Github user mridulm commented on the issue:
https://github.com/apache/spark/pull/20998
@squito My concern is, in large workloads, some nodes simply become bad for
some tasks (transient env or hardware issues, colocating containers, etc) while
being fine for others; speculative tasks should alleviate performance concerns
and not increase chances of application failure due to locality preference
affinity. For cluster sizes which are very small, speculative execution is less
relevant than for those which are large - and here we are tuning for the former.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
