vanzin commented on a change in pull request #24817: [SPARK-27963][core] Allow
dynamic allocation without a shuffle service.
URL: https://github.com/apache/spark/pull/24817#discussion_r298358761
##########
File path:
core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala
##########
@@ -264,7 +426,18 @@ private[spark] class ExecutorMonitor(
def updateTimeout(): Unit = {
val oldDeadline = timeoutAt
val newDeadline = if (idleStart >= 0) {
- idleStart + (if (cachedBlocks.nonEmpty) storageTimeoutMs else
idleTimeoutMs)
+ val timeout = if (cachedBlocks.nonEmpty || (shuffleIds != null &&
shuffleIds.nonEmpty)) {
+ val _cacheTimeout = if (cachedBlocks.nonEmpty) storageTimeoutMs else
Long.MaxValue
+ val _shuffleTimeout = if (shuffleIds != null && shuffleIds.nonEmpty)
{
Review comment:
This was intentional. The `hasActiveShuffle` tracking is not meant to change
when the executor times out, just when it should not be removed.
Doing that would mean that every time a job starts with a shuffle that this
executor has data, its timeout would be reset, even if it ends up not running
any tasks. (Which is unlikely, but well.)
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
