Github user andrewor14 commented on the pull request:
https://github.com/apache/spark/pull/2746#issuecomment-59857732
@kayousterhout @sryza Correct me if I'm wrong, but I believe grabbing as
many executors as needed in MR/Tez comes for free in these two frameworks
because a container is supposed to be short-lived (unless you explicitly
specify to reuse containers). Thus, it is much easier to understand this policy
in these contexts because the same policy is used for normal scheduling of
tasks. In Spark, however, executors are mostly coarse-grained, so we have to
use the ratio of pending tasks per executor to achieve something similar.
However, this is more complicated for Spark because the correlation between
number of executors added and the number of pending tasks these executors will
be scheduled is less clear.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
