Github user jinfengni commented on a diff in the pull request:
https://github.com/apache/drill/pull/156#discussion_r39463605
--- Diff:
exec/java-exec/src/main/java/org/apache/drill/exec/planner/logical/partition/PruneScanRule.java
---
@@ -176,81 +177,103 @@ protected void doOnMatch(RelOptRuleCall call,
DrillFilterRel filterRel, DrillPro
RexNode pruneCondition = c.getFinalCondition();
if (pruneCondition == null) {
+ logger.debug("No conditions were found eligible for partition
pruning.");
return;
}
// set up the partitions
- final GroupScan groupScan = scanRel.getGroupScan();
- List<PartitionLocation> partitions = descriptor.getPartitions();
-
- if (partitions.size() > Character.MAX_VALUE) {
- return;
- }
-
- final NullableBitVector output = new
NullableBitVector(MaterializedField.create("", Types.optional(MinorType.BIT)),
allocator);
- final VectorContainer container = new VectorContainer();
-
- try {
- final ValueVector[] vectors = new
ValueVector[descriptor.getMaxHierarchyLevel()];
- for (int partitionColumnIndex :
BitSets.toIter(partitionColumnBitSet)) {
- SchemaPath column =
SchemaPath.getSimplePath(fieldNameMap.get(partitionColumnIndex));
- MajorType type = descriptor.getVectorType(column, settings);
- MaterializedField field = MaterializedField.create(column, type);
- ValueVector v = TypeHelper.getNewVector(field, allocator);
- v.allocateNew();
- vectors[partitionColumnIndex] = v;
- container.add(v);
- }
-
- // populate partition vectors.
- descriptor.populatePartitionVectors(vectors, partitions,
partitionColumnBitSet, fieldNameMap);
-
- // materialize the expression
- logger.debug("Attempting to prune {}", pruneCondition);
- final LogicalExpression expr = DrillOptiq.toDrill(new
DrillParseContext(settings), scanRel, pruneCondition);
- final ErrorCollectorImpl errors = new ErrorCollectorImpl();
-
- LogicalExpression materializedExpr =
ExpressionTreeMaterializer.materialize(expr, container, errors,
optimizerContext.getFunctionRegistry());
- // Make sure pruneCondition's materialized expression is always of
BitType, so that
- // it's same as the type of output vector.
- if (materializedExpr.getMajorType().getMode() ==
TypeProtos.DataMode.REQUIRED) {
- materializedExpr =
ExpressionTreeMaterializer.convertToNullableType(
- materializedExpr,
- materializedExpr.getMajorType().getMinorType(),
- optimizerContext.getFunctionRegistry(),
- errors);
+ List<String> newFiles = Lists.newArrayList();
+ long numTotal = 0; // total number of partitions
+ int batchIndex = 0;
+ String firstLocation = null;
+
+ // Outer loop: iterate over a list of batches of PartitionLocations
+ for (List<PartitionLocation> partitions : descriptor) {
+ numTotal += partitions.size();
+ logger.debug("Evaluating partition pruning for batch {}",
batchIndex);
+ if (batchIndex == 0) { // save the first location in case everything
is pruned
+ firstLocation = partitions.get(0).getEntirePartitionLocation();
}
+ final NullableBitVector output = new
NullableBitVector(MaterializedField.create("", Types.optional(MinorType.BIT)),
allocator);
+ final VectorContainer container = new VectorContainer();
+
+ try {
+ final ValueVector[] vectors = new
ValueVector[descriptor.getMaxHierarchyLevel()];
+ for (int partitionColumnIndex :
BitSets.toIter(partitionColumnBitSet)) {
+ SchemaPath column =
SchemaPath.getSimplePath(fieldNameMap.get(partitionColumnIndex));
+ MajorType type = descriptor.getVectorType(column, settings);
+ MaterializedField field = MaterializedField.create(column, type);
+ ValueVector v = TypeHelper.getNewVector(field, allocator);
+ v.allocateNew();
+ vectors[partitionColumnIndex] = v;
+ container.add(v);
+ }
- if (errors.getErrorCount() != 0) {
- logger.warn("Failure while materializing expression [{}]. Errors:
{}", expr, errors);
- }
+ // populate partition vectors.
+ descriptor.populatePartitionVectors(vectors, partitions,
partitionColumnBitSet, fieldNameMap);
+
+ // materialize the expression
+ logger.debug("Attempting to prune {}", pruneCondition);
+ final LogicalExpression expr = DrillOptiq.toDrill(new
DrillParseContext(settings), scanRel, pruneCondition);
+ final ErrorCollectorImpl errors = new ErrorCollectorImpl();
+
+ LogicalExpression materializedExpr =
ExpressionTreeMaterializer.materialize(expr, container, errors,
optimizerContext.getFunctionRegistry());
+ // Make sure pruneCondition's materialized expression is always of
BitType, so that
+ // it's same as the type of output vector.
+ if (materializedExpr.getMajorType().getMode() ==
TypeProtos.DataMode.REQUIRED) {
+ materializedExpr =
ExpressionTreeMaterializer.convertToNullableType(
+ materializedExpr,
+ materializedExpr.getMajorType().getMinorType(),
+ optimizerContext.getFunctionRegistry(),
+ errors);
+ }
- output.allocateNew(partitions.size());
- InterpreterEvaluator.evaluate(partitions.size(), optimizerContext,
container, output, materializedExpr);
- int record = 0;
+ if (errors.getErrorCount() != 0) {
--- End diff --
If expression materializer reports error, is it better to stop the
execution of partition pruning rule, and raise Exception here? In such case, I
feel it's likely that the Interperter would hit error as well.
Also, is the condition expression same across multiple sub-list of
partition locations? If that's the case, is it better to move the logic of
expression materialization out of this for loop? We do not have to do
materialization every time when we process one sublist.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
