were required.
*
* @todo This might need to use a static direct buffer pool to avoid leaking
* temporary direct buffers since Java probably will attempt to
* allocate a "temporary" direct buffer if [data] is not already a
* direct buffer. There is logic in {@link TemporaryRawStore} for
* handling the overflow from a heap buffer onto the disk which
* already does this.
*
* @deprecated by {@link #writeAll(IReopenChannel, ByteBuffer, long)}
*/
static public int writeAll(final FileChannel channel, final ByteBuffer data,
final long pos) throws IOException {
return writeAll(new NOPReopener(channel),data,pos);
// final int nbytes = data.remaining();
//
// int count = 0;
// int nwrites = 0;
//
// while (data.remaining() > 0) {
//
// final int nwritten = channel.write(data, pos + count);
//
// count += nwritten;
//
// nwrites++;
//
// if (nwrites == 100) {
//
// log.warn("writing on channel: remaining=" + data.remaining()
// + ", nwrites=" + nwrites + ", written=" + count);
//
// } else if (nwrites == 1000) {
//
// log.error("writing on channel: remaining=" + data.remaining()
// + ", nwrites=" + nwrites + ", written=" + count);
//
// } else if (nwrites > 10000) {
//
// throw new RuntimeException("writing on channel: remaining="
// + data.remaining() + ", nwrites=" + nwrites
// + ", written=" + count);
//
// }
//
// }
//
// if (count != nbytes) {
//
// throw new RuntimeException("Expecting to write " + nbytes
// + " bytes, but wrote " + count + " bytes in " + nwrites);
//
// }
//
// return nwrites;
}
/**
* Write bytes in data from the position to the limit on the channel
* starting at pos. The position of the buffer will be advanced to
* the limit. The position of the channel is not changed by this method. If
* the backing channel is asynchronously in another thread closed then it
* will be re-opened and the write will continue.
*
* Note: I have seen count != remaining() for a single invocation of
* FileChannel#write(). This occurred 5 hours into a run with the write cache
* disabled (so lots of small record writes). All of a sudden, several
* writes wound up reporting too few bytes written - this persisted until
* the end of the run (Fedora core 6 with Sun JDK 1.6.0_03). I have since
* modified this code to use a loop to ensure that all bytes get written.
*
* @param opener
* @param data
* @param pos
*
* @return The #of disk write operations that were required.
*
* @throws IOException
*/
static public int writeAll(final IReopenChannel opener,
final ByteBuffer data, final long pos) throws IOException {
final long begin = System.nanoTime();
final int nbytes = data.remaining();
int count = 0;
int nwrites = 0;
while (data.remaining() > 0) {
final FileChannel channel;
try {
channel = opener.reopenChannel();
if(channel == null) {
throw new AssertionError("Channel is null?");
}
} catch (IllegalStateException ex) {
// the channel could not be re-opened.
throw ex;
}
final int nwritten;
try {
nwritten = channel.write(data, pos + count);
} catch (ClosedByInterruptException ex) {
/*
* This indicates that this thread was interrupted. We always
* abort in this case.
*/
throw ex;
} catch (AsynchronousCloseException ex) {
/*
* The channel was closed asynchronously while blocking during
* the write. We will continue to write if the channel can be
* reopened.
*/
continue;
} catch (ClosedChannelException ex) {
/*
* The channel is closed. This could have occurred between the
* moment when we got the FileChannel reference and the moment
* when we tried to write on the FileChannel. We will continue
* to write if the channel can be reopened.
*/
continue;
} catch (IOException ex) {
// anything else.
throw ex;
}
count += nwritten;
nwrites++;
if (nwrites == 100) {
log.warn("writing on channel: remaining=" + data.remaining()
+ ", nwrites=" + nwrites + ", written=" + count);
} else if (nwrites == 1000) {
log.error("writing on channel: remaining=" + data.remaining()
+ ", nwrites=" + nwrites + ", written=" + count);
} else if (nwrites > 10000) {
throw new RuntimeException("writing on channel: remaining="
+ data.remaining() + ", nwrites=" + nwrites
+ ", written=" + count);
}
}
if (count != nbytes) {
throw new RuntimeException("Expecting to write " + nbytes
+ " bytes, but wrote " + count + " bytes in " + nwrites);
}
final long elapsed = System.nanoTime() - begin;
if (INFO) {
log.info("wrote on disk: address: " + pos + ", bytes=" + nbytes + ", elapsed="
+ TimeUnit.NANOSECONDS.toMillis(elapsed) + "ms");
}
return nwrites;
}
/**
* {@link FileChannel} to {@link FileChannel} transfer of count bytes
* from the sourceChannel starting at the fromPosition onto
* the out file starting at its current position. The position on the
* sourceChannel is updated by this method and will be the positioned
* after the last byte transferred (this is done by
* {@link FileChannel#transferFrom(java.nio.channels.ReadableByteChannel, long, long)}
* so we have no choice about that). The position on the out file is
* updated by this method and will be located after the last byte
* transferred (this is done since we may have to extend the output file in
* which case its position will have been changed anyway so this makes the
* API more uniform).
*
* @param sourceChannel
* The source {@link FileChannel}.
* @param fromPosition
* The first byte to transfer from the sourceChannel.
* @param count
* The #of bytes to transfer.
* @param out
* The output file.
* @param toPosition
* The data will be transferred onto the output channel starting
* at this position.
*
* @return #of IO operations required to effect the transfer.
*
* @throws IOException
*
* @issue There is a known bug with large transfers under Windows. See FileChannel.transferTo() doesn't work if address space runs out
*
*
* @todo as a workaround, detect the Windows platform and do the IO
* ourselves when the #of bytes to transfer exceeds 500M.
*
* @todo does not handle asynchronous closes due to interrupts in other
* threads during a concurrent NIO operation on the same source/dest
* file channel (I have never seen this problem and perhaps the NIO
* API excludes it).
*
* @todo I have seen this exception when the JVM was allocated to much of
* the total RAM on the machine (in this case 2600m out of 3G
* available with 4G installed). The workaround is to reduce the -Xmx
* parameter value. This could also be corrected by failing over to a
* copy based on normal file IO.
*
*
* Caused by: java.io.IOException: Map failed
* at sun.nio.ch.FileChannelImpl.map(FileChannelImpl.java:758)
* at sun.nio.ch.FileChannelImpl.transferFromFileChannel(FileChannelImpl.java:537)
* at sun.nio.ch.FileChannelImpl.transferFrom(FileChannelImpl.java:600)
* at com.bigdata.io.FileChannelUtility.transferAll(FileChannelUtility.java:553)
* at com.bigdata.journal.AbstractBufferStrategy.transferFromDiskTo(AbstractBufferStrategy.java:420)
* at com.bigdata.journal.DiskOnlyStrategy.transferTo(DiskOnlyStrategy.java:2442)
* at com.bigdata.btree.IndexSegmentBuilder.writeIndexSegment(IndexSegmentBuilder.java:1865)
* at com.bigdata.btree.IndexSegmentBuilder.call(IndexSegmentBuilder.java:1148)
* at com.bigdata.resources.IndexManager.buildIndexSegment(IndexManager.java:1755)
*
*/
static public int transferAll(final FileChannel sourceChannel,
final long fromPosition, final long count,
final RandomAccessFile out, final long toPosition) throws IOException {
final long begin = System.currentTimeMillis();
// the output channel.
final FileChannel outChannel = out.getChannel();
// Set the fromPosition on source channel.
sourceChannel.position(fromPosition);
/*
* Extend the output file.
*
* Note: This is required at least for some circumstances.
*
* Note:This can also have a side-effect on the file position.
*/
out.setLength(toPosition + count);
/*
* Transfer the data. It is possible that this will take multiple writes
* for at least some implementations.
*/
if (INFO)
log.info("fromPosition=" + fromPosition + ", count=" + count
+ ", toPosition=" + toPosition);
int nwrites = 0; // #of write operations.
long nwritten = 0;
long n = count;
{
long to = toPosition;
while (n > 0) {
final long nbytes = outChannel.transferFrom(sourceChannel, to,
n);
to += nbytes;
nwritten += nbytes;
n -= nbytes;
nwrites++;
if (nwrites == 100) {
log.warn("writing on channel: remaining=" + n
+ ", nwrites=" + nwrites + ", written=" + nwritten
+ " of " + count + " bytes");
} else if (nwrites == 1000) {
log.error("writing on channel: remaining=" + n
+ ", nwrites=" + nwrites + ", written=" + nwritten
+ " of " + count + " bytes");
} else if (nwrites > 10000) {
throw new RuntimeException("writing on channel: remaining="
+ n + ", nwrites=" + nwrites + ", written="
+ nwritten + " of " + count + " bytes");
}
}
}
// Verify transfer is complete.
if (n != 0) {
throw new IOException("Expected to transfer " + count
+ ", but transferred " + (count - n));
}
/*
* Update the position on the output channel since transferFrom does NOT
* do this itself.
*/
outChannel.position(toPosition + count);
final long elapsed = System.currentTimeMillis() - begin;
if (INFO)
log.info("Transferred " + count
+ " bytes from disk channel at offset " + fromPosition
+ " to disk channel at offset=" + toPosition + " in "
+ nwrites + " writes and " + elapsed + "ms");
return nwrites;
}
/**
* An InputStream implementation that utilizes the resilient readAll method of this class
*/
public static class ReopenerInputStream extends InputStream {
final private IReopenChannel m_opener;
final private ByteBuffer m_buffer;
final private IBufferAccess m_bufferAccess;
final private long m_eof;
private long m_cursor = 0;
final private byte[] m_singleByte = new byte[1];
public ReopenerInputStream(final IReopenChannel opener) throws IOException {
m_opener = opener;
m_eof = opener.reopenChannel().size();
try {
m_bufferAccess = DirectBufferPool.INSTANCE.acquire();
m_buffer = m_bufferAccess.buffer();
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
}
@Override
public void close() {
try {
m_bufferAccess.release();
} catch (InterruptedException e) {
log.warn(e);
}
}
@Override
public int read() throws IOException {
final int ret = read(m_singleByte, 0, 1);
if (ret == -1) {
return -1;
} else {
return m_singleByte[0];
}
}
@Override
public int read(final byte[] dst) throws IOException {
return read(dst, 0, dst.length);
}
@Override
public int read(final byte[] dst, final int off, final int len) throws IOException {
final int rdlen = (int) ((m_cursor + len) < m_eof ? len : m_eof-m_cursor);
if (rdlen == 0)
return -1; // eof!
m_buffer.position(0);
m_buffer.limit(rdlen);
final int ret = FileChannelUtility.readAll(m_opener, m_buffer, m_cursor);
if (ret > 0) {
m_cursor += rdlen;
if (log.isTraceEnabled())
log.trace("Request for " + len + " bytes");
m_buffer.position(0);
m_buffer.limit(rdlen);
m_buffer.get(dst, off, rdlen);
return rdlen;
}
return -1; // EOF
}
}
}