You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@spark.apache.org by kundan kumar <ii...@gmail.com> on 2016/04/20 18:26:56 UTC
Executor still on the UI even if the worker is dead
Hi TD/Cody,
Why does it happen so in Spark Streaming that the executors are still shown
on the UI even when the worker is killed and not in the cluster.
This severely impacts my running jobs which takes too longer and the stages
failing with the exception
java.io.IOException: Failed to connect to --- (dead worker)
Is this a bug in Spark ??
Version is 1.4.0
This is entirely against the fault tolerance of the workers. Killing a
worker in a cluster of 5 impacts the entire job.
Thanks,
Kundan