Github user mridulm commented on the issue:
https://github.com/apache/spark/pull/17238
This is something to be configured appropriately at the lower level (namely
topology config), and should be handled there Special casing this in spark is
not right.
Two scenarios :
- Default rack for cpu nodes, and a special rack for gpu nodes
(specifically for rdma comm) - will cause rack locality to be disabled for the
cpu nodes (since they are on the default rack).
- Specializations of NetworkTopology need not honor DEFAULT_RACK -
resulting in limited value in those scenarios (IIRC the api does not require
unknown rack to be specified as DEFAULT_RACK) - is that incorrect assessment
@squito ?
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
