Description
The kryo output buffer size for SArg is hardcoded right now to 100000. The hive implementation sets this to an initial size of 4096 and max size to 10485760.
Spark started using the apache orc implementation from 2.3 compared to the hive implementation. Spark jobs are failing with buffer overflow error due to the small size of the buffer here. We should make this configurable so that frameworks using the orc implementation can pass the size configs.
Attachments
Issue Links
- is related to
-
SPARK-27107 Spark SQL Job failing because of Kryo buffer overflow with ORC
-
- Resolved
-
- links to
