Github user xccui commented on a diff in the pull request:
https://github.com/apache/flink/pull/4625#discussion_r139983961
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/runtime/join/TimeBoundedStreamInnerJoin.scala
---
@@ -131,340 +116,308 @@ class TimeBoundedStreamInnerJoin(
// Initialize the data caches.
val leftListTypeInfo: TypeInformation[JList[Row]] = new
ListTypeInfo[Row](leftType)
val leftStateDescriptor: MapStateDescriptor[Long, JList[Row]] =
- new MapStateDescriptor[Long, JList[Row]](timeIndicator +
"InnerJoinLeftCache",
- BasicTypeInfo.LONG_TYPE_INFO.asInstanceOf[TypeInformation[Long]],
leftListTypeInfo)
+ new MapStateDescriptor[Long, JList[Row]](
+ timeIndicator + "InnerJoinLeftCache",
+ BasicTypeInfo.LONG_TYPE_INFO.asInstanceOf[TypeInformation[Long]],
+ leftListTypeInfo)
leftCache = getRuntimeContext.getMapState(leftStateDescriptor)
val rightListTypeInfo: TypeInformation[JList[Row]] = new
ListTypeInfo[Row](rightType)
val rightStateDescriptor: MapStateDescriptor[Long, JList[Row]] =
- new MapStateDescriptor[Long, JList[Row]](timeIndicator +
"InnerJoinRightCache",
- BasicTypeInfo.LONG_TYPE_INFO.asInstanceOf[TypeInformation[Long]],
rightListTypeInfo)
+ new MapStateDescriptor[Long, JList[Row]](
+ timeIndicator + "InnerJoinRightCache",
+ BasicTypeInfo.LONG_TYPE_INFO.asInstanceOf[TypeInformation[Long]],
+ rightListTypeInfo)
rightCache = getRuntimeContext.getMapState(rightStateDescriptor)
// Initialize the timer states.
val leftTimerStateDesc: ValueStateDescriptor[Long] =
- new ValueStateDescriptor[Long](timeIndicator +
"InnerJoinLeftTimerState",
- classOf[Long])
+ new ValueStateDescriptor[Long](timeIndicator +
"InnerJoinLeftTimerState", classOf[Long])
leftTimerState = getRuntimeContext.getState(leftTimerStateDesc)
val rightTimerStateDesc: ValueStateDescriptor[Long] =
- new ValueStateDescriptor[Long](timeIndicator +
"InnerJoinRightTimerState",
- classOf[Long])
+ new ValueStateDescriptor[Long](timeIndicator +
"InnerJoinRightTimerState", classOf[Long])
rightTimerState = getRuntimeContext.getState(rightTimerStateDesc)
}
/**
- * Process records from the left stream.
- *
- * @param cRowValue the input record
- * @param ctx the context to register timer or get current time
- * @param out the collector for outputting results
- *
+ * Process rows from the left stream.
*/
override def processElement1(
- cRowValue: CRow,
- ctx: CoProcessFunction[CRow, CRow, CRow]#Context,
- out: Collector[CRow]): Unit = {
- val timeForRecord: Long = getTimeForRecord(ctx, cRowValue, true)
- getCurrentOperatorTime(ctx)
+ cRowValue: CRow,
+ ctx: CoProcessFunction[CRow, CRow, CRow]#Context,
+ out: Collector[CRow]): Unit = {
+ updateOperatorTime(ctx)
+ val rowTime: Long = getTimeForLeftStream(ctx, cRowValue)
+ val oppositeLowerBound: Long = rowTime - rightRelativeSize
+ val oppositeUpperBound: Long = rowTime + leftRelativeSize
processElement(
cRowValue,
- timeForRecord,
+ rowTime,
ctx,
out,
leftOperatorTime,
+ oppositeLowerBound,
+ oppositeUpperBound,
rightOperatorTime,
rightTimerState,
leftCache,
rightCache,
- true
+ leftRow = true
)
}
/**
- * Process records from the right stream.
- *
- * @param cRowValue the input record
- * @param ctx the context to get current time
- * @param out the collector for outputting results
- *
+ * Process rows from the right stream.
*/
override def processElement2(
- cRowValue: CRow,
- ctx: CoProcessFunction[CRow, CRow, CRow]#Context,
- out: Collector[CRow]): Unit = {
- val timeForRecord: Long = getTimeForRecord(ctx, cRowValue, false)
- getCurrentOperatorTime(ctx)
+ cRowValue: CRow,
+ ctx: CoProcessFunction[CRow, CRow, CRow]#Context,
+ out: Collector[CRow]): Unit = {
+ updateOperatorTime(ctx)
+ val rowTime: Long = getTimeForRightStream(ctx, cRowValue)
+ val oppositeLowerBound: Long = rowTime - leftRelativeSize
+ val oppositeUpperBound: Long = rowTime + rightRelativeSize
processElement(
cRowValue,
- timeForRecord,
+ rowTime,
ctx,
out,
rightOperatorTime,
+ oppositeLowerBound,
+ oppositeUpperBound,
leftOperatorTime,
leftTimerState,
rightCache,
leftCache,
- false
+ leftRow = false
)
}
/**
- * Put a record from the input stream into the cache and iterate the
opposite cache to
- * output records meeting the join conditions. If there is no timer set
for the OPPOSITE
+ * Put a row from the input stream into the cache and iterate the
opposite cache to
+ * output join results meeting the conditions. If there is no timer set
for the OPPOSITE
* STREAM, register one.
*/
private def processElement(
- cRowValue: CRow,
- timeForRecord: Long,
- ctx: CoProcessFunction[CRow, CRow, CRow]#Context,
- out: Collector[CRow],
- myWatermark: Long,
- oppositeWatermark: Long,
- oppositeTimeState: ValueState[Long],
- recordListCache: MapState[Long, JList[Row]],
- oppositeCache: MapState[Long, JList[Row]],
- leftRecord: Boolean): Unit = {
- if (relativeWindowSize > 0) {
- //TODO Shall we consider adding a method for initialization with the
context and collector?
- cRowWrapper.out = out
-
- val record = cRowValue.row
-
- //TODO Only if the time of the record is greater than the watermark,
can we continue.
- if (timeForRecord >= myWatermark - allowedLateness) {
- val oppositeLowerBound: Long =
- if (leftRecord) timeForRecord - rightRelativeSize else
timeForRecord - leftRelativeSize
-
- val oppositeUpperBound: Long =
- if (leftRecord) timeForRecord + leftRelativeSize else
timeForRecord + rightRelativeSize
-
- // Put the record into the cache for later use.
- val recordList = if (recordListCache.contains(timeForRecord)) {
- recordListCache.get(timeForRecord)
- } else {
- new util.ArrayList[Row]()
- }
- recordList.add(record)
- recordListCache.put(timeForRecord, recordList)
-
- // Register a timer on THE OTHER STREAM to remove records from the
cache once they are
- // expired.
- if (oppositeTimeState.value == 0) {
- registerCleanUpTimer(
- ctx, timeForRecord, oppositeWatermark, oppositeTimeState,
leftRecord, true)
- }
+ cRowValue: CRow,
+ timeForRow: Long,
+ ctx: CoProcessFunction[CRow, CRow, CRow]#Context,
+ out: Collector[CRow],
+ myWatermark: Long,
+ oppositeLowerBound: Long,
+ oppositeUpperBound: Long,
+ oppositeWatermark: Long,
+ oppositeTimeState: ValueState[Long],
+ rowListCache: MapState[Long, JList[Row]],
+ oppositeCache: MapState[Long, JList[Row]],
+ leftRow: Boolean): Unit = {
+ cRowWrapper.out = out
+ val row = cRowValue.row
+ if (!checkRowOutOfDate(timeForRow, myWatermark)) {
--- End diff --
Hi @fhueske, it takes me a little time to rethink these conditions. As you
said, it's really quite easy to get confused ð
In general, I'm in favour of your suggestions. Besides, I've got some extra
ideas.
1. About storing a record in state. Only if `oppositeOperatorTime <
oppositeQualifedUpperBound`, can a record be stored. For example, suppose
there's a record from the left stream with `time = 10` and qualified time range
of the right stream is calculated to be `[8, 15]`. Then only if the
`rightOperatorTime < 15`, need we cache the left record. Otherwise, it means
all the qualified records from the right stream should have been seen.
2. We need two conditions here, but the "lateness condition" should be
applied out of the "storing condition", i.e., we first check the lateness of a
record and only if it passes can we continue processing it (checking if it
should be stored, etc.). That's because even if there exist records from the
opposite stream that can be joined with a late record, we can not guarantee the
**completeness** of the results. In the above example, if `rightOperatorTime =
16`, the left record with `time = 10` will not be stored. After that, here
comes a right record with `time = 15`, but we can never join it with the left
record just seen before since it is not stored.
"Incomplete" or "empty", which one to choose?
---
