aokolnychyi opened a new pull request #569: Use broadcast variables in 
IcebergSource
URL: https://github.com/apache/incubator-iceberg/pull/569
 
 
   In some Spark jobs, we see a substantial scheduler delay, which happens in 
`TaskSetManager` when Spark serializes Iceberg `ReadTask`. The latter contains 
a reference to `FileIO` (which can contain a full Hadoop conf).
   
   The current idea is to broadcast `EncryptionManager` and `FileIO` in 
`IcebergSource`. Then `Reader` and `ReadTask` can store references to the 
broadcasted values and fetch actual ones in `createPartitionReader` while 
creating `TaskDataReader`.
   
   All broadcasted values are automatically registered for clean-up in 
`SparkContext$broadcast`.
   
   **Note!** Using this approach with Kryo requires providing a custom 
registrator to instruct Spark to apply Java serialization for 
`SerializableConfiguration`. We should extend #549 for this.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to