[
https://issues.apache.org/jira/browse/HIVE-24746?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
László Bodor updated HIVE-24746:
--------------------------------
Description:
During range computation, timestamp ranges become a hotspot due to "TimeStamp"
comparisons. It has to construct the entire TimeStamp object via OI (which
incurs LocalTime computation etc internally).
All these are done for "equals" comparison which can be done with "seconds &
nanoseconds" present in TimeStamp.
[https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/udf/ptf/ValueBoundaryScanner.java#L852]
Request is to explore optimising this code path, so that equals() can be
performed with "seconds/nanoseconds" instead of entire timestamp
{noformat}
at
org.apache.hadoop.hive.common.type.Timestamp.setTimeInSeconds(Timestamp.java:133)
at
org.apache.hadoop.hive.serde2.io.TimestampWritableV2.populateTimestamp(TimestampWritableV2.java:401)
at
org.apache.hadoop.hive.serde2.io.TimestampWritableV2.getTimestamp(TimestampWritableV2.java:210)
at
org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectInspectorUtils.getTimestamp(PrimitiveObjectInspectorUtils.java:1239)
at
org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectInspectorUtils.getTimestamp(PrimitiveObjectInspectorUtils.java:1181)
at
org.apache.hadoop.hive.ql.udf.ptf.TimestampValueBoundaryScanner.isEqual(ValueBoundaryScanner.java:848)
at
org.apache.hadoop.hive.ql.udf.ptf.SingleValueBoundaryScanner.computeEndCurrentRow(ValueBoundaryScanner.java:593)
at
org.apache.hadoop.hive.ql.udf.ptf.SingleValueBoundaryScanner.computeEnd(ValueBoundaryScanner.java:530)
at
org.apache.hadoop.hive.ql.udf.ptf.BasePartitionEvaluator.getRange(BasePartitionEvaluator.java:273)
at
org.apache.hadoop.hive.ql.udf.ptf.BasePartitionEvaluator.iterate(BasePartitionEvaluator.java:219)
at
org.apache.hadoop.hive.ql.udf.ptf.WindowingTableFunction.evaluateWindowFunction(WindowingTableFunction.java:147)
at
org.apache.hadoop.hive.ql.udf.ptf.WindowingTableFunction.access$100(WindowingTableFunction.java:61)
at
org.apache.hadoop.hive.ql.udf.ptf.WindowingTableFunction$WindowingIterator.next(WindowingTableFunction.java:755)
at
org.apache.hadoop.hive.ql.exec.PTFOperator$PTFInvocation.finishPartition(PTFOperator.java:373)
at
org.apache.hadoop.hive.ql.exec.PTFOperator.closeOp(PTFOperator.java:104)
at org.apache.hadoop.hive.ql.exec.Operator.close(Operator.java:732)
at org.apache.hadoop.hive.ql.exec.Operator.close(Operator.java:756)
at
org.apache.hadoop.hive.ql.exec.tez.ReduceRecordProcessor.close(ReduceRecordProcessor.java:383)
at
org.apache.hadoop.hive.ql.exec.tez.TezProcessor.initializeAndRunProcessor(TezProcessor.java:284)
at
org.apache.hadoop.hive.ql.exec.tez.TezProcessor.run(TezProcessor.java:250)
{noformat}
> PTF: TimestampValueBoundaryScanner can be optimised during range computation
> ----------------------------------------------------------------------------
>
> Key: HIVE-24746
> URL: https://issues.apache.org/jira/browse/HIVE-24746
> Project: Hive
> Issue Type: Improvement
> Reporter: László Bodor
> Assignee: László Bodor
> Priority: Major
>
> During range computation, timestamp ranges become a hotspot due to
> "TimeStamp" comparisons. It has to construct the entire TimeStamp object via
> OI (which incurs LocalTime computation etc internally).
>
> All these are done for "equals" comparison which can be done with "seconds &
> nanoseconds" present in TimeStamp.
>
> [https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/udf/ptf/ValueBoundaryScanner.java#L852]
>
>
> Request is to explore optimising this code path, so that equals() can be
> performed with "seconds/nanoseconds" instead of entire timestamp
>
> {noformat}
> at
> org.apache.hadoop.hive.common.type.Timestamp.setTimeInSeconds(Timestamp.java:133)
> at
> org.apache.hadoop.hive.serde2.io.TimestampWritableV2.populateTimestamp(TimestampWritableV2.java:401)
> at
> org.apache.hadoop.hive.serde2.io.TimestampWritableV2.getTimestamp(TimestampWritableV2.java:210)
> at
> org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectInspectorUtils.getTimestamp(PrimitiveObjectInspectorUtils.java:1239)
> at
> org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectInspectorUtils.getTimestamp(PrimitiveObjectInspectorUtils.java:1181)
> at
> org.apache.hadoop.hive.ql.udf.ptf.TimestampValueBoundaryScanner.isEqual(ValueBoundaryScanner.java:848)
> at
> org.apache.hadoop.hive.ql.udf.ptf.SingleValueBoundaryScanner.computeEndCurrentRow(ValueBoundaryScanner.java:593)
> at
> org.apache.hadoop.hive.ql.udf.ptf.SingleValueBoundaryScanner.computeEnd(ValueBoundaryScanner.java:530)
> at
> org.apache.hadoop.hive.ql.udf.ptf.BasePartitionEvaluator.getRange(BasePartitionEvaluator.java:273)
> at
> org.apache.hadoop.hive.ql.udf.ptf.BasePartitionEvaluator.iterate(BasePartitionEvaluator.java:219)
> at
> org.apache.hadoop.hive.ql.udf.ptf.WindowingTableFunction.evaluateWindowFunction(WindowingTableFunction.java:147)
> at
> org.apache.hadoop.hive.ql.udf.ptf.WindowingTableFunction.access$100(WindowingTableFunction.java:61)
> at
> org.apache.hadoop.hive.ql.udf.ptf.WindowingTableFunction$WindowingIterator.next(WindowingTableFunction.java:755)
> at
> org.apache.hadoop.hive.ql.exec.PTFOperator$PTFInvocation.finishPartition(PTFOperator.java:373)
> at
> org.apache.hadoop.hive.ql.exec.PTFOperator.closeOp(PTFOperator.java:104)
> at org.apache.hadoop.hive.ql.exec.Operator.close(Operator.java:732)
> at org.apache.hadoop.hive.ql.exec.Operator.close(Operator.java:756)
> at
> org.apache.hadoop.hive.ql.exec.tez.ReduceRecordProcessor.close(ReduceRecordProcessor.java:383)
> at
> org.apache.hadoop.hive.ql.exec.tez.TezProcessor.initializeAndRunProcessor(TezProcessor.java:284)
> at
> org.apache.hadoop.hive.ql.exec.tez.TezProcessor.run(TezProcessor.java:250)
> {noformat}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
