Skip to content
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

[VL] Creation of Java Arrow complex type vector leaks memory #1084

Merged
merged 2 commits into from
Mar 8, 2023
Merged

[VL] Creation of Java Arrow complex type vector leaks memory #1084

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
1485ad9
Creation of Java Arrow complex type vector leaks memory
zhztheplayer Mar 7, 2023
d4d14e1
style
zhztheplayer Mar 7, 2023
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
2 changes: 1 addition & 1 deletion backends-velox/src/test/scala/io/glutenproject/execution/VeloxDataTypeValidationSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ class VeloxDataTypeValidationSuite extends WholeStageTransformerSuite {
.set("spark.sql.files.maxPartitionBytes", "1g")
.set("spark.sql.shuffle.partitions", "1")
.set("spark.memory.offHeap.size", "2g")
.set("spark.unsafe.exceptionOnMemoryLeak", "false")
.set("spark.unsafe.exceptionOnMemoryLeak", "true")
.set("spark.sql.autoBroadcastJoinThreshold", "10M")
.set("spark.sql.sources.useV1SourceList", "avro")
}
Expand Down
6 changes: 3 additions & 3 deletions backends-velox/src/test/scala/io/glutenproject/execution/VeloxStringFunctionsSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ class VeloxStringFunctionsSuite extends WholeStageTransformerSuite {
.set("spark.sql.files.maxPartitionBytes", "1g")
.set("spark.sql.shuffle.partitions", "1")
.set("spark.memory.offHeap.size", "2g")
.set("spark.unsafe.exceptionOnMemoryLeak", "false")
.set("spark.unsafe.exceptionOnMemoryLeak", "true")
.set("spark.sql.autoBroadcastJoinThreshold", "-1")
.set("spark.sql.sources.useV1SourceList", "avro")
.set("spark.sql.optimizer.excludedRules", ConstantFolding.ruleName + "," +
Expand Down Expand Up @@ -289,7 +289,7 @@ class VeloxStringFunctionsSuite extends WholeStageTransformerSuite {
s"from $LINEITEM_TABLE limit $LENGTH") { checkOperatorMatch[ProjectExecTransformer] }
runQueryAndCompare(s"select l_orderkey, like(l_comment, 'a_%b') " +
s"from $LINEITEM_TABLE limit $LENGTH") { checkOperatorMatch[ProjectExecTransformer] }
runQueryAndCompare(s"select l_orderkey, like('l_comment', 'a\\__b') " +
runQueryAndCompare(s"select l_orderkey, like(l_comment, 'a\\__b') " +
s"from $LINEITEM_TABLE limit $LENGTH") { checkOperatorMatch[ProjectExecTransformer] }
runQueryAndCompare(s"select l_orderkey, like(l_comment, 'abc_') " +
s"from $LINEITEM_TABLE limit $LENGTH") { checkOperatorMatch[ProjectExecTransformer] }
Expand Down Expand Up @@ -353,7 +353,7 @@ class VeloxStringFunctionsSuite extends WholeStageTransformerSuite {
}

test("regexp_extract_all") {
runQueryAndCompare(s"select l_orderkey, regexp_extract_all('l_comment', '([a-z])', 1) " +
runQueryAndCompare(s"select l_orderkey, regexp_extract_all(l_comment, '([a-z])', 1) " +
s"from $LINEITEM_TABLE limit 5") { checkOperatorMatch[ProjectExecTransformer] }
// fall back because of unsupported cast(array)
runQueryAndCompare(s"select l_orderkey, l_comment, " +
Expand Down
28 changes: 17 additions & 11 deletions gluten-data/src/main/java/io/glutenproject/vectorized/ArrowWritableColumnVector.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,6 +48,7 @@
import org.apache.arrow.vector.types.pojo.Schema;
import org.apache.spark.sql.catalyst.util.DateTimeUtils;
import org.apache.spark.sql.execution.datasources.v2.arrow.SparkSchemaUtil;
import org.apache.spark.sql.execution.vectorized.WritableColumnVector;
import org.apache.spark.sql.execution.vectorized.WritableColumnVectorShim;
import org.apache.spark.sql.types.ArrayType;
import org.apache.spark.sql.types.DataType;
Expand Down Expand Up @@ -244,7 +245,7 @@ private void createVectorAccessor(ValueVector vector, ValueVector dictionary) {
} else if (vector instanceof MapVector) {
MapVector mapVector = (MapVector) vector;
accessor = new MapAccessor(mapVector);
childColumns = new ArrowWritableColumnVector[2];
reallocateChildColumns(2);
final StructVector structVector = (StructVector) mapVector.getDataVector();
final FieldVector keyChild = structVector.getChild(MapVector.KEY_NAME);
final FieldVector valueChild = structVector.getChild(MapVector.VALUE_NAME);
Expand All @@ -255,14 +256,14 @@ private void createVectorAccessor(ValueVector vector, ValueVector dictionary) {
} else if (vector instanceof ListVector) {
ListVector listVector = (ListVector) vector;
accessor = new ArrayAccessor(listVector);
childColumns = new ArrowWritableColumnVector[1];
reallocateChildColumns(1);
childColumns[0] = new ArrowWritableColumnVector(
listVector.getDataVector(), 0, listVector.size(), false);
} else if (vector instanceof StructVector) {
StructVector structVector = (StructVector) vector;
accessor = new StructAccessor(structVector);

childColumns = new ArrowWritableColumnVector[structVector.size()];
reallocateChildColumns(structVector.size());
for (int i = 0; i < childColumns.length; ++i) {
childColumns[i] = new ArrowWritableColumnVector(structVector.getVectorById(i),
i, structVector.size(), false);
Expand All @@ -275,6 +276,19 @@ private void createVectorAccessor(ValueVector vector, ValueVector dictionary) {
}
}

// The child columns may already be created in super class's constructor
// org.apache.spark.sql.execution.vectorized
// .WritableColumnVector#WritableColumnVector(int, DataType).
// So we close them then create new ones.
private void reallocateChildColumns(int width) {
if (childColumns != null) {
for (WritableColumnVector column : childColumns) {
column.close();
}
}
childColumns = new ArrowWritableColumnVector[width];
}

private ArrowVectorWriter createVectorWriter(ValueVector vector) {
if (vector instanceof BitVector) {
return new BooleanWriter((BitVector) vector);
Expand Down Expand Up @@ -355,14 +369,6 @@ public void close() {
closed = true;
vectorCount.getAndDecrement();
super.close();
// TODO: close Arrow Allocated Memory
if (childColumns != null) {
for (int i = 0; i < childColumns.length; i++) {
childColumns[i].close();
Copy link
Contributor

Choose a reason for hiding this comment

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

Why we delete here?

Copy link
Member Author

Choose a reason for hiding this comment

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

It's included in super.close()

childColumns[i] = null;
}
childColumns = null;
}
vector.close();
if (dictionaryVector != null) {
dictionaryVector.close();
Expand Down
8 changes: 5 additions & 3 deletions gluten-data/src/main/scala/io/glutenproject/execution/GlutenColumnarRules.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@

package io.glutenproject.execution

import io.glutenproject.backendsapi.BackendsApiManager
import io.glutenproject.columnarbatch.ArrowColumnarBatches
import io.glutenproject.memory.arrowalloc.ArrowBufferAllocators
import io.glutenproject.utils.{LogicalPlanSelector, QueryPlanSelector}
Expand Down Expand Up @@ -132,9 +133,10 @@ case class LoadArrowData(child: SparkPlan) extends UnaryExecNode {

override protected def doExecuteColumnar(): RDD[ColumnarBatch] = {
child.executeColumnar().mapPartitions { itr =>
itr.map { cb =>
ArrowColumnarBatches.ensureLoaded(ArrowBufferAllocators.contextInstance(), cb)
}
BackendsApiManager.getIteratorApiInstance.genCloseableColumnBatchIterator(
itr.map { cb =>
ArrowColumnarBatches.ensureLoaded(ArrowBufferAllocators.contextInstance(), cb)
})
}
}

Expand Down