Skip to content

TEZ-4745: Optimize some write paths in UnorderedPartitionedKVWriter #534

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Aggarwal-Raghav wants to merge 3 commits into
base: master
Choose a base branch
from
Open

TEZ-4745: Optimize some write paths in UnorderedPartitionedKVWriter #534

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9055e5d
TEZ-4745: Optimize some write paths in UnorderedPartitionedKVWriter
Aggarwal-Raghav Aug 24, 2026
bd9a32e
fix checkstyle
Aggarwal-Raghav Aug 25, 2026
26dec99
further optimization
Aggarwal-Raghav Aug 25, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
76 changes: 43 additions & 33 deletions ...main/java/org/apache/tez/runtime/library/common/writers/UnorderedPartitionedKVWriter.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -407,25 +407,29 @@ public void write(Object key, Object value) throws IOException {

@SuppressWarnings("unchecked")
private void write(Object key, Object value, int partition) throws IOException {
WrappedBuffer locBuffer = this.currentBuffer;

// Wrap to 4 byte (Int) boundary for metaData
int mod = currentBuffer.nextPosition % INT_SIZE;
// Equivalent to modulo 4 (locBuffer.nextPosition % INT_SIZE)
int mod = locBuffer.nextPosition & (INT_SIZE - 1);
int metaSkip = mod == 0 ? 0 : (INT_SIZE - mod);
if ((currentBuffer.availableSize < (META_SIZE + metaSkip)) || (currentBuffer.full)) {
if ((locBuffer.availableSize < (META_SIZE + metaSkip)) || (locBuffer.full)) {
// Move over to the next buffer.
metaSkip = 0;
setupNextBuffer();
locBuffer = this.currentBuffer;
}
currentBuffer.nextPosition += metaSkip;
int metaStart = currentBuffer.nextPosition;
currentBuffer.availableSize -= (META_SIZE + metaSkip);
currentBuffer.nextPosition += META_SIZE;
locBuffer.nextPosition += metaSkip;
int metaStart = locBuffer.nextPosition;
locBuffer.availableSize -= (META_SIZE + metaSkip);
locBuffer.nextPosition += META_SIZE;

keySerializer.serialize(key);

if (currentBuffer.full) {
if (locBuffer.full) {
if (metaStart == 0) { // Started writing at the start of the buffer. Write Key to disk.
// Key too large for any buffer. Write entire record to disk.
currentBuffer.reset();
locBuffer.reset();
writeLargeRecord(key, value, partition);
return;
} else { // Exceeded length on current buffer.
Expand All @@ -437,14 +441,14 @@ private void write(Object key, Object value, int partition) throws IOException {
}
}

int valStart = currentBuffer.nextPosition;
int valStart = locBuffer.nextPosition;
valSerializer.serialize(value);

if (currentBuffer.full) {
if (locBuffer.full) {
// Value too large for current buffer, or K-V too large for entire buffer.
if (metaStart == 0) {
// Key + Value too large for a single buffer.
currentBuffer.reset();
locBuffer.reset();
writeLargeRecord(key, value, partition);
return;
} else { // Exceeded length on current buffer.
Expand All @@ -455,26 +459,30 @@ private void write(Object key, Object value, int partition) throws IOException {
}
}

// Meta-data updates
int metaIndex = metaStart / INT_SIZE;
int indexNext = currentBuffer.partitionPositions[partition];
int pos = locBuffer.nextPosition;

currentBuffer.metaBuffer.put(metaIndex + INDEX_KEYLEN, (valStart - (metaStart + META_SIZE)));
currentBuffer.metaBuffer.put(metaIndex + INDEX_VALLEN, (currentBuffer.nextPosition - valStart));
currentBuffer.metaBuffer.put(metaIndex + INDEX_NEXT, indexNext);
currentBuffer.skipSize += metaSkip; // For size estimation
// Meta-data updates
// Equivalent to division by 4 (metaStart / INT_SIZE)
int metaIndex = metaStart >>> 2;
int indexNext = locBuffer.partitionPositions[partition];
int keyLength = valStart - (metaStart + META_SIZE);
int valLength = pos - valStart;
int recordBytes = keyLength + valLength;

locBuffer.metaBuffer.put(metaIndex + INDEX_KEYLEN, keyLength);
locBuffer.metaBuffer.put(metaIndex + INDEX_VALLEN, valLength);
locBuffer.metaBuffer.put(metaIndex + INDEX_NEXT, indexNext);
locBuffer.skipSize += metaSkip; // For size estimation
// Update stats on number of records
localOutputRecordBytesCounter += (currentBuffer.nextPosition - (metaStart + META_SIZE));
localOutputBytesWithOverheadCounter += ((currentBuffer.nextPosition - metaStart) + metaSkip);
localOutputRecordsCounter++;
if (localOutputRecordBytesCounter % NOTIFY_THRESHOLD == 0) {
localOutputRecordBytesCounter += recordBytes;
localOutputBytesWithOverheadCounter += recordBytes + (META_SIZE + metaSkip);
if (++localOutputRecordsCounter == NOTIFY_THRESHOLD) {
updateTezCountersAndNotify();
}
currentBuffer.partitionPositions[partition] = metaStart;
currentBuffer.recordsPerPartition[partition]++;
currentBuffer.sizePerPartition[partition] +=
currentBuffer.nextPosition - (metaStart + META_SIZE);
currentBuffer.numRecords++;
locBuffer.partitionPositions[partition] = metaStart;
locBuffer.recordsPerPartition[partition]++;
locBuffer.sizePerPartition[partition] += recordBytes;
locBuffer.numRecords++;

}

Expand Down Expand Up @@ -689,16 +697,18 @@ protected SpillResult callInternal() throws IOException {
private long writePartition(int pos, WrappedBuffer wrappedBuffer, Writer writer,
DataInputBuffer keyBuffer, DataInputBuffer valBuffer) throws IOException {
long numRecords = 0;
IntBuffer metaBuffer = wrappedBuffer.metaBuffer;
byte[] buffer = wrappedBuffer.buffer;
while (pos != WrappedBuffer.PARTITION_ABSENT_POSITION) {
int metaIndex = pos / INT_SIZE;
int keyLength = wrappedBuffer.metaBuffer.get(metaIndex + INDEX_KEYLEN);
int valLength = wrappedBuffer.metaBuffer.get(metaIndex + INDEX_VALLEN);
keyBuffer.reset(wrappedBuffer.buffer, pos + META_SIZE, keyLength);
valBuffer.reset(wrappedBuffer.buffer, pos + META_SIZE + keyLength, valLength);
int metaIndex = pos >>> 2;
int keyLength = metaBuffer.get(metaIndex + INDEX_KEYLEN);
int valLength = metaBuffer.get(metaIndex + INDEX_VALLEN);
keyBuffer.reset(buffer, pos + META_SIZE, keyLength);
valBuffer.reset(buffer, pos + META_SIZE + keyLength, valLength);

writer.append(keyBuffer, valBuffer);
numRecords++;
pos = wrappedBuffer.metaBuffer.get(metaIndex + INDEX_NEXT);
pos = metaBuffer.get(metaIndex + INDEX_NEXT);
}
return numRecords;
}
Expand Down