org.apache.hadoop.fs.azure.PageBlobOutputStream Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of hadoop-apache Show documentation
Show all versions of hadoop-apache Show documentation
Shaded version of Apache Hadoop for Presto
/**
* 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.fs.azure;
import static org.apache.hadoop.fs.azure.PageBlobFormatHelpers.PAGE_DATA_SIZE;
import static org.apache.hadoop.fs.azure.PageBlobFormatHelpers.PAGE_HEADER_SIZE;
import static org.apache.hadoop.fs.azure.PageBlobFormatHelpers.PAGE_SIZE;
import static org.apache.hadoop.fs.azure.PageBlobFormatHelpers.fromShort;
import static org.apache.hadoop.fs.azure.PageBlobFormatHelpers.withMD5Checking;
import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.OutputStream;
import java.util.Arrays;
import java.util.Locale;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.ThreadPoolExecutor;
import java.util.concurrent.TimeUnit;
import org.apache.hadoop.fs.StreamCapabilities;
import org.apache.hadoop.fs.Syncable;
import org.apache.hadoop.fs.azure.StorageInterface.CloudPageBlobWrapper;
import io.prestosql.hadoop.$internal.org.apache.commons.lang3.exception.ExceptionUtils;
import io.prestosql.hadoop.$internal.org.apache.commons.logging.Log;
import io.prestosql.hadoop.$internal.org.apache.commons.logging.LogFactory;
import org.apache.hadoop.conf.Configuration;
import io.prestosql.hadoop.$internal.com.google.common.annotations.VisibleForTesting;
import io.prestosql.hadoop.$internal.com.microsoft.azure.storage.OperationContext;
import io.prestosql.hadoop.$internal.com.microsoft.azure.storage.StorageException;
import io.prestosql.hadoop.$internal.com.microsoft.azure.storage.blob.BlobRequestOptions;
import io.prestosql.hadoop.$internal.com.microsoft.azure.storage.blob.CloudPageBlob;
/**
* An output stream that write file data to a page blob stored using ASV's
* custom format.
*/
final class PageBlobOutputStream extends OutputStream implements Syncable, StreamCapabilities {
/**
* The maximum number of raw bytes Azure Storage allows us to upload in a
* single request (4 MB).
*/
private static final int MAX_RAW_BYTES_PER_REQUEST = 4 * 1024 * 1024;
/**
* The maximum number of pages Azure Storage allows us to upload in a
* single request.
*/
private static final int MAX_PAGES_IN_REQUEST =
MAX_RAW_BYTES_PER_REQUEST / PAGE_SIZE;
/**
* The maximum number of data bytes (header not included) we can upload
* in a single request. I'm limiting it to (N - 1) pages to account for
* the possibility that we may have to rewrite the previous request's
* last page.
*/
private static final int MAX_DATA_BYTES_PER_REQUEST =
PAGE_DATA_SIZE * (MAX_PAGES_IN_REQUEST - 1);
private final CloudPageBlobWrapper blob;
private final OperationContext opContext;
/**
* If the IO thread encounters an error, it'll store it here.
*/
private volatile IOException lastError;
/**
* Current size of the page blob in bytes. It may be extended if the file
* gets full.
*/
private long currentBlobSize;
/**
* The current byte offset we're at in the blob (how many bytes we've
* uploaded to the server).
*/
private long currentBlobOffset;
/**
* The data in the last page that we wrote to the server, in case we have to
* overwrite it in the new request.
*/
private byte[] previousLastPageDataWritten = new byte[0];
/**
* The current buffer we're writing to before sending to the server.
*/
private ByteArrayOutputStream outBuffer;
/**
* The task queue for writing to the server.
*/
private final LinkedBlockingQueue ioQueue;
/**
* The thread pool we're using for writing to the server. Note that the IO
* write is NOT designed for parallelism, so there can only be one thread
* in that pool (I'm using the thread pool mainly for the lifetime management
* capabilities, otherwise I'd have just used a simple Thread).
*/
private final ThreadPoolExecutor ioThreadPool;
// The last task given to the ioThreadPool to execute, to allow
// waiting until it's done.
private WriteRequest lastQueuedTask;
// Whether the stream has been closed.
private boolean closed = false;
public static final Log LOG = LogFactory.getLog(AzureNativeFileSystemStore.class);
// Set the minimum page blob file size to 128MB, which is >> the default
// block size of 32MB. This default block size is often used as the
// hbase.regionserver.hlog.blocksize.
// The goal is to have a safe minimum size for HBase log files to allow them
// to be filled and rolled without exceeding the minimum size. A larger size
// can be used by setting the fs.azure.page.blob.size configuration variable.
public static final long PAGE_BLOB_MIN_SIZE = 128L * 1024L * 1024L;
// The default and minimum amount to extend a page blob by if it starts
// to get full.
public static final long
PAGE_BLOB_DEFAULT_EXTENSION_SIZE = 128L * 1024L * 1024L;
// The configured page blob extension size (either the default, or if greater,
// the value configured in fs.azure.page.blob.extension.size
private long configuredPageBlobExtensionSize;
/**
* Constructs an output stream over the given page blob.
*
* @param blob the blob that this stream is associated with.
* @param opContext an object used to track the execution of the operation
* @throws StorageException if anything goes wrong creating the blob.
*/
public PageBlobOutputStream(final CloudPageBlobWrapper blob,
final OperationContext opContext,
final Configuration conf) throws StorageException {
this.blob = blob;
this.outBuffer = new ByteArrayOutputStream();
this.opContext = opContext;
this.lastQueuedTask = null;
this.ioQueue = new LinkedBlockingQueue();
// As explained above: the IO writes are not designed for parallelism,
// so we only have one thread in this thread pool.
this.ioThreadPool = new ThreadPoolExecutor(1, 1, 2, TimeUnit.SECONDS,
ioQueue);
// Make page blob files have a size that is the greater of a
// minimum size, or the value of fs.azure.page.blob.size from configuration.
long pageBlobConfigSize = conf.getLong("fs.azure.page.blob.size", 0);
LOG.debug("Read value of fs.azure.page.blob.size as " + pageBlobConfigSize
+ " from configuration (0 if not present).");
long pageBlobSize = Math.max(PAGE_BLOB_MIN_SIZE, pageBlobConfigSize);
// Ensure that the pageBlobSize is a multiple of page size.
if (pageBlobSize % PAGE_SIZE != 0) {
pageBlobSize += PAGE_SIZE - pageBlobSize % PAGE_SIZE;
}
blob.create(pageBlobSize, new BlobRequestOptions(), opContext);
currentBlobSize = pageBlobSize;
// Set the page blob extension size. It must be a minimum of the default
// value.
configuredPageBlobExtensionSize =
conf.getLong("fs.azure.page.blob.extension.size", 0);
if (configuredPageBlobExtensionSize < PAGE_BLOB_DEFAULT_EXTENSION_SIZE) {
configuredPageBlobExtensionSize = PAGE_BLOB_DEFAULT_EXTENSION_SIZE;
}
// make sure it is a multiple of the page size
if (configuredPageBlobExtensionSize % PAGE_SIZE != 0) {
configuredPageBlobExtensionSize +=
PAGE_SIZE - configuredPageBlobExtensionSize % PAGE_SIZE;
}
}
private void checkStreamState() throws IOException {
if (lastError != null) {
throw lastError;
}
}
/**
* Query the stream for a specific capability.
*
* @param capability string to query the stream support for.
* @return true for hsync and hflush.
*/
@Override
public boolean hasCapability(String capability) {
switch (capability.toLowerCase(Locale.ENGLISH)) {
case StreamCapabilities.HSYNC:
case StreamCapabilities.HFLUSH:
return true;
default:
return false;
}
}
/**
* Closes this output stream and releases any system resources associated with
* this stream. If any data remains in the buffer it is committed to the
* service.
*/
@Override
public synchronized void close() throws IOException {
if (closed) {
return;
}
LOG.debug("Closing page blob output stream.");
flush();
checkStreamState();
ioThreadPool.shutdown();
try {
LOG.debug(ioThreadPool.toString());
if (!ioThreadPool.awaitTermination(10, TimeUnit.MINUTES)) {
LOG.debug("Timed out after 10 minutes waiting for IO requests to finish");
NativeAzureFileSystemHelper.logAllLiveStackTraces();
LOG.debug(ioThreadPool.toString());
throw new IOException("Timed out waiting for IO requests to finish");
}
} catch (InterruptedException e) {
LOG.debug("Caught InterruptedException");
// Restore the interrupted status
Thread.currentThread().interrupt();
}
closed = true;
}
/**
* A single write request for data to write to Azure storage.
*/
private class WriteRequest implements Runnable {
private final byte[] dataPayload;
private final CountDownLatch doneSignal = new CountDownLatch(1);
public WriteRequest(byte[] dataPayload) {
this.dataPayload = dataPayload;
}
public void waitTillDone() throws InterruptedException {
doneSignal.await();
}
@Override
public void run() {
try {
LOG.debug("before runInternal()");
runInternal();
LOG.debug("after runInternal()");
} finally {
doneSignal.countDown();
}
}
private void runInternal() {
if (lastError != null) {
// We're already in an error state, no point doing anything.
return;
}
if (dataPayload.length == 0) {
// Nothing to do.
return;
}
// Since we have to rewrite the last request's last page's data
// (may be empty), total data size is our data plus whatever was
// left from there.
final int totalDataBytes = dataPayload.length
+ previousLastPageDataWritten.length;
// Calculate the total number of pages we're writing to the server.
final int numberOfPages = (totalDataBytes / PAGE_DATA_SIZE)
+ (totalDataBytes % PAGE_DATA_SIZE == 0 ? 0 : 1);
// Fill up the raw bytes we're writing.
byte[] rawPayload = new byte[numberOfPages * PAGE_SIZE];
// Keep track of the size of the last page we uploaded.
int currentLastPageDataSize = -1;
for (int page = 0; page < numberOfPages; page++) {
// Our current byte offset in the data.
int dataOffset = page * PAGE_DATA_SIZE;
// Our current byte offset in the raw buffer.
int rawOffset = page * PAGE_SIZE;
// The size of the data in the current page.
final short currentPageDataSize = (short) Math.min(PAGE_DATA_SIZE,
totalDataBytes - dataOffset);
// Save off this page's size as the potential last page's size.
currentLastPageDataSize = currentPageDataSize;
// Write out the page size in the header.
final byte[] header = fromShort(currentPageDataSize);
System.arraycopy(header, 0, rawPayload, rawOffset, header.length);
rawOffset += header.length;
int bytesToCopyFromDataPayload = currentPageDataSize;
if (dataOffset < previousLastPageDataWritten.length) {
// First write out the last page's data.
final int bytesToCopyFromLastPage = Math.min(currentPageDataSize,
previousLastPageDataWritten.length - dataOffset);
System.arraycopy(previousLastPageDataWritten, dataOffset,
rawPayload, rawOffset, bytesToCopyFromLastPage);
bytesToCopyFromDataPayload -= bytesToCopyFromLastPage;
rawOffset += bytesToCopyFromLastPage;
dataOffset += bytesToCopyFromLastPage;
}
if (dataOffset >= previousLastPageDataWritten.length) {
// Then write the current payload's data.
System.arraycopy(dataPayload,
dataOffset - previousLastPageDataWritten.length,
rawPayload, rawOffset, bytesToCopyFromDataPayload);
}
}
// Raw payload constructed, ship it off to the server.
writePayloadToServer(rawPayload);
// Post-send bookkeeping.
currentBlobOffset += rawPayload.length;
if (currentLastPageDataSize < PAGE_DATA_SIZE) {
// Partial page, save it off so it's overwritten in the next request.
final int startOffset = (numberOfPages - 1) * PAGE_SIZE + PAGE_HEADER_SIZE;
previousLastPageDataWritten = Arrays.copyOfRange(rawPayload,
startOffset,
startOffset + currentLastPageDataSize);
// Since we're rewriting this page, set our current offset in the server
// to that page's beginning.
currentBlobOffset -= PAGE_SIZE;
} else {
// It wasn't a partial page, we won't need to rewrite it.
previousLastPageDataWritten = new byte[0];
}
// Extend the file if we need more room in the file. This typically takes
// less than 200 milliseconds if it has to actually be done,
// so it is okay to include it in a write and won't cause a long pause.
// Other writes can be queued behind this write in any case.
conditionalExtendFile();
}
/**
* Writes the given raw payload to Azure Storage at the current blob
* offset.
*/
private void writePayloadToServer(byte[] rawPayload) {
final ByteArrayInputStream wrapperStream =
new ByteArrayInputStream(rawPayload);
LOG.debug("writing payload of " + rawPayload.length + " bytes to Azure page blob");
try {
long start = System.currentTimeMillis();
blob.uploadPages(wrapperStream, currentBlobOffset, rawPayload.length,
withMD5Checking(), PageBlobOutputStream.this.opContext);
long end = System.currentTimeMillis();
LOG.trace("Azure uploadPages time for " + rawPayload.length + " bytes = " + (end - start));
} catch (IOException ex) {
LOG.debug(ExceptionUtils.getStackTrace(ex));
lastError = ex;
} catch (StorageException ex) {
LOG.debug(ExceptionUtils.getStackTrace(ex));
lastError = new IOException(ex);
}
if (lastError != null) {
LOG.debug("Caught error in PageBlobOutputStream#writePayloadToServer()");
}
}
}
private synchronized void flushIOBuffers() {
if (outBuffer.size() == 0) {
return;
}
lastQueuedTask = new WriteRequest(outBuffer.toByteArray());
ioThreadPool.execute(lastQueuedTask);
outBuffer = new ByteArrayOutputStream();
}
@VisibleForTesting
synchronized void waitForLastFlushCompletion() throws IOException {
try {
if (lastQueuedTask != null) {
lastQueuedTask.waitTillDone();
}
} catch (InterruptedException e1) {
// Restore the interrupted status
Thread.currentThread().interrupt();
}
}
/**
* Extend the page blob file if we are close to the end.
*/
private void conditionalExtendFile() {
// maximum allowed size of an Azure page blob (1 terabyte)
final long MAX_PAGE_BLOB_SIZE = 1024L * 1024L * 1024L * 1024L;
// If blob is already at the maximum size, then don't try to extend it.
if (currentBlobSize == MAX_PAGE_BLOB_SIZE) {
return;
}
// If we are within the maximum write size of the end of the file,
if (currentBlobSize - currentBlobOffset <= MAX_RAW_BYTES_PER_REQUEST) {
// Extend the file. Retry up to 3 times with back-off.
CloudPageBlob cloudPageBlob = (CloudPageBlob) blob.getBlob();
long newSize = currentBlobSize + configuredPageBlobExtensionSize;
// Make sure we don't exceed maximum blob size.
if (newSize > MAX_PAGE_BLOB_SIZE) {
newSize = MAX_PAGE_BLOB_SIZE;
}
final int MAX_RETRIES = 3;
int retries = 1;
boolean resizeDone = false;
while(!resizeDone && retries <= MAX_RETRIES) {
try {
cloudPageBlob.resize(newSize);
resizeDone = true;
currentBlobSize = newSize;
} catch (StorageException e) {
LOG.warn("Failed to extend size of " + cloudPageBlob.getUri());
try {
// sleep 2, 8, 18 seconds for up to 3 retries
Thread.sleep(2000 * retries * retries);
} catch (InterruptedException e1) {
// Restore the interrupted status
Thread.currentThread().interrupt();
}
} finally {
retries++;
}
}
}
}
/**
* Flushes this output stream and forces any buffered output bytes to be
* written out. If any data remains in the buffer it is committed to the
* service. Data is queued for writing but not forced out to the service
* before the call returns.
*/
@Override
public void flush() throws IOException {
checkStreamState();
flushIOBuffers();
}
/**
* Writes b.length bytes from the specified byte array to this output stream.
*
* @param data
* the byte array to write.
*
* @throws IOException
* if an I/O error occurs. In particular, an IOException may be
* thrown if the output stream has been closed.
*/
@Override
public void write(final byte[] data) throws IOException {
write(data, 0, data.length);
}
/**
* Writes length bytes from the specified byte array starting at offset to
* this output stream.
*
* @param data
* the byte array to write.
* @param offset
* the start offset in the data.
* @param length
* the number of bytes to write.
* @throws IOException
* if an I/O error occurs. In particular, an IOException may be
* thrown if the output stream has been closed.
*/
@Override
public void write(final byte[] data, final int offset, final int length)
throws IOException {
if (offset < 0 || length < 0 || length > data.length - offset) {
throw new IndexOutOfBoundsException();
}
writeInternal(data, offset, length);
}
/**
* Writes the specified byte to this output stream. The general contract for
* write is that one byte is written to the output stream. The byte to be
* written is the eight low-order bits of the argument b. The 24 high-order
* bits of b are ignored.
*
* @param byteVal
* the byteValue to write.
* @throws IOException
* if an I/O error occurs. In particular, an IOException may be
* thrown if the output stream has been closed.
*/
@Override
public void write(final int byteVal) throws IOException {
write(new byte[] { (byte) (byteVal & 0xFF) });
}
/**
* Writes the data to the buffer and triggers writes to the service as needed.
*
* @param data
* the byte array to write.
* @param offset
* the start offset in the data.
* @param length
* the number of bytes to write.
* @throws IOException
* if an I/O error occurs. In particular, an IOException may be
* thrown if the output stream has been closed.
*/
private synchronized void writeInternal(final byte[] data, int offset,
int length) throws IOException {
while (length > 0) {
checkStreamState();
final int availableBufferBytes = MAX_DATA_BYTES_PER_REQUEST
- this.outBuffer.size();
final int nextWrite = Math.min(availableBufferBytes, length);
outBuffer.write(data, offset, nextWrite);
offset += nextWrite;
length -= nextWrite;
if (outBuffer.size() > MAX_DATA_BYTES_PER_REQUEST) {
throw new RuntimeException("Internal error: maximum write size " +
Integer.toString(MAX_DATA_BYTES_PER_REQUEST) + "exceeded.");
}
if (outBuffer.size() == MAX_DATA_BYTES_PER_REQUEST) {
flushIOBuffers();
}
}
}
/**
* Force all data in the output stream to be written to Azure storage.
* Wait to return until this is complete.
*/
@Override
public synchronized void hsync() throws IOException {
LOG.debug("Entering PageBlobOutputStream#hsync().");
long start = System.currentTimeMillis();
flush();
LOG.debug(ioThreadPool.toString());
try {
if (lastQueuedTask != null) {
lastQueuedTask.waitTillDone();
}
} catch (InterruptedException e1) {
// Restore the interrupted status
Thread.currentThread().interrupt();
}
LOG.debug("Leaving PageBlobOutputStream#hsync(). Total hsync duration = "
+ (System.currentTimeMillis() - start) + " msec.");
}
@Override
public void hflush() throws IOException {
// hflush is required to force data to storage, so call hsync,
// which does that.
hsync();
}
@Deprecated
public void sync() throws IOException {
// Sync has been deprecated in favor of hflush.
hflush();
}
// For unit testing purposes: kill the IO threads.
@VisibleForTesting
void killIoThreads() {
ioThreadPool.shutdownNow();
}
}