This is an automated email from the ASF dual-hosted git repository.
wchevreuil pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/hbase.git
The following commit(s) were added to refs/heads/master by this push:
new 370efd04775 HBASE-30225 Performance degradation observed on ycsb reads
benchmark after HBASE-29727 (#8353)
370efd04775 is described below
commit 370efd04775cb7169e4657bd5ab1744b4a4ad6b2
Author: Wellington Ramos Chevreuil <[email protected]>
AuthorDate: Thu Jun 18 21:05:58 2026 +0100
HBASE-30225 Performance degradation observed on ycsb reads benchmark after
HBASE-29727 (#8353)
Signed-off-by: Peter Somogyi <[email protected]>
Reviewed-by: Duo Zhang <[email protected]>
Reviewed-by: Vladimir Rodionov <[email protected]>
---
.../hadoop/hbase/io/hfile/BlockCacheKey.java | 18 ++++++++++--------
.../hadoop/hbase/io/hfile/HFileReaderImpl.java | 22 +++++++++++++++++-----
.../hadoop/hbase/io/hfile/HFileWriterImpl.java | 14 ++++++++++++--
3 files changed, 39 insertions(+), 15 deletions(-)
diff --git
a/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/BlockCacheKey.java
b/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/BlockCacheKey.java
index 254f0adbd5f..526a6cded4c 100644
---
a/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/BlockCacheKey.java
+++
b/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/BlockCacheKey.java
@@ -17,11 +17,11 @@
*/
package org.apache.hadoop.hbase.io.hfile;
+import com.google.errorprone.annotations.RestrictedApi;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hbase.io.HeapSize;
import org.apache.hadoop.hbase.util.ClassSize;
import org.apache.hadoop.hbase.util.FastStringPool;
-import org.apache.hadoop.hbase.util.HFileArchiveUtil;
import org.apache.yetus.audience.InterfaceAudience;
/**
@@ -38,8 +38,6 @@ public class BlockCacheKey implements HeapSize,
java.io.Serializable {
private static final FastStringPool CF_NAME_POOL = new FastStringPool();
- // New compressed format using integer file ID (when codec is available)
-
private final String hfileName;
private final String regionName;
@@ -90,6 +88,8 @@ public class BlockCacheKey implements HeapSize,
java.io.Serializable {
* @param offset Offset of the block into the file
* @param isPrimaryReplica Whether this is from primary replica
* @param blockType Type of block
+ * @param archived Whether the file is archived or not. This is
relevant for proper cache
+ * metrics computation.
*/
public BlockCacheKey(String hfileName, String cfName, String regionName,
long offset,
boolean isPrimaryReplica, BlockType blockType, boolean archived) {
@@ -105,17 +105,20 @@ public class BlockCacheKey implements HeapSize,
java.io.Serializable {
/**
* Construct a new BlockCacheKey using a file path. File, column family and
region information
- * will be extracted from the passed path. This should be used when
inserting keys into the cache,
- * so that region cache metrics are recorded properly.
+ * will be extracted from the passed path, so that region cache metrics are
recorded properly.
+ * NOTE: This should be used when inserting keys into the cache from tests
only, as the region and
+ * cf name parsing from the path can be costly and induce a 10% performance
degradation on the
+ * read path, when this parsing is performed on each key creation time.
* @param hfilePath The path to the HFile
* @param offset Offset of the block into the file
* @param isPrimaryReplica Whether this is from primary replica
* @param blockType Type of block
*/
+ @RestrictedApi(explanation = "Should only be called in tests due to costly
path parsing",
+ link = "", allowedOnPath = ".*/src/test/.*")
public BlockCacheKey(Path hfilePath, long offset, boolean isPrimaryReplica,
BlockType blockType) {
this(hfilePath.getName(), hfilePath.getParent().getName(),
- hfilePath.getParent().getParent().getName(), offset, isPrimaryReplica,
blockType,
- HFileArchiveUtil.isHFileArchived(hfilePath));
+ hfilePath.getParent().getParent().getName(), offset, isPrimaryReplica,
blockType, false);
}
@Override
@@ -194,5 +197,4 @@ public class BlockCacheKey implements HeapSize,
java.io.Serializable {
public boolean isArchived() {
return archived;
}
-
}
diff --git
a/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileReaderImpl.java
b/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileReaderImpl.java
index 6e8f11711ae..000fd4acf00 100644
---
a/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileReaderImpl.java
+++
b/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileReaderImpl.java
@@ -52,6 +52,7 @@ import org.apache.hadoop.hbase.nio.ByteBuff;
import org.apache.hadoop.hbase.regionserver.KeyValueScanner;
import org.apache.hadoop.hbase.util.ByteBufferUtils;
import org.apache.hadoop.hbase.util.Bytes;
+import org.apache.hadoop.hbase.util.HFileArchiveUtil;
import org.apache.hadoop.hbase.util.IdLock;
import org.apache.hadoop.hbase.util.ObjectIntPair;
import org.apache.hadoop.io.WritableUtils;
@@ -124,6 +125,14 @@ public abstract class HFileReaderImpl implements
HFile.Reader, Configurable {
/** Minor versions starting with this number have faked index key */
static final int MINOR_VERSION_WITH_FAKED_KEY = 3;
+ private final String fileName;
+
+ private final String region;
+
+ private final String family;
+
+ private final boolean archived;
+
/**
* Opens a HFile.
* @param context Reader context info
@@ -149,6 +158,10 @@ public abstract class HFileReaderImpl implements
HFile.Reader, Configurable {
fsBlockReader.setDataBlockEncoder(dataBlockEncoder, conf);
dataBlockIndexReader = fileInfo.getDataBlockIndexReader();
metaBlockIndexReader = fileInfo.getMetaBlockIndexReader();
+ fileName = path.getName();
+ family = path.getParent().getName();
+ region = path.getParent().getParent().getName();
+ archived = HFileArchiveUtil.isHFileArchived(path);
}
@SuppressWarnings("serial")
@@ -1246,8 +1259,8 @@ public abstract class HFileReaderImpl implements
HFile.Reader, Configurable {
synchronized (metaBlockIndexReader.getRootBlockKey(block)) {
// Check cache for block. If found return.
long metaBlockOffset = metaBlockIndexReader.getRootBlockOffset(block);
- BlockCacheKey cacheKey =
- new BlockCacheKey(path, metaBlockOffset,
this.isPrimaryReplicaReader(), BlockType.META);
+ BlockCacheKey cacheKey = new BlockCacheKey(fileName, family, region,
metaBlockOffset,
+ this.isPrimaryReplicaReader(), BlockType.META, archived);
cacheBlock &=
cacheConf.shouldCacheBlockOnRead(BlockType.META.getCategory(),
getHFileInfo(), conf);
@@ -1336,9 +1349,8 @@ public abstract class HFileReaderImpl implements
HFile.Reader, Configurable {
// the other choice is to duplicate work (which the cache would prevent you
// from doing).
- BlockCacheKey cacheKey =
- new BlockCacheKey(path, dataBlockOffset, this.isPrimaryReplicaReader(),
expectedBlockType);
-
+ BlockCacheKey cacheKey = new BlockCacheKey(fileName, family, region,
dataBlockOffset,
+ this.isPrimaryReplicaReader(), expectedBlockType, archived);
boolean useLock = false;
IdLock.Entry lockEntry = null;
final Span span = Span.current();
diff --git
a/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileWriterImpl.java
b/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileWriterImpl.java
index b635c2cfec6..7435685292f 100644
---
a/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileWriterImpl.java
+++
b/hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileWriterImpl.java
@@ -189,11 +189,21 @@ public class HFileWriterImpl implements HFile.Writer {
private final TimeRangeTracker timeRangeTracker;
private long earliestPutTs = HConstants.LATEST_TIMESTAMP;
+ private String regionName;
+
+ private String familyName;
+
public HFileWriterImpl(final Configuration conf, CacheConfig cacheConf, Path
path,
FSDataOutputStream outputStream, HFileContext fileContext) {
this.outputStream = outputStream;
this.path = path;
- this.name = path != null ? path.getName() : outputStream.toString();
+ if (path != null) {
+ this.name = path.getName();
+ this.regionName = path.getParent().getParent().getName();
+ this.familyName = path.getParent().getName();
+ } else {
+ this.name = outputStream.toString();
+ }
this.hFileContext = fileContext;
// TODO: Move this back to upper layer
this.timeRangeTracker =
TimeRangeTracker.create(TimeRangeTracker.Type.NON_SYNC);
@@ -593,7 +603,7 @@ public class HFileWriterImpl implements HFile.Writer {
private BlockCacheKey buildCacheBlockKey(long offset, BlockType blockType) {
if (path != null) {
- return new BlockCacheKey(path, offset, true, blockType);
+ return new BlockCacheKey(name, familyName, regionName, offset, true,
blockType, false);
}
return new BlockCacheKey(name, offset, true, blockType);
}
