diff --git common/src/java/org/apache/hadoop/hive/conf/HiveConf.java common/src/java/org/apache/hadoop/hive/conf/HiveConf.java
index d26573e..8637142 100644
--- common/src/java/org/apache/hadoop/hive/conf/HiveConf.java
+++ common/src/java/org/apache/hadoop/hive/conf/HiveConf.java
@@ -550,6 +550,9 @@
// Define the default block padding
HIVE_ORC_DEFAULT_BLOCK_PADDING("hive.exec.orc.default.block.padding",
true),
+ // Define the tolerance for block padding. By default, 5% of stripe size is
+ // tolerated. For 256MB stripe size, maximum padding will be ~13MB.
+ HIVE_ORC_BLOCK_PADDING_TOLERANCE("hive.exec.orc.block.padding.tolerance", 0.05f),
// Define the default compression codec for ORC file
HIVE_ORC_DEFAULT_COMPRESS("hive.exec.orc.default.compress", "ZLIB"),
HIVE_ORC_INCLUDE_FILE_FOOTER_IN_SPLITS("hive.orc.splits.include.file.footer", false),
diff --git conf/hive-default.xml.template conf/hive-default.xml.template
index 8a74e4e..f7af01e 100644
--- conf/hive-default.xml.template
+++ conf/hive-default.xml.template
@@ -1962,6 +1962,14 @@
+ hive.exec.orc.block.padding.tolerance
+ 0.05
+
+ Define the tolerance for block padding.
+
+
+
+
hive.exec.orc.default.compress
ZLIB
diff --git ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcFile.java ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcFile.java
index 7c542c1..cde66d3 100644
--- ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcFile.java
+++ ql/src/java/org/apache/hadoop/hive/ql/io/orc/OrcFile.java
@@ -221,6 +221,7 @@ public static Reader createReader(Path path,
private MemoryManager memoryManagerValue;
private Version versionValue;
private WriterCallback callback;
+ private float paddingTolerance;
WriterOptions(Configuration conf) {
configuration = conf;
@@ -250,6 +251,9 @@ public static Reader createReader(Path path,
} else {
versionValue = Version.byName(versionName);
}
+ paddingTolerance =
+ conf.getFloat(HiveConf.ConfVars.HIVE_ORC_BLOCK_PADDING_TOLERANCE.varname,
+ HiveConf.ConfVars.HIVE_ORC_BLOCK_PADDING_TOLERANCE.defaultFloatVal);
}
/**
@@ -301,6 +305,14 @@ public WriterOptions blockPadding(boolean value) {
}
/**
+ * Sets the tolerance for block padding.
+ */
+ public WriterOptions paddingTolerance(float value) {
+ paddingTolerance = value;
+ return this;
+ }
+
+ /**
* Sets the generic compression that is used to compress the data.
*/
public WriterOptions compress(CompressionKind value) {
@@ -370,7 +382,7 @@ public static Writer createWriter(Path path,
opts.stripeSizeValue, opts.compressValue,
opts.bufferSizeValue, opts.rowIndexStrideValue,
opts.memoryManagerValue, opts.blockPaddingValue,
- opts.versionValue, opts.callback);
+ opts.versionValue, opts.callback, opts.paddingTolerance);
}
/**
diff --git ql/src/java/org/apache/hadoop/hive/ql/io/orc/WriterImpl.java ql/src/java/org/apache/hadoop/hive/ql/io/orc/WriterImpl.java
index b3ab96f..d3ba88a 100644
--- ql/src/java/org/apache/hadoop/hive/ql/io/orc/WriterImpl.java
+++ ql/src/java/org/apache/hadoop/hive/ql/io/orc/WriterImpl.java
@@ -97,13 +97,15 @@
private final FileSystem fs;
private final Path path;
- private final long stripeSize;
+ private long adjustedStripeSize;
+ private final long defaultStripeSize;
private final int rowIndexStride;
private final CompressionKind compress;
private final CompressionCodec codec;
private final boolean addBlockPadding;
private final int bufferSize;
private final long blockSize;
+ private final float paddingTolerance;
// the streams that make up the current stripe
private final Map streams =
new TreeMap();
@@ -146,7 +148,8 @@
MemoryManager memoryManager,
boolean addBlockPadding,
OrcFile.Version version,
- OrcFile.WriterCallback callback) throws IOException {
+ OrcFile.WriterCallback callback,
+ float paddingTolerance) throws IOException {
this.fs = fs;
this.path = path;
this.conf = conf;
@@ -162,11 +165,13 @@ public Writer getWriter() {
} else {
callbackContext = null;
}
- this.stripeSize = stripeSize;
+ this.adjustedStripeSize = stripeSize;
+ this.defaultStripeSize = stripeSize;
this.version = version;
this.addBlockPadding = addBlockPadding;
// pick large block size to minimize block over or under hangs
this.blockSize = Math.min(MAX_BLOCK_SIZE, 2 * stripeSize);
+ this.paddingTolerance = paddingTolerance;
this.compress = compress;
this.bufferSize = bufferSize;
this.rowIndexStride = rowIndexStride;
@@ -211,7 +216,7 @@ static CompressionCodec createCodec(CompressionKind kind) {
@Override
public synchronized boolean checkMemory(double newScale) throws IOException {
- long limit = (long) Math.round(stripeSize * newScale);
+ long limit = (long) Math.round(adjustedStripeSize * newScale);
long size = estimateStripeSize();
if (LOG.isDebugEnabled()) {
LOG.debug("ORC writer " + path + " size = " + size + " limit = " +
@@ -1802,10 +1807,17 @@ private void flushStripe() throws IOException {
// Do we need to pad the file so the stripe doesn't straddle a block
// boundary?
long start = rawWriter.getPos();
- long stripeSize = indexSize + dataSize + footer.getSerializedSize();
+ long currentStripeSize = indexSize + dataSize + footer.getSerializedSize();
+ long available = blockSize - (start % blockSize);
+ float availRatio = (float) available / (float) adjustedStripeSize;
+ if (availRatio > 0.0f && availRatio < 1.0f && availRatio > paddingTolerance) {
+ // adjust default stripe size to fit into remaining space
+ adjustedStripeSize = (long) (availRatio * adjustedStripeSize);
+ }
if (addBlockPadding &&
- stripeSize < blockSize &&
- (start % blockSize) + stripeSize > blockSize) {
+ currentStripeSize < blockSize &&
+ (start % blockSize) + currentStripeSize > blockSize &&
+ availRatio < paddingTolerance ) {
long padding = blockSize - (start % blockSize);
byte[] pad = new byte[(int) Math.min(HDFS_BUFFER_SIZE, padding)];
start += padding;
@@ -1814,6 +1826,9 @@ private void flushStripe() throws IOException {
rawWriter.write(pad, 0, writeLen);
padding -= writeLen;
}
+
+ // after padding reset the default stripe size
+ adjustedStripeSize = defaultStripeSize;
}
// write out the data streams
diff --git ql/src/test/resources/orc-file-dump-dictionary-threshold.out ql/src/test/resources/orc-file-dump-dictionary-threshold.out
index 0a06481..f6e7a50 100644
--- ql/src/test/resources/orc-file-dump-dictionary-threshold.out
+++ ql/src/test/resources/orc-file-dump-dictionary-threshold.out
@@ -90,15 +90,15 @@ Stripes:
Encoding column 1: DIRECT_V2
Encoding column 2: DIRECT_V2
Encoding column 3: DIRECT_V2
- Stripe: offset: 1800000 data: 177935 rows: 1000 tail: 67 index: 813
- Stream: column 0 section ROW_INDEX start: 1800000 length 10
- Stream: column 1 section ROW_INDEX start: 1800010 length 36
- Stream: column 2 section ROW_INDEX start: 1800046 length 39
- Stream: column 3 section ROW_INDEX start: 1800085 length 728
- Stream: column 1 section DATA start: 1800813 length 4007
- Stream: column 2 section DATA start: 1804820 length 8007
- Stream: column 3 section DATA start: 1812827 length 164661
- Stream: column 3 section LENGTH start: 1977488 length 1260
+ Stripe: offset: 1751794 data: 177935 rows: 1000 tail: 67 index: 813
+ Stream: column 0 section ROW_INDEX start: 1751794 length 10
+ Stream: column 1 section ROW_INDEX start: 1751804 length 36
+ Stream: column 2 section ROW_INDEX start: 1751840 length 39
+ Stream: column 3 section ROW_INDEX start: 1751879 length 728
+ Stream: column 1 section DATA start: 1752607 length 4007
+ Stream: column 2 section DATA start: 1756614 length 8007
+ Stream: column 3 section DATA start: 1764621 length 164661
+ Stream: column 3 section LENGTH start: 1929282 length 1260
Encoding column 0: DIRECT
Encoding column 1: DIRECT_V2
Encoding column 2: DIRECT_V2
diff --git ql/src/test/resources/orc-file-dump.out ql/src/test/resources/orc-file-dump.out
index 7a5b907..f46a89b 100644
--- ql/src/test/resources/orc-file-dump.out
+++ ql/src/test/resources/orc-file-dump.out
@@ -80,30 +80,30 @@ Stripes:
Encoding column 1: DIRECT_V2
Encoding column 2: DIRECT_V2
Encoding column 3: DICTIONARY_V2
- Stripe: offset: 200000 data: 63796 rows: 5000 tail: 74 index: 123
- Stream: column 0 section ROW_INDEX start: 200000 length 10
- Stream: column 1 section ROW_INDEX start: 200010 length 35
- Stream: column 2 section ROW_INDEX start: 200045 length 39
- Stream: column 3 section ROW_INDEX start: 200084 length 39
- Stream: column 1 section DATA start: 200123 length 20029
- Stream: column 2 section DATA start: 220152 length 40035
- Stream: column 3 section DATA start: 260187 length 3574
- Stream: column 3 section LENGTH start: 263761 length 25
- Stream: column 3 section DICTIONARY_DATA start: 263786 length 133
+ Stripe: offset: 191881 data: 63796 rows: 5000 tail: 74 index: 123
+ Stream: column 0 section ROW_INDEX start: 191881 length 10
+ Stream: column 1 section ROW_INDEX start: 191891 length 35
+ Stream: column 2 section ROW_INDEX start: 191926 length 39
+ Stream: column 3 section ROW_INDEX start: 191965 length 39
+ Stream: column 1 section DATA start: 192004 length 20029
+ Stream: column 2 section DATA start: 212033 length 40035
+ Stream: column 3 section DATA start: 252068 length 3574
+ Stream: column 3 section LENGTH start: 255642 length 25
+ Stream: column 3 section DICTIONARY_DATA start: 255667 length 133
Encoding column 0: DIRECT
Encoding column 1: DIRECT_V2
Encoding column 2: DIRECT_V2
Encoding column 3: DICTIONARY_V2
- Stripe: offset: 263993 data: 12940 rows: 1000 tail: 71 index: 123
- Stream: column 0 section ROW_INDEX start: 263993 length 10
- Stream: column 1 section ROW_INDEX start: 264003 length 36
- Stream: column 2 section ROW_INDEX start: 264039 length 39
- Stream: column 3 section ROW_INDEX start: 264078 length 38
- Stream: column 1 section DATA start: 264116 length 4007
- Stream: column 2 section DATA start: 268123 length 8007
- Stream: column 3 section DATA start: 276130 length 768
- Stream: column 3 section LENGTH start: 276898 length 25
- Stream: column 3 section DICTIONARY_DATA start: 276923 length 133
+ Stripe: offset: 255874 data: 12940 rows: 1000 tail: 71 index: 123
+ Stream: column 0 section ROW_INDEX start: 255874 length 10
+ Stream: column 1 section ROW_INDEX start: 255884 length 36
+ Stream: column 2 section ROW_INDEX start: 255920 length 39
+ Stream: column 3 section ROW_INDEX start: 255959 length 38
+ Stream: column 1 section DATA start: 255997 length 4007
+ Stream: column 2 section DATA start: 260004 length 8007
+ Stream: column 3 section DATA start: 268011 length 768
+ Stream: column 3 section LENGTH start: 268779 length 25
+ Stream: column 3 section DICTIONARY_DATA start: 268804 length 133
Encoding column 0: DIRECT
Encoding column 1: DIRECT_V2
Encoding column 2: DIRECT_V2