-
Notifications
You must be signed in to change notification settings - Fork 28.2k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[SPARK-34892][SS] Introduce MergingSortWithSessionWindowStateIterator…
… sorting input rows and rows in state efficiently Introduction: this PR is a part of SPARK-10816 (EventTime based sessionization (session window)). Please refer #31937 to see the overall view of the code change. (Note that code diff could be diverged a bit.) ### What changes were proposed in this pull request? This PR introduces MergingSortWithSessionWindowStateIterator, which does "merge sort" between input rows and sessions in state based on group key and session's start time. Note that the iterator does merge sort among input rows and sessions grouped by grouping key. The iterator doesn't provide sessions in state which keys don't exist in input rows. For input rows, the iterator will provide all rows regardless of the existence of matching sessions in state. MergingSortWithSessionWindowStateIterator works on the precondition that given iterator is sorted by "group keys + start time of session window", and the iterator still retains the characteristic of the sort. ### Why are the changes needed? This part is a one of required on implementing SPARK-10816. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? New UT added. Closes #33077 from HeartSaVioR/SPARK-34892-SPARK-10816-PR-31570-part-4. Authored-by: Jungtaek Lim <kabhwan.opensource@gmail.com> Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com> (cherry picked from commit 12a576f) Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
- Loading branch information
1 parent
5bc06fd
commit dcee7a6
Showing
3 changed files
with
400 additions
and
1 deletion.
There are no files selected for viewing
168 changes: 168 additions & 0 deletions
168
.../org/apache/spark/sql/execution/streaming/MergingSortWithSessionWindowStateIterator.scala
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,168 @@ | ||
/* | ||
* Licensed to the Apache Software Foundation (ASF) under one or more | ||
* contributor license agreements. See the NOTICE file distributed with | ||
* this work for additional information regarding copyright ownership. | ||
* The ASF licenses this file to You under the Apache License, Version 2.0 | ||
* (the "License"); you may not use this file except in compliance with | ||
* the License. You may obtain a copy of the License at | ||
* | ||
* http://www.apache.org/licenses/LICENSE-2.0 | ||
* | ||
* Unless required by applicable law or agreed to in writing, software | ||
* distributed under the License is distributed on an "AS IS" BASIS, | ||
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
* See the License for the specific language governing permissions and | ||
* limitations under the License. | ||
*/ | ||
|
||
package org.apache.spark.sql.execution.streaming | ||
|
||
import org.apache.spark.internal.Logging | ||
import org.apache.spark.sql.catalyst.InternalRow | ||
import org.apache.spark.sql.catalyst.expressions.{Attribute, UnsafeProjection, UnsafeRow} | ||
import org.apache.spark.sql.catalyst.expressions.codegen.GenerateUnsafeProjection | ||
import org.apache.spark.sql.execution.streaming.state.{ReadStateStore, StreamingSessionWindowStateManager} | ||
|
||
/** | ||
* This class technically does the merge sort between input rows and existing sessions in state, | ||
* to optimize the cost of sort on "input rows + existing sessions". This is based on the | ||
* precondition that input rows are sorted by "group keys + start time of session window". | ||
* | ||
* This only materializes the existing sessions into memory, which are tend to be not many per | ||
* group key. The cost of sorting existing sessions would be also minor based on the assumption. | ||
* | ||
* The output rows are sorted with "group keys + start time of session window", which is same as | ||
* the sort condition on input rows. | ||
*/ | ||
class MergingSortWithSessionWindowStateIterator( | ||
iter: Iterator[InternalRow], | ||
stateManager: StreamingSessionWindowStateManager, | ||
store: ReadStateStore, | ||
groupWithoutSessionExpressions: Seq[Attribute], | ||
sessionExpression: Attribute, | ||
inputSchema: Seq[Attribute]) extends Iterator[InternalRow] with Logging { | ||
|
||
private val keysProjection: UnsafeProjection = GenerateUnsafeProjection.generate( | ||
groupWithoutSessionExpressions, inputSchema) | ||
private val sessionProjection: UnsafeProjection = | ||
GenerateUnsafeProjection.generate(Seq(sessionExpression), inputSchema) | ||
|
||
private case class SessionRowInformation( | ||
keys: UnsafeRow, | ||
sessionStart: Long, | ||
sessionEnd: Long, | ||
row: InternalRow) | ||
|
||
private object SessionRowInformation { | ||
def of(row: InternalRow): SessionRowInformation = { | ||
val keys = keysProjection(row).copy() | ||
val session = sessionProjection(row).copy() | ||
val sessionRow = session.getStruct(0, 2) | ||
val sessionStart = sessionRow.getLong(0) | ||
val sessionEnd = sessionRow.getLong(1) | ||
|
||
SessionRowInformation(keys, sessionStart, sessionEnd, row) | ||
} | ||
} | ||
|
||
// Holds the latest fetched row from input side iterator. | ||
private var currentRowFromInput: SessionRowInformation = _ | ||
|
||
// Holds the latest fetched row from state side iterator. | ||
private var currentRowFromState: SessionRowInformation = _ | ||
|
||
// Holds the iterator of rows (sessions) in state for the session key. | ||
private var sessionIterFromState: Iterator[InternalRow] = _ | ||
|
||
// Holds the current session key. | ||
private var currentSessionKey: UnsafeRow = _ | ||
|
||
override def hasNext: Boolean = { | ||
currentRowFromInput != null || currentRowFromState != null || | ||
(sessionIterFromState != null && sessionIterFromState.hasNext) || iter.hasNext | ||
} | ||
|
||
override def next(): InternalRow = { | ||
if (currentRowFromInput == null) { | ||
mayFillCurrentRow() | ||
} | ||
|
||
if (currentRowFromState == null) { | ||
mayFillCurrentStateRow() | ||
} | ||
|
||
if (currentRowFromInput == null && currentRowFromState == null) { | ||
throw new IllegalStateException("No Row to provide in next() which should not happen!") | ||
} | ||
|
||
// return current row vs current state row, should return smaller key, earlier session start | ||
val returnCurrentRow: Boolean = { | ||
if (currentRowFromInput == null) { | ||
false | ||
} else if (currentRowFromState == null) { | ||
true | ||
} else { | ||
// compare | ||
if (currentRowFromInput.keys != currentRowFromState.keys) { | ||
// state row cannot advance to row in input, so state row should be lower | ||
false | ||
} else { | ||
currentRowFromInput.sessionStart < currentRowFromState.sessionStart | ||
} | ||
} | ||
} | ||
|
||
val ret: SessionRowInformation = { | ||
if (returnCurrentRow) { | ||
val toRet = currentRowFromInput | ||
currentRowFromInput = null | ||
toRet | ||
} else { | ||
val toRet = currentRowFromState | ||
currentRowFromState = null | ||
toRet | ||
} | ||
} | ||
|
||
ret.row | ||
} | ||
|
||
private def mayFillCurrentRow(): Unit = { | ||
if (iter.hasNext) { | ||
currentRowFromInput = SessionRowInformation.of(iter.next()) | ||
} | ||
} | ||
|
||
private def mayFillCurrentStateRow(): Unit = { | ||
if (sessionIterFromState != null && sessionIterFromState.hasNext) { | ||
currentRowFromState = SessionRowInformation.of(sessionIterFromState.next()) | ||
} else { | ||
sessionIterFromState = null | ||
|
||
if (currentRowFromInput != null && currentRowFromInput.keys != currentSessionKey) { | ||
// We expect a small number of sessions per group key, so materializing them | ||
// and sorting wouldn't hurt much. The important thing is that we shouldn't buffer input | ||
// rows to sort with existing sessions. | ||
val unsortedIter = stateManager.getSessions(store, currentRowFromInput.keys) | ||
val unsortedList = unsortedIter.map(_.copy()).toList | ||
|
||
val sortedList = unsortedList.sortWith((row1, row2) => { | ||
def getSessionStart(r: InternalRow): Long = { | ||
val session = sessionProjection(r) | ||
val sessionRow = session.getStruct(0, 2) | ||
sessionRow.getLong(0) | ||
} | ||
|
||
// here sorting is based on the fact that keys are same | ||
getSessionStart(row1).compareTo(getSessionStart(row2)) < 0 | ||
}) | ||
sessionIterFromState = sortedList.iterator | ||
|
||
currentSessionKey = currentRowFromInput.keys | ||
if (sessionIterFromState.hasNext) { | ||
currentRowFromState = SessionRowInformation.of(sessionIterFromState.next()) | ||
} | ||
} | ||
} | ||
} | ||
} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.