showuon commented on code in PR #13487:
URL: https://github.com/apache/kafka/pull/13487#discussion_r1157196981
##########
core/src/main/scala/kafka/log/remote/RemoteLogManager.scala:
##########
@@ -272,15 +364,270 @@ class RemoteLogManager(rlmConfig: RemoteLogManagerConfig,
None
}
+ trait CancellableRunnable extends Runnable {
+ @volatile private var cancelled = false
+
+ def cancel(): Unit = {
+ cancelled = true
+ }
+
+ def isCancelled(): Boolean = {
+ cancelled
+ }
+ }
+
+ /**
+ * Returns the leader epoch checkpoint by truncating with the given
start[exclusive] and end[inclusive] offset
+ *
+ * @param log The actual log from where to take the leader-epoch
checkpoint
+ * @param startOffset The start offset of the checkpoint file (exclusive in
the truncation).
+ * If start offset is 6, then it will retain an entry at
offset 6.
+ * @param endOffset The end offset of the checkpoint file (inclusive in
the truncation)
+ * If end offset is 100, then it will remove the entries
greater than or equal to 100.
+ * @return the truncated leader epoch checkpoint
+ */
+ private[remote] def getLeaderEpochCheckpoint(log: UnifiedLog, startOffset:
Long, endOffset: Long): InMemoryLeaderEpochCheckpoint = {
+ val checkpoint = new InMemoryLeaderEpochCheckpoint()
+ log.leaderEpochCache
+ .map(cache => cache.writeTo(checkpoint))
+ .foreach { x =>
+ if (startOffset >= 0) {
+ x.truncateFromStart(startOffset)
+ }
+ x.truncateFromEnd(endOffset)
+ }
+ checkpoint
+ }
+
+ private[remote] class RLMTask(tpId: TopicIdPartition) extends
CancellableRunnable with Logging {
+ this.logIdent = s"[RemoteLogManager=$brokerId partition=$tpId] "
+ @volatile private var leaderEpoch: Int = -1
+
+ private def isLeader(): Boolean = leaderEpoch >= 0
+
+ // The readOffset is None initially for a new leader RLMTask, and needs to
be fetched inside the task's run() method.
+ @volatile private var copiedOffsetOption: Option[Long] = None
+
+ def convertToLeader(leaderEpochVal: Int): Unit = {
+ if (leaderEpochVal < 0) {
+ throw new KafkaException(s"leaderEpoch value for topic partition $tpId
can not be negative")
+ }
+ if (this.leaderEpoch != leaderEpochVal) {
+ leaderEpoch = leaderEpochVal
+ }
+ // Reset readOffset, so that it is set in next run of RLMTask
+ copiedOffsetOption = None
+ }
+
+ def convertToFollower(): Unit = {
+ leaderEpoch = -1
+ }
+
+ def handleCopyLogSegmentsToRemote(): Unit = {
+ if (isCancelled())
+ return
+
+ def maybeUpdateReadOffset(): Unit = {
+ if (copiedOffsetOption.isEmpty) {
+ info(s"Find the highest remote offset for partition: $tpId after
becoming leader, leaderEpoch: $leaderEpoch")
+
+ // This is found by traversing from the latest leader epoch from
leader epoch history and find the highest offset
+ // of a segment with that epoch copied into remote storage. If it
can not find an entry then it checks for the
+ // previous leader epoch till it finds an entry, If there are no
entries till the earliest leader epoch in leader
+ // epoch cache then it starts copying the segments from the earliest
epoch entry’s offset.
+ copiedOffsetOption = Some(findHighestRemoteOffset(tpId))
+ }
+ }
+
+ try {
+ maybeUpdateReadOffset()
+ val copiedOffset = copiedOffsetOption.get
+ fetchLog(tpId.topicPartition()).foreach { log =>
+ // LSO indicates the offset below are ready to be
consumed(high-watermark or committed)
+ val lso = log.lastStableOffset
+ if (lso < 0) {
+ warn(s"lastStableOffset for partition $tpId is $lso, which should
not be negative.")
+ } else if (lso > 0 && copiedOffset < lso) {
+ // Copy segments only till the min of high-watermark or
stable-offset as remote storage should contain
+ // only committed/acked messages
+ val toOffset = lso
+ debug(s"Checking for segments to copy, copiedOffset: $copiedOffset
and toOffset: $toOffset")
+ val activeSegBaseOffset = log.activeSegment.baseOffset
+ // log-start-offset can be ahead of the read-offset, when:
+ // 1) log-start-offset gets incremented via delete-records API (or)
+ // 2) enabling the remote log for the first time
+ val fromOffset = Math.max(copiedOffset + 1, log.logStartOffset)
+ val sortedSegments = log.logSegments(fromOffset,
toOffset).toSeq.sortBy(_.baseOffset)
+ val activeSegIndex: Int = sortedSegments.map(x =>
x.baseOffset).search(activeSegBaseOffset) match {
+ case Found(x) => x
+ case InsertionPoint(y) => y - 1
+ }
+ // sortedSegments becomes empty list when fromOffset and toOffset
are same, and activeSegIndex becomes -1
+ if (activeSegIndex < 0) {
+ debug(s"No segments found to be copied for partition $tpId with
copiedOffset: $copiedOffset and " +
+ s"active segment's base-offset: $activeSegBaseOffset")
+ } else {
+ sortedSegments.slice(0, activeSegIndex).foreach { segment =>
+ if (isCancelled() || !isLeader()) {
+ info(s"Skipping copying log segments as the current task
state is changed, cancelled: " +
+ s"${isCancelled()} leader:${isLeader()}")
+ return
+ }
+
+ val logFile = segment.log.file()
+ val logFileName = logFile.getName
+
+ info(s"Copying $logFileName to remote storage.")
+ val id = new RemoteLogSegmentId(tpId, Uuid.randomUuid())
+
+ val nextOffset = segment.readNextOffset
+ val endOffset = nextOffset - 1
+ val producerIdSnapshotFile: File =
log.producerStateManager.fetchSnapshot(nextOffset).orElse(null)
+
+ val epochEntries = getLeaderEpochCheckpoint(log,
segment.baseOffset, nextOffset).read()
+ val segmentLeaderEpochs: util.Map[Integer, lang.Long] = new
util.HashMap[Integer, lang.Long](epochEntries.size())
+ epochEntries.forEach(entry =>
+ segmentLeaderEpochs.put(Integer.valueOf(entry.epoch),
lang.Long.valueOf(entry.startOffset))
+ )
+
+ val remoteLogSegmentMetadata = new
RemoteLogSegmentMetadata(id, segment.baseOffset, endOffset,
+ segment.largestTimestamp, brokerId, time.milliseconds(),
segment.log.sizeInBytes(),
+ segmentLeaderEpochs)
+
+
remoteLogMetadataManager.addRemoteLogSegmentMetadata(remoteLogSegmentMetadata).get()
+
+ val leaderEpochsIndex = getLeaderEpochCheckpoint(log,
startOffset = -1, nextOffset).readAsByteBuffer()
+ val segmentData = new LogSegmentData(logFile.toPath,
toPathIfExists(segment.lazyOffsetIndex.get.file),
+ toPathIfExists(segment.lazyTimeIndex.get.file),
Optional.ofNullable(toPathIfExists(segment.txnIndex.file)),
+ producerIdSnapshotFile.toPath, leaderEpochsIndex)
+
remoteLogStorageManager.copyLogSegmentData(remoteLogSegmentMetadata,
segmentData)
+
+ val rlsmAfterSegmentCopy = new
RemoteLogSegmentMetadataUpdate(id, time.milliseconds(),
+ RemoteLogSegmentState.COPY_SEGMENT_FINISHED, brokerId)
+
+
remoteLogMetadataManager.updateRemoteLogSegmentMetadata(rlsmAfterSegmentCopy).get()
+
+ copiedOffsetOption = Some(endOffset)
+ log.updateHighestOffsetInRemoteStorage(endOffset)
+ info(s"Copied $logFileName to remote storage with segment-id:
${rlsmAfterSegmentCopy.remoteLogSegmentId()}")
+ }
+ }
+ } else {
+ debug(s"Skipping copying segments, current
read-offset:$copiedOffset, and LSO:$lso")
+ }
+ }
+ } catch {
+ case ex: Exception =>
+ if (!isCancelled()) {
+ error(s"Error occurred while copying log segments of partition:
$tpId", ex)
+ }
+ }
+ }
+
+ private def toPathIfExists(file: File): Path = {
+ if (file.exists()) file.toPath else null
+ }
+
+ override def run(): Unit = {
+ if (isCancelled())
+ return
+
+ try {
+ if (isLeader()) {
+ // Copy log segments to remote storage
+ handleCopyLogSegmentsToRemote()
+ }
+ } catch {
+ case ex: InterruptedException =>
+ if (!isCancelled()) {
+ warn(s"Current thread for topic-partition-id $tpId is interrupted,
this task won't be rescheduled. " +
+ s"Reason: ${ex.getMessage}")
+ }
+ case ex: Exception =>
+ if (!isCancelled()) {
+ warn(s"Current task for topic-partition $tpId received error but
it will be scheduled. " +
+ s"Reason: ${ex.getMessage}")
+ }
+ }
+ }
+
+ override def toString: String = {
+ this.getClass.toString + s"[$tpId]"
+ }
+ }
+
+ def findHighestRemoteOffset(topicIdPartition: TopicIdPartition): Long = {
+ var offset: Optional[lang.Long] = Optional.empty()
+ fetchLog(topicIdPartition.topicPartition()).foreach { log =>
+ log.leaderEpochCache.foreach(cache => {
+ var epoch = cache.latestEpoch
+ while (!offset.isPresent && epoch.isPresent) {
+ offset =
remoteLogMetadataManager.highestOffsetForEpoch(topicIdPartition, epoch.getAsInt)
+ epoch = cache.previousEpoch(epoch.getAsInt)
+ }
+ })
+ }
+ offset.orElse(-1L)
+ }
+
+
+ private def doHandleLeaderOrFollowerPartitions(topicPartition:
TopicIdPartition,
+ convertToLeaderOrFollower:
RLMTask => Unit): Unit = {
+ val rlmTaskWithFuture =
leaderOrFollowerTasks.computeIfAbsent(topicPartition, (tp: TopicIdPartition) =>
{
+ val task = new RLMTask(tp)
+ // set this upfront when it is getting initialized instead of doing it
after scheduling.
+ convertToLeaderOrFollower(task)
+ info(s"Created a new task: $task and getting scheduled")
+ val future = rlmScheduledThreadPool.scheduleWithFixedDelay(task, 0,
delayInMs, TimeUnit.MILLISECONDS)
+ RLMTaskWithFuture(task, future)
+ })
+ convertToLeaderOrFollower(rlmTaskWithFuture.rlmTask)
Review Comment:
Will there be race condition when the topic partition is new inserted, so
it'll set `convertToLeaderOrFollower` once before scheduling. And then the
scheduled thread could be run immediately, later, run this line
`convertToLeaderOrFollower(rlmTaskWithFuture.rlmTask)`, so it reset the leader
epoch if it got updated.
Could we add a null condition check for `rlmTaskWithFuture` here? that is:
```
if (rlmTaskWithFuture != null) {
// not new created task, should convert it here
convertToLeaderOrFollower(rlmTaskWithFuture.rlmTask)
}
```
##########
core/src/main/scala/kafka/log/remote/RemoteLogManager.scala:
##########
@@ -272,15 +364,270 @@ class RemoteLogManager(rlmConfig: RemoteLogManagerConfig,
None
}
+ trait CancellableRunnable extends Runnable {
+ @volatile private var cancelled = false
+
+ def cancel(): Unit = {
+ cancelled = true
+ }
+
+ def isCancelled(): Boolean = {
+ cancelled
+ }
+ }
+
+ /**
+ * Returns the leader epoch checkpoint by truncating with the given
start[exclusive] and end[inclusive] offset
+ *
+ * @param log The actual log from where to take the leader-epoch
checkpoint
+ * @param startOffset The start offset of the checkpoint file (exclusive in
the truncation).
+ * If start offset is 6, then it will retain an entry at
offset 6.
+ * @param endOffset The end offset of the checkpoint file (inclusive in
the truncation)
+ * If end offset is 100, then it will remove the entries
greater than or equal to 100.
+ * @return the truncated leader epoch checkpoint
+ */
+ private[remote] def getLeaderEpochCheckpoint(log: UnifiedLog, startOffset:
Long, endOffset: Long): InMemoryLeaderEpochCheckpoint = {
+ val checkpoint = new InMemoryLeaderEpochCheckpoint()
+ log.leaderEpochCache
+ .map(cache => cache.writeTo(checkpoint))
+ .foreach { x =>
+ if (startOffset >= 0) {
+ x.truncateFromStart(startOffset)
+ }
+ x.truncateFromEnd(endOffset)
+ }
+ checkpoint
+ }
+
+ private[remote] class RLMTask(tpId: TopicIdPartition) extends
CancellableRunnable with Logging {
+ this.logIdent = s"[RemoteLogManager=$brokerId partition=$tpId] "
+ @volatile private var leaderEpoch: Int = -1
+
+ private def isLeader(): Boolean = leaderEpoch >= 0
+
+ // The readOffset is None initially for a new leader RLMTask, and needs to
be fetched inside the task's run() method.
+ @volatile private var copiedOffsetOption: Option[Long] = None
+
+ def convertToLeader(leaderEpochVal: Int): Unit = {
+ if (leaderEpochVal < 0) {
+ throw new KafkaException(s"leaderEpoch value for topic partition $tpId
can not be negative")
+ }
+ if (this.leaderEpoch != leaderEpochVal) {
+ leaderEpoch = leaderEpochVal
+ }
+ // Reset readOffset, so that it is set in next run of RLMTask
+ copiedOffsetOption = None
+ }
+
+ def convertToFollower(): Unit = {
+ leaderEpoch = -1
+ }
+
+ def handleCopyLogSegmentsToRemote(): Unit = {
+ if (isCancelled())
+ return
+
+ def maybeUpdateReadOffset(): Unit = {
+ if (copiedOffsetOption.isEmpty) {
+ info(s"Find the highest remote offset for partition: $tpId after
becoming leader, leaderEpoch: $leaderEpoch")
+
+ // This is found by traversing from the latest leader epoch from
leader epoch history and find the highest offset
+ // of a segment with that epoch copied into remote storage. If it
can not find an entry then it checks for the
+ // previous leader epoch till it finds an entry, If there are no
entries till the earliest leader epoch in leader
+ // epoch cache then it starts copying the segments from the earliest
epoch entry’s offset.
+ copiedOffsetOption = Some(findHighestRemoteOffset(tpId))
+ }
+ }
+
+ try {
+ maybeUpdateReadOffset()
+ val copiedOffset = copiedOffsetOption.get
+ fetchLog(tpId.topicPartition()).foreach { log =>
+ // LSO indicates the offset below are ready to be
consumed(high-watermark or committed)
+ val lso = log.lastStableOffset
+ if (lso < 0) {
+ warn(s"lastStableOffset for partition $tpId is $lso, which should
not be negative.")
+ } else if (lso > 0 && copiedOffset < lso) {
+ // Copy segments only till the min of high-watermark or
stable-offset as remote storage should contain
+ // only committed/acked messages
+ val toOffset = lso
+ debug(s"Checking for segments to copy, copiedOffset: $copiedOffset
and toOffset: $toOffset")
+ val activeSegBaseOffset = log.activeSegment.baseOffset
+ // log-start-offset can be ahead of the read-offset, when:
+ // 1) log-start-offset gets incremented via delete-records API (or)
+ // 2) enabling the remote log for the first time
+ val fromOffset = Math.max(copiedOffset + 1, log.logStartOffset)
+ val sortedSegments = log.logSegments(fromOffset,
toOffset).toSeq.sortBy(_.baseOffset)
+ val activeSegIndex: Int = sortedSegments.map(x =>
x.baseOffset).search(activeSegBaseOffset) match {
+ case Found(x) => x
+ case InsertionPoint(y) => y - 1
+ }
+ // sortedSegments becomes empty list when fromOffset and toOffset
are same, and activeSegIndex becomes -1
+ if (activeSegIndex < 0) {
+ debug(s"No segments found to be copied for partition $tpId with
copiedOffset: $copiedOffset and " +
+ s"active segment's base-offset: $activeSegBaseOffset")
+ } else {
+ sortedSegments.slice(0, activeSegIndex).foreach { segment =>
+ if (isCancelled() || !isLeader()) {
+ info(s"Skipping copying log segments as the current task
state is changed, cancelled: " +
+ s"${isCancelled()} leader:${isLeader()}")
+ return
+ }
+
+ val logFile = segment.log.file()
+ val logFileName = logFile.getName
+
+ info(s"Copying $logFileName to remote storage.")
+ val id = new RemoteLogSegmentId(tpId, Uuid.randomUuid())
+
+ val nextOffset = segment.readNextOffset
+ val endOffset = nextOffset - 1
+ val producerIdSnapshotFile: File =
log.producerStateManager.fetchSnapshot(nextOffset).orElse(null)
+
+ val epochEntries = getLeaderEpochCheckpoint(log,
segment.baseOffset, nextOffset).read()
+ val segmentLeaderEpochs: util.Map[Integer, lang.Long] = new
util.HashMap[Integer, lang.Long](epochEntries.size())
+ epochEntries.forEach(entry =>
+ segmentLeaderEpochs.put(Integer.valueOf(entry.epoch),
lang.Long.valueOf(entry.startOffset))
+ )
+
+ val remoteLogSegmentMetadata = new
RemoteLogSegmentMetadata(id, segment.baseOffset, endOffset,
+ segment.largestTimestamp, brokerId, time.milliseconds(),
segment.log.sizeInBytes(),
+ segmentLeaderEpochs)
+
+
remoteLogMetadataManager.addRemoteLogSegmentMetadata(remoteLogSegmentMetadata).get()
+
+ val leaderEpochsIndex = getLeaderEpochCheckpoint(log,
startOffset = -1, nextOffset).readAsByteBuffer()
+ val segmentData = new LogSegmentData(logFile.toPath,
toPathIfExists(segment.lazyOffsetIndex.get.file),
+ toPathIfExists(segment.lazyTimeIndex.get.file),
Optional.ofNullable(toPathIfExists(segment.txnIndex.file)),
+ producerIdSnapshotFile.toPath, leaderEpochsIndex)
+
remoteLogStorageManager.copyLogSegmentData(remoteLogSegmentMetadata,
segmentData)
+
+ val rlsmAfterSegmentCopy = new
RemoteLogSegmentMetadataUpdate(id, time.milliseconds(),
+ RemoteLogSegmentState.COPY_SEGMENT_FINISHED, brokerId)
+
+
remoteLogMetadataManager.updateRemoteLogSegmentMetadata(rlsmAfterSegmentCopy).get()
+
+ copiedOffsetOption = Some(endOffset)
+ log.updateHighestOffsetInRemoteStorage(endOffset)
+ info(s"Copied $logFileName to remote storage with segment-id:
${rlsmAfterSegmentCopy.remoteLogSegmentId()}")
+ }
+ }
+ } else {
+ debug(s"Skipping copying segments, current
read-offset:$copiedOffset, and LSO:$lso")
+ }
+ }
+ } catch {
+ case ex: Exception =>
+ if (!isCancelled()) {
+ error(s"Error occurred while copying log segments of partition:
$tpId", ex)
+ }
+ }
+ }
+
+ private def toPathIfExists(file: File): Path = {
+ if (file.exists()) file.toPath else null
+ }
+
+ override def run(): Unit = {
+ if (isCancelled())
+ return
+
+ try {
+ if (isLeader()) {
+ // Copy log segments to remote storage
+ handleCopyLogSegmentsToRemote()
+ }
+ } catch {
+ case ex: InterruptedException =>
+ if (!isCancelled()) {
+ warn(s"Current thread for topic-partition-id $tpId is interrupted,
this task won't be rescheduled. " +
+ s"Reason: ${ex.getMessage}")
+ }
+ case ex: Exception =>
+ if (!isCancelled()) {
+ warn(s"Current task for topic-partition $tpId received error but
it will be scheduled. " +
+ s"Reason: ${ex.getMessage}")
+ }
+ }
+ }
+
+ override def toString: String = {
+ this.getClass.toString + s"[$tpId]"
+ }
+ }
+
+ def findHighestRemoteOffset(topicIdPartition: TopicIdPartition): Long = {
+ var offset: Optional[lang.Long] = Optional.empty()
+ fetchLog(topicIdPartition.topicPartition()).foreach { log =>
+ log.leaderEpochCache.foreach(cache => {
+ var epoch = cache.latestEpoch
+ while (!offset.isPresent && epoch.isPresent) {
+ offset =
remoteLogMetadataManager.highestOffsetForEpoch(topicIdPartition, epoch.getAsInt)
+ epoch = cache.previousEpoch(epoch.getAsInt)
+ }
+ })
+ }
+ offset.orElse(-1L)
+ }
+
+
+ private def doHandleLeaderOrFollowerPartitions(topicPartition:
TopicIdPartition,
+ convertToLeaderOrFollower:
RLMTask => Unit): Unit = {
+ val rlmTaskWithFuture =
leaderOrFollowerTasks.computeIfAbsent(topicPartition, (tp: TopicIdPartition) =>
{
+ val task = new RLMTask(tp)
+ // set this upfront when it is getting initialized instead of doing it
after scheduling.
+ convertToLeaderOrFollower(task)
+ info(s"Created a new task: $task and getting scheduled")
+ val future = rlmScheduledThreadPool.scheduleWithFixedDelay(task, 0,
delayInMs, TimeUnit.MILLISECONDS)
+ RLMTaskWithFuture(task, future)
+ })
+ convertToLeaderOrFollower(rlmTaskWithFuture.rlmTask)
Review Comment:
Will there be race condition when the topic partition is new inserted, so
it'll set `convertToLeaderOrFollower` once before scheduling. And then the
scheduled thread could be run immediately, later, run this line
`convertToLeaderOrFollower(rlmTaskWithFuture.rlmTask)`, so it reset the leader
epoch if it got updated.
Could we add a null condition check for `rlmTaskWithFuture` here? that is:
```
if (rlmTaskWithFuture != null) {
// not new created task, should convert it here
convertToLeaderOrFollower(rlmTaskWithFuture.rlmTask)
}
```
WDYT?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: jira-unsubscr...@kafka.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
