Is your feature request related to a problem? Please describe.
We are looking at the max input bytes in the auto tuner to decide what to set spark.sql.files.maxPartitionBytes to. But the value we get is the compressed size. Many times these sizes can explode and our setting is to high.
For GPU event logs see if we can get the uncompressed size. We may need to add another metric to the Spark rapids plugin for this.
Is your feature request related to a problem? Please describe. We are looking at the max input bytes in the auto tuner to decide what to set
spark.sql.files.maxPartitionBytes
to. But the value we get is the compressed size. Many times these sizes can explode and our setting is to high.For GPU event logs see if we can get the uncompressed size. We may need to add another metric to the Spark rapids plugin for this.