Github user fidato13 commented on the issue:
https://github.com/apache/spark/pull/15327
Hi @srowen
The Spark SQL property/algo that binaryFiles partition calculation is now
implementing is the property "spark.files.openCostInBytes" (in
org.apache.spark.sql.internal.SQLConf) utilized in
org.apache.spark.sql.execution.FileSourceScanExec inside method
createNonBucketedReadRDD .
As the original issue states that sc.binaryFiles is always creating an RDD
with number of partitions as 2 (irrespective of number and size of files say
for 1000 files or for even GB's of file , It will only create two partitions
only).
To resolve , Reynold suggested that we can probably implement somthing like
Spark SQL does i.e. to consider the cost of opening a file(more description in
the JIRA issue) so that it doesn't create large number of partitions for small
files.
Please let me know your suggestions on this, I have now verified that it's
now creating an optimized number of partitions . Also, since both properties
are used for their respective config's and for the SPARK CORE it's used just
once for this issue , Can you please advise about if it would be okay to keep
them or How would you like to proceed on this.
Thanks
---
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]
