HADOOP-18347. S3A Vectored IO to use bounded thread pool. (#4918)

part of HADOOP-18103.

Also introducing a config fs.s3a.vectored.active.ranged.reads
to configure the maximum number of number of range reads a
single input stream can have active (downloading, or queued)
to the central FileSystem instance's pool of queued operations.
This stops a single stream overloading the shared thread pool.

Contributed by: Mukund Thakur
 Conflicts:
	hadoop-tools/hadoop-aws/src/main/java/org/apache/hadoop/fs/s3a/Constants.java
	hadoop-tools/hadoop-aws/src/main/java/org/apache/hadoop/fs/s3a/S3AFileSystem.java
This commit is contained in:
Mukund Thakur 2022-09-27 21:13:07 +05:30
parent e5a566c91f
commit bbe841e601
5 changed files with 55 additions and 16 deletions

View File

@ -84,7 +84,7 @@ public interface StreamCapabilities {
* Support for vectored IO api.
* See {@code PositionedReadable#readVectored(List, IntFunction)}.
*/
String VECTOREDIO = "readvectored";
String VECTOREDIO = "in:readvectored";
/**
* Stream abort() capability implemented by {@link Abortable#abort()}.

View File

@ -1203,4 +1203,23 @@ public final class Constants {
* Default maximum read size in bytes during vectored reads : {@value}.
*/
public static final int DEFAULT_AWS_S3_VECTOR_READS_MAX_MERGED_READ_SIZE = 1253376; //1M
/**
* Maximum number of range reads a single input stream can have
* active (downloading, or queued) to the central FileSystem
* instance's pool of queued operations.
* This stops a single stream overloading the shared thread pool.
* {@value}
* <p>
* Default is {@link #DEFAULT_AWS_S3_VECTOR_ACTIVE_RANGE_READS}
*/
public static final String AWS_S3_VECTOR_ACTIVE_RANGE_READS =
"fs.s3a.vectored.active.ranged.reads";
/**
* Limit of queued range data download operations during vectored
* read. Value: {@value}
*/
public static final int DEFAULT_AWS_S3_VECTOR_ACTIVE_RANGE_READS = 4;
}

View File

@ -319,6 +319,12 @@ public class S3AFileSystem extends FileSystem implements StreamCapabilities,
/** Vectored IO context. */
private VectoredIOContext vectoredIOContext;
/**
* Maximum number of active range read operation a single
* input stream can have.
*/
private int vectoredActiveRangeReads;
private long readAhead;
private ChangeDetectionPolicy changeDetectionPolicy;
private final AtomicBoolean closed = new AtomicBoolean(false);
@ -597,6 +603,8 @@ public class S3AFileSystem extends FileSystem implements StreamCapabilities,
longBytesOption(conf, ASYNC_DRAIN_THRESHOLD,
DEFAULT_ASYNC_DRAIN_THRESHOLD, 0),
inputPolicy);
vectoredActiveRangeReads = intOption(conf,
AWS_S3_VECTOR_ACTIVE_RANGE_READS, DEFAULT_AWS_S3_VECTOR_ACTIVE_RANGE_READS, 1);
vectoredIOContext = populateVectoredIOContext(conf);
} catch (AmazonClientException e) {
// amazon client exception: stop all services then throw the translation
@ -1504,7 +1512,10 @@ public class S3AFileSystem extends FileSystem implements StreamCapabilities,
createObjectAttributes(path, fileStatus),
createInputStreamCallbacks(auditSpan),
inputStreamStats,
unboundedThreadPool));
new SemaphoredDelegatingExecutor(
boundedThreadPool,
vectoredActiveRangeReads,
true)));
}
/**

View File

@ -27,7 +27,7 @@ import java.net.SocketTimeoutException;
import java.nio.ByteBuffer;
import java.util.List;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.ThreadPoolExecutor;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.function.IntFunction;
@ -139,7 +139,7 @@ public class S3AInputStream extends FSInputStream implements CanSetReadahead,
/**
* Thread pool used for vectored IO operation.
*/
private final ThreadPoolExecutor unboundedThreadPool;
private final ExecutorService boundedThreadPool;
private final String bucket;
private final String key;
private final String pathStr;
@ -196,13 +196,13 @@ public class S3AInputStream extends FSInputStream implements CanSetReadahead,
* @param s3Attributes object attributes
* @param client S3 client to use
* @param streamStatistics stream io stats.
* @param unboundedThreadPool thread pool to use.
* @param boundedThreadPool thread pool to use.
*/
public S3AInputStream(S3AReadOpContext ctx,
S3ObjectAttributes s3Attributes,
InputStreamCallbacks client,
S3AInputStreamStatistics streamStatistics,
ThreadPoolExecutor unboundedThreadPool) {
ExecutorService boundedThreadPool) {
Preconditions.checkArgument(isNotEmpty(s3Attributes.getBucket()),
"No Bucket");
Preconditions.checkArgument(isNotEmpty(s3Attributes.getKey()), "No Key");
@ -224,7 +224,7 @@ public class S3AInputStream extends FSInputStream implements CanSetReadahead,
setInputPolicy(ctx.getInputPolicy());
setReadahead(ctx.getReadahead());
this.asyncDrainThreshold = ctx.getAsyncDrainThreshold();
this.unboundedThreadPool = unboundedThreadPool;
this.boundedThreadPool = boundedThreadPool;
this.vectoredIOContext = context.getVectoredIOContext();
this.threadIOStatistics = requireNonNull(ctx.getIOStatisticsAggregator());
}
@ -882,7 +882,7 @@ public class S3AInputStream extends FSInputStream implements CanSetReadahead,
streamStatistics.readVectoredOperationStarted(sortedRanges.size(), sortedRanges.size());
for (FileRange range: sortedRanges) {
ByteBuffer buffer = allocate.apply(range.getLength());
unboundedThreadPool.submit(() -> readSingleRange(range, buffer));
boundedThreadPool.submit(() -> readSingleRange(range, buffer));
}
} else {
LOG.debug("Trying to merge the ranges as they are not disjoint");
@ -893,7 +893,7 @@ public class S3AInputStream extends FSInputStream implements CanSetReadahead,
LOG.debug("Number of original ranges size {} , Number of combined ranges {} ",
ranges.size(), combinedFileRanges.size());
for (CombinedFileRange combinedFileRange: combinedFileRanges) {
unboundedThreadPool.submit(
boundedThreadPool.submit(
() -> readCombinedRangeAndUpdateChildren(combinedFileRange, allocate));
}
}

View File

@ -75,13 +75,22 @@ on the client requirements.
</description>
</property>
<property>
<name>fs.s3a.vectored.read.max.merged.size</name>
<value>1M</value>
<description>
What is the largest merged read size in bytes such
that we group ranges together during vectored read.
Setting this value to 0 will disable merging of ranges.
</description>
<name>fs.s3a.vectored.read.max.merged.size</name>
<value>1M</value>
<description>
What is the largest merged read size in bytes such
that we group ranges together during vectored read.
Setting this value to 0 will disable merging of ranges.
</description>
<property>
<name>fs.s3a.vectored.active.ranged.reads</name>
<value>4</value>
<description>
Maximum number of range reads a single input stream can have
active (downloading, or queued) to the central FileSystem
instance's pool of queued operations.
This stops a single stream overloading the shared thread pool.
</description>
</property>
```