[CELEBORN-555][REFACTOR] Avoid prin noisy blacklist info when record blacklist (#1460)
* [CELEBORN-555][REFACTOR] Avoid prin noisy blacklist info when record blacklist
This commit is contained in:
parent
13ce04f8a1
commit
4bbc8aec4f
@ -99,25 +99,37 @@ class WorkerStatusTracker(
|
||||
case _ =>
|
||||
}
|
||||
}
|
||||
if (!failedWorker.isEmpty) {
|
||||
recordWorkerFailure(failedWorker)
|
||||
}
|
||||
recordWorkerFailure(failedWorker)
|
||||
}
|
||||
|
||||
def recordWorkerFailure(failures: ShuffleFailedWorkers): Unit = {
|
||||
val failedWorker = new ShuffleFailedWorkers(failures)
|
||||
logInfo(s"Report Worker Failure: ${failedWorker.asScala}, current blacklist $blacklist")
|
||||
failedWorker.asScala.foreach { case (worker, (statusCode, registerTime)) =>
|
||||
if (!blacklist.containsKey(worker)) {
|
||||
blacklist.put(worker, (statusCode, registerTime))
|
||||
} else {
|
||||
statusCode match {
|
||||
case StatusCode.WORKER_SHUTDOWN |
|
||||
StatusCode.NO_AVAILABLE_WORKING_DIR |
|
||||
StatusCode.RESERVE_SLOTS_FAILED |
|
||||
StatusCode.UNKNOWN_WORKER =>
|
||||
blacklist.put(worker, (statusCode, blacklist.get(worker)._2))
|
||||
case _ => // Not cover
|
||||
if (!failures.isEmpty) {
|
||||
val failedWorker = new ShuffleFailedWorkers(failures)
|
||||
val failedWorkerMsg = failedWorker.asScala.map { case (worker, (status, time)) =>
|
||||
s"${worker.readableAddress()} ${status.name()} $time"
|
||||
}.mkString("\n")
|
||||
val blacklistMsg = blacklist.asScala.map { case (worker, (status, time)) =>
|
||||
s"${worker.readableAddress()} ${status.name()} $time"
|
||||
}.mkString("\n")
|
||||
logInfo(
|
||||
s"""
|
||||
|Reporting Worker Failure:
|
||||
|$failedWorkerMsg
|
||||
|Current blacklist:
|
||||
|$blacklistMsg
|
||||
""".stripMargin)
|
||||
failedWorker.asScala.foreach { case (worker, (statusCode, registerTime)) =>
|
||||
if (!blacklist.containsKey(worker)) {
|
||||
blacklist.put(worker, (statusCode, registerTime))
|
||||
} else {
|
||||
statusCode match {
|
||||
case StatusCode.WORKER_SHUTDOWN |
|
||||
StatusCode.NO_AVAILABLE_WORKING_DIR |
|
||||
StatusCode.RESERVE_SLOTS_FAILED |
|
||||
StatusCode.UNKNOWN_WORKER =>
|
||||
blacklist.put(worker, (statusCode, blacklist.get(worker)._2))
|
||||
case _ => // Not cover
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user