Github user mateiz commented on a diff in the pull request:
https://github.com/apache/spark/pull/159#discussion_r10683261
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala ---
@@ -59,6 +59,15 @@ private[spark] class TaskSetManager(
// CPUs to request per task
val CPUS_PER_TASK = conf.getInt("spark.task.cpus", 1)
+ /*
+ * Sometimes if an executor is dead or in an otherwise invalid state,
the driver
+ * does not realize right away leading to repeated task failures. If
enabled,
+ * this temporarily prevents a task from re-launching on an executor
where
+ * it just failed.
+ */
+ private[this] val EXECUTOR_TASK_BLACKLIST_TIMEOUT =
+ conf.getLong("spark.task.executorBlacklistTimeout", 0L)
--- End diff --
Regarding `private[this]`, we don't really use it elsewhere, so I'd skip it
here. It would be a bigger change to convert the whole codebase to
`private[this]`. Its main benefit seems to be in terms of code generated but
that kind of stuff usually gets inlined anyway.
I now see that this is per-task, not a blacklist for the whole executor, so
it makes sense for the variable name to contain "task". In that case maybe call
this `spark.scheduler.executorTaskBlacklistTime` -- I think it's good to put
"scheduler" as the first component there since it's not really a property of
tasks.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
