Start to use explicit memory limits in the parquet chunked reader #9991
Conversation
Signed-off-by: Robert (Bobby) Evans <bobby@apache.org>
revans2
added
the
cudf_dependency
sameerz
added
the
reliability
| val passReadLimit = if (useSubPageChunked) { | ||
| 4 * chunkSizeByteLimit |
There was a problem hiding this comment.
Can we set some multiplier constant or configurable constant for it, instead of hard coding 4X like this? In the (near) future, chunked ORC reader may have benefit from it.
There was a problem hiding this comment.
I'm not sure where would be a good place to put this. It is a magic number based on our estimates that we will take 4x the target batch size as the amount of memory that we are allowed to use. GpuDataProducer is the only place that might be common between them, but that is not a proper place for it. I could create a static object to hold it somewhere. GpuConventionMagicNumbers or something.
There was a problem hiding this comment.
Okay then let's see what can we do the next time with the ORC chunked reader. I'm fine to leave this as-is for now.
sql-plugin/src/main/scala/com/nvidia/spark/rapids/RapidsConf.scala
Outdated
Show resolved
Hide resolved
Signed-off-by: Robert (Bobby) Evans <bobby@apache.org>
|
build |
|
@ttnghia could you please take another look? |
| .doc("Enable a chunked reader where possible for reading data that is smaller " + | ||
| "than the typical row group/page limit. Currently this only works for parquet.") | ||
| .booleanConf | ||
| .createWithDefault(false) |
There was a problem hiding this comment.
We should have a followup issue to turn this on in 24.04.
This depends on rapidsai/cudf#14360