Skip to content

[PIXELS-1367] Align PixelsBlockLoader with Trino 466 native blocks #178

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
gengdy1545 wants to merge 2 commits into
base: master
Choose a base branch
from
Open

[PIXELS-1367] Align PixelsBlockLoader with Trino 466 native blocks #178

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d232fcf
fix: align Pixels block loading with Trino 466
gengdy1545 Aug 3, 2026
0604b52
refactor: build VariableWidthBlock directly in PixelsBlockLoader
gengdy1545 Aug 4, 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
62 changes: 50 additions & 12 deletions connector/src/main/java/io/pixelsdb/pixels/trino/PixelsBlockLoader.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,9 +22,6 @@
import io.airlift.slice.Slices;
import io.pixelsdb.pixels.core.TypeDescription;
import io.pixelsdb.pixels.core.vector.*;
import io.pixelsdb.pixels.trino.block.TimeArrayBlock;
import io.pixelsdb.pixels.trino.block.VarcharArrayBlock;
import io.pixelsdb.pixels.trino.block.VarcharArrayBlockEncoding;
import io.trino.spi.block.*;
import io.trino.spi.type.Type;

Expand Down Expand Up @@ -70,7 +67,26 @@ public Block load()
switch (typeCategory)
{
case BYTE:
ByteColumnVector bytecv = (ByteColumnVector) vector;
block = new ByteArrayBlock(batchSize, Optional.ofNullable(bytecv.isNull), bytecv.vector);
break;
case SHORT:
IntColumnVector shortcv = (IntColumnVector) vector;
short[] shortValues = new short[batchSize];
for (int i = 0; i < batchSize; ++i)
{
if (!shortcv.isNull[i])
{
int value = shortcv.vector[i];
if (value < Short.MIN_VALUE || value > Short.MAX_VALUE)
{
throw new IllegalArgumentException("SMALLINT value out of range: " + value);
}
shortValues[i] = (short) value;
}
}
block = new ShortArrayBlock(batchSize, Optional.ofNullable(shortcv.isNull), shortValues);
break;
case INT:
IntColumnVector icv = (IntColumnVector) vector;
block = new IntArrayBlock(batchSize, Optional.ofNullable(icv.isNull), icv.vector);
Expand Down Expand Up @@ -109,9 +125,31 @@ public Block load()
case VARBINARY:
if (vector instanceof BinaryColumnVector scv)
{
block = VarcharArrayBlockEncoding.Instance()
.replacementBlockForWrite(new VarcharArrayBlock(batchSize, scv.vector, scv.start, scv.lens, !scv.noNulls, scv.isNull).getLoadedBlock())
.get();
int totalLength = 0;
for (int i = 0; i < batchSize; ++i)
{
if (!scv.noNulls && scv.isNull[i])
{
continue;
}
totalLength += scv.lens[i];
}
byte[] content = new byte[totalLength];

@bianhq bianhq Aug 5, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

More memory allocation leads to higher GC pressure.

int[] offsets = new int[batchSize + 1];
int curOffset = 0;
for (int i = 0; i < batchSize; ++i)
{
offsets[i] = curOffset;
if (!scv.noNulls && scv.isNull[i])
{
continue;
}
int len = scv.lens[i];
System.arraycopy(scv.vector[i], scv.start[i], content, curOffset, len);

@bianhq bianhq Aug 5, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Additional memory copy is inefficient.

curOffset += len;
}
offsets[batchSize] = curOffset;
block = new VariableWidthBlock(batchSize, Slices.wrappedBuffer(content), offsets, Optional.of(scv.isNull));
} else
{
DictionaryColumnVector dscv = (DictionaryColumnVector) vector;
Expand Down Expand Up @@ -146,12 +184,12 @@ public Block load()
case TIME:
// PIXELS-94: add time type.
TimeColumnVector tcv = (TimeColumnVector) vector;
/**
* In Presto, LongArrayBlock is used for time type. However, in Pixels,
* Time value is stored as int, so here we use TimeArrayBlock, which
* accepts int values but provides getLong method same as LongArrayBlock.
*/
block = new TimeArrayBlock(batchSize, tcv.times, !tcv.noNulls, tcv.isNull);
long[] timeValues = new long[batchSize];
for (int i = 0; i < batchSize; ++i)
{
timeValues[i] = (long) tcv.times[i] * 1_000_000_000L;

@bianhq bianhq Aug 5, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

More memory read is inefficient.

}
block = new LongArrayBlock(batchSize, Optional.ofNullable(tcv.isNull), timeValues);
break;
case TIMESTAMP:
TimestampColumnVector tscv = (TimestampColumnVector) vector;
Expand Down
9 changes: 0 additions & 9 deletions connector/src/main/java/io/pixelsdb/pixels/trino/PixelsPlugin.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,20 +20,11 @@
package io.pixelsdb.pixels.trino;

import com.google.common.collect.ImmutableList;
import io.pixelsdb.pixels.trino.block.TimeArrayBlockEncoding;
import io.pixelsdb.pixels.trino.block.VarcharArrayBlockEncoding;
import io.trino.spi.Plugin;
import io.trino.spi.block.BlockEncoding;
import io.trino.spi.connector.ConnectorFactory;

public class PixelsPlugin implements Plugin
{
@Override
public Iterable<BlockEncoding> getBlockEncodings()
{
return ImmutableList.of(VarcharArrayBlockEncoding.Instance(), TimeArrayBlockEncoding.Instance());
}

@Override
public Iterable<ConnectorFactory> getConnectorFactories()
{
Expand Down
Loading