Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
1 change: 1 addition & 0 deletions sdk/cosmos/azure-cosmos-spark_3-3_2-12/CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
#### Breaking Changes

#### Bugs Fixed
* Fixed bounded change feed reads hanging after partition splits when stale feed-range metadata collapsed child continuation LSNs into a single `endLsn`. - See [PR 49883](https://github.com/Azure/azure-sdk-for-java/pull/49883)

#### Other Changes

Expand Down
1 change: 1 addition & 0 deletions sdk/cosmos/azure-cosmos-spark_3-4_2-12/CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
#### Breaking Changes

#### Bugs Fixed
* Fixed bounded change feed reads hanging after partition splits when stale feed-range metadata collapsed child continuation LSNs into a single `endLsn`. - See [PR 49883](https://github.com/Azure/azure-sdk-for-java/pull/49883)

#### Other Changes

Expand Down
1 change: 1 addition & 0 deletions sdk/cosmos/azure-cosmos-spark_3-5_2-12/CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
#### Breaking Changes

#### Bugs Fixed
* Fixed bounded change feed reads hanging after partition splits when stale feed-range metadata collapsed child continuation LSNs into a single `endLsn`. - See [PR 49883](https://github.com/Azure/azure-sdk-for-java/pull/49883)

#### Other Changes

Expand Down
1 change: 1 addition & 0 deletions sdk/cosmos/azure-cosmos-spark_3-5_2-13/CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
#### Breaking Changes

#### Bugs Fixed
* Fixed bounded change feed reads hanging after partition splits when stale feed-range metadata collapsed child continuation LSNs into a single `endLsn`. - See [PR 49883](https://github.com/Azure/azure-sdk-for-java/pull/49883)

#### Other Changes

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -139,8 +139,14 @@ private[cosmos] object SparkBridgeImplementationInternal extends BasicLoggingTra

def extractContinuationTokensFromChangeFeedStateJson(stateJsonBase64: String): Array[(NormalizedRange, Long)] = {
assert(!Strings.isNullOrWhiteSpace(stateJsonBase64), s"Argument 'stateJsonBase64' must not be null or empty.")
val state = ChangeFeedState.fromString(stateJsonBase64)
state
extractContinuationTokensFromChangeFeedState(ChangeFeedState.fromString(stateJsonBase64))
}

def extractContinuationTokensFromChangeFeedState(
changeFeedState: ChangeFeedState
): Array[(NormalizedRange, Long)] = {
assert(changeFeedState != null, "Argument 'changeFeedState' must not be null.")
changeFeedState
.extractContinuationTokens() // already sorted
.asScala
.map(t => Tuple2(rangeToNormalizedRange(t.getRange), toLsn(t.getToken)))
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -414,17 +414,25 @@ private object CosmosPartitionPlanner extends BasicLoggingTrait {
logDebug(endOffsetDebug.toString)
}

createLatestOffset(startOffset, inputPartitions)
}
// scalastyle:on method.length
// scalastyle:on parameter.number

private[spark] def createLatestOffset
(
startOffset: ChangeFeedOffset,
inputPartitions: Array[CosmosInputPartition]
): ChangeFeedOffset = {
val changeFeedStateJson = SparkBridgeImplementationInternal
.createChangeFeedStateJson(
startOffset.changeFeedState,
inputPartitions.map(m => (m.feedRange, m.endLsn.get)))

ChangeFeedOffset(changeFeedStateJson, Some(inputPartitions))
}
// scalastyle:on method.length
// scalastyle:on parameter.number

private[this] def getOrderedPartitionMetadataWithStartLsn
private[spark] def getOrderedPartitionMetadataWithStartLsn
(
stateJson: String,
latestPartitionMetadata: Array[PartitionMetadata]
Expand Down Expand Up @@ -834,12 +842,48 @@ private object CosmosPartitionPlanner extends BasicLoggingTrait {
0,
None,
new AtomicLong(0),
new AtomicLong(0)
new AtomicLong(0),
None
))
})
.collectSeq()
.map(metadata =>
filterPartitionMetadataByFeedRanges(
expandPartitionMetadataByLatestLsn(metadata, isChangeFeed),
partitionConfig.feedRangeFiler))
})
.block()
.toArray
}

private[spark] def expandPartitionMetadataByLatestLsn(
metadata: Seq[PartitionMetadata],
isChangeFeed: Boolean
): Seq[PartitionMetadata] = {
if (isChangeFeed) {
metadata.foreach(partitionMetadata => {
if (partitionMetadata.fromNowContinuationState.isEmpty && partitionMetadata.latestLsn > 0) {
logWarning(
s"Change-feed metadata for range '${partitionMetadata.feedRange}' is missing the FromNow continuation " +
"state; composite split expansion cannot be performed.")
}
})
metadata.flatMap(_.splitByLatestLsn())
Comment thread
tvaron3 marked this conversation as resolved.
} else {
metadata
}
}

private[spark] def filterPartitionMetadataByFeedRanges(
metadata: Seq[PartitionMetadata],
feedRangeFilter: Option[Array[NormalizedRange]]
): Seq[PartitionMetadata] = {
feedRangeFilter match {
case Some(epkRangesInScope) =>
metadata.filter(partitionMetadata =>
epkRangesInScope.exists(epk =>
SparkBridgeImplementationInternal.doRangesOverlap(epk, partitionMetadata.feedRange)))
case None => metadata
}
}
}
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,36 @@ private object PartitionMetadata {
s"$databaseId/$containerId/${feedRange.min}-${feedRange.max}"

// scalastyle:off parameter.number
def apply(userConfig: Map[String, String],
cosmosClientConfig: CosmosClientConfiguration,
cosmosClientStateHandles: Option[Broadcast[CosmosClientMetadataCachesSnapshots]],
cosmosContainerConfig: CosmosContainerConfig,
feedRange: NormalizedRange,
documentCount: Long,
totalDocumentSizeInKB: Long,
firstLsn: Option[Long],
latestLsn: Long,
startLsn: Long,
endLsn: Option[Long],
lastRetrieved: AtomicLong,
lastUpdated: AtomicLong): PartitionMetadata = {
new PartitionMetadata(
userConfig,
cosmosClientConfig,
cosmosClientStateHandles,
cosmosContainerConfig,
feedRange,
documentCount,
totalDocumentSizeInKB,
firstLsn,
latestLsn,
startLsn,
endLsn,
lastRetrieved,
lastUpdated,
None)
}

def apply(userConfig: Map[String, String],
cosmosClientConfig: CosmosClientConfiguration,
cosmosClientStateHandles: Option[Broadcast[CosmosClientMetadataCachesSnapshots]],
Expand Down Expand Up @@ -51,7 +81,8 @@ private object PartitionMetadata {
startLsn,
endLsn,
new AtomicLong(nowEpochMs),
new AtomicLong(nowEpochMs)
new AtomicLong(nowEpochMs),
Some(fromNowContinuationToken)
)
}
}
Expand All @@ -70,7 +101,8 @@ private[cosmos] case class PartitionMetadata
startLsn: Long,
endLsn: Option[Long],
lastRetrieved: AtomicLong,
lastUpdated: AtomicLong
lastUpdated: AtomicLong,
fromNowContinuationState: Option[String]
) extends BasicLoggingTrait {

requireNotNull(feedRange, "feedRange")
Expand All @@ -94,7 +126,8 @@ private[cosmos] case class PartitionMetadata
startLsn,
this.endLsn,
new AtomicLong(this.lastRetrieved.get),
new AtomicLong(this.lastUpdated.get)
new AtomicLong(this.lastUpdated.get),
this.getFromNowContinuationStateForRange(subRange)
)
}

Expand All @@ -112,10 +145,55 @@ private[cosmos] case class PartitionMetadata
startLsn,
Some(explicitEndLsn),
new AtomicLong(this.lastRetrieved.get),
new AtomicLong(this.lastUpdated.get)
new AtomicLong(this.lastUpdated.get),
this.fromNowContinuationState
)
}

def splitByLatestLsn(): Seq[PartitionMetadata] = {
this.fromNowContinuationState match {
case None => Seq(this)
case Some(stateJson) =>
val parsedState = SparkBridgeImplementationInternal.parseChangeFeedState(stateJson)
val latestLsnsByRange =
SparkBridgeImplementationInternal.extractContinuationTokensFromChangeFeedState(parsedState)

if (latestLsnsByRange.length <= 1) {
Seq(this)
} else {
logInfo(
s"FromNow continuation for range '${this.feedRange}' resolved to " +
s"${latestLsnsByRange.length} effective ranges. Planning each range independently.")

val effectiveRangesAndLsns = latestLsnsByRange
Comment thread
tvaron3 marked this conversation as resolved.
.flatMap(rangeAndLsn =>
this.intersect(this.feedRange, rangeAndLsn._1)
.map(effectiveRange => effectiveRange -> rangeAndLsn._2))
.toSeq
if (effectiveRangesAndLsns.length != latestLsnsByRange.length) {
throw new IllegalStateException(
s"FromNow continuation for range '${this.feedRange}' contained non-overlapping effective ranges.")
}
val effectiveStates = SparkBridgeImplementationInternal.extractChangeFeedStateForRanges(
parsedState,
effectiveRangesAndLsns.map(_._1))

require(
effectiveStates.length == effectiveRangesAndLsns.length,
"Expected one continuation state for every effective range.")

effectiveRangesAndLsns
.zip(effectiveStates)
.map { case ((effectiveRange, effectiveLatestLsn), effectiveState) =>
this.withFeedRangeAndLatestLsn(
effectiveRange,
effectiveLatestLsn,
effectiveState)
}
}
}
}

def getWeightedLsnGap(
isChangeFeed: Boolean,
partitionMetricsMap: Option[ConcurrentHashMap[NormalizedRange, ChangeFeedMetricsTracker]] = None): Long = {
Expand Down Expand Up @@ -180,4 +258,50 @@ private[cosmos] case class PartitionMetadata
}
}
}

private def withFeedRangeAndLatestLsn(
effectiveRange: NormalizedRange,
effectiveLatestLsn: Long,
effectiveFromNowContinuationState: String
): PartitionMetadata = {
new PartitionMetadata(
this.userConfig,
this.cosmosClientConfig,
this.cosmosClientStateHandles,
this.cosmosContainerConfig,
effectiveRange,
0,
0,
None,
effectiveLatestLsn,
this.startLsn,
this.endLsn,
new AtomicLong(this.lastRetrieved.get),
new AtomicLong(this.lastUpdated.get),
Some(effectiveFromNowContinuationState)
)
}

private def getFromNowContinuationStateForRange(
effectiveRange: NormalizedRange
): Option[String] = {
this.fromNowContinuationState.map(state => {
if (effectiveRange == this.feedRange) {
state
} else {
SparkBridgeImplementationInternal.extractChangeFeedStateForRange(
SparkBridgeImplementationInternal.parseChangeFeedState(state),
effectiveRange)
}
})
}

private def intersect(
left: NormalizedRange,
right: NormalizedRange
): Option[NormalizedRange] = {
val min = if (left.min.compareTo(right.min) >= 0) left.min else right.min
val max = if (left.max.compareTo(right.max) <= 0) left.max else right.max
if (min.compareTo(max) < 0) Some(NormalizedRange(min, max)) else None
}
}
Loading
Loading