Github user zsxwing commented on a diff in the pull request:
https://github.com/apache/spark/pull/9143#discussion_r42720937
--- Diff:
streaming/src/main/scala/org/apache/spark/streaming/scheduler/ReceivedBlockTracker.scala
---
@@ -235,8 +350,67 @@ private[streaming] class ReceivedBlockTracker(
}
}
+ /**
+ * Creates a WAL Writer in a separate thread to enable batching of log
events.
+ * Exposed for tests.
+ */
+ protected def createBatchWriteAheadLogWriter(): Option[BatchLogWriter] =
{
+ if (!WriteAheadLogUtils.isBatchingEnabled(conf)) return None
+ val writer = checkpointDirOption.map(_ => new BatchLogWriter)
+ writer.foreach { runnable =>
+ new Thread(runnable, "Batch WAL Writer").start()
+ Runtime.getRuntime.addShutdownHook(new Thread {
+ override def run(): Unit = runnable.stop()
+ })
+ }
+ writer
+ }
+
/** Check if the write ahead log is enabled. This is only used for
testing purposes. */
private[streaming] def isWriteAheadLogEnabled: Boolean =
writeAheadLogOption.nonEmpty
+
+ /** A helper class that writes LogEvents in a separate thread to allow
for batching. */
+ private[streaming] class BatchLogWriter extends Runnable {
+
+ var active: Boolean = true
+
+ private def writeRecords(records: List[ReceivedBlockTrackerLogEvent]):
Unit = {
+ writeAheadLogOption.foreach { logManager =>
+ if (records.nonEmpty) {
+ logDebug(s"Batched ${records.length} records for WAL write")
+
logManager.write(ByteBuffer.wrap(Utils.serialize(CombinedRBTLogEvent(records))),
+ clock.getTimeMillis())
+ }
+ }
+ }
+
+ def stop(): Unit = {
+ active = false
+ }
+
+ private def flushRecords(): Unit = {
+ val buffer = new ArrayBuffer[ReceivedBlockTrackerLogEvent]()
+ while (!walWriteQueue.isEmpty) {
+ buffer.append(walWriteQueue.poll())
+ }
+ val records = buffer.toList
+ try {
+ writeRecords(records)
+ records.foreach(walWriteStatusMap.put(_, Success))
+ } catch {
+ case e: Exception =>
+ logWarning(s"Batch WAL Writer failed to write $records")
+ records.foreach(walWriteStatusMap.put(_, Fail))
+ }
+ }
+
+ override def run(): Unit = {
+ while (active) {
+ flushRecords()
+ Thread.sleep(WAL_WRITE_STATUS_CHECK_BACKOFF)
--- End diff --
If you can change `walWriteQueue` to `LinkedBlockingQueue`, you can use
`take` to wait for new records rather than using a while loop of `Thread.sleep`.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
