Github user sryza commented on the pull request:
https://github.com/apache/spark/pull/6652#issuecomment-109444936
My worry is the following situation:
* Map outputs from stage 1 exhibit little skew and are distributed evenly
across nodes.
* 5 nodes are chosen as preferred for stage 2. The choice is basically
random because of the lack of skew.
* Stage 2 has fewer tasks than the number of slots.
* Instead of stage 2's tasks evenly distributed across the cluster, all of
them end up on the 5 nodes, which are more likely to become bottlenecked on
incoming network traffic.
This isn't a disaster situation by any means. But it also seems like it
could be easily avoided with a policy that only prefers locations when it would
make a difference.
---
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]
