Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
viirya commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2091997670 I think it makes more sense to use Arrow types as a bridge between Comet and Parquet reader in Iceberg. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
parthchandra commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2091903091 > For the Parquet part, we may need to define something like `CometDataType` which gets converted from the Parquet schema, and from which we can derive Spark catalyst data type or Iceberg data type. How about Arrow types as the canonical data types for Comet? ` org.apache.spark.sql.util.ArrowUtils` has conversions between Arrow and Spark schema/types. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
sunchao commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2089524991 The original purpose of `comet-common` module is to make it engine-agnostic so it can be used for other use cases like Iceberg. Unfortunately we didn't have time to make it completely isolated, so it is still tightly coupled with Spark in several ways like Parquet -> catalyst schema conversion, ColumnVector, and later on a bunch of shuffle related stuff which are all closely related to Spark. If necessary, we can perhaps consider splitting the module further into `comet-parquet`, `comet-spark-shuffle` etc. For the Parquet part, we may need to define something like `CometDataType` which gets converted from the Parquet schema, and from which we can derive Spark catalyst data type or Iceberg data type. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
advancedxy commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2084810855 > @advancedxy Good suggestions. I believe this Issue is to address point 3 above while 1 and 2 are in progress. Thanks for the clarification, it makes totally sense then. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
parthchandra commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2083237844 @advancedxy Good suggestions. I believe this Issue is to address point 3 above while 1 and 2 are in progress. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
advancedxy commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-208203 I'm +1 for this direction in the long term and I can help review the Iceberg integration if needed. In the short term, I think Iceberg could integrate Comet in its iceberg-spark module though, which doesn't require Comet's common module to be engine independent? So it would be great that we can make this work incrementally, such as: 1. release Comet 0.1(or any other first version) first 2. integrate Comet in Iceberg's spark module 3. refactor and making comet common module engine independent incrementally in the next release or various releases 4. integrate Comet in Iceberg's arrow/comet module and make the vectorized reader generally available for other engines in the iceberg repo. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
snmvaughan commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2079553631 +1 for this direction. We can start migrating in this direction by moving a subset of `Utils.scala` which is specific to the mapping between Spark and Arrow. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
viirya commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2078371802 This sounds a good direction to go. In the short term it might add some additional works that require us to refactor common and spark modules, though. Currently I'm still not sure about integrations with other engines. It is a great target, I think. Although to me it seems a little too far from current project status and bandwidth. 😄 I would like to focus on Spark integration at current stage. But if this refactoring is necessary to move Iceberg integration forward for now, I will support it. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
andygrove commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2078337974 I am +1 for making comet-common Arrow-native and easier to integrate with other engines. Let me know how I can help. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
parthchandra commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2078218574 Seems to me it would be a step in the right direction. The idea that comet-common should be independent of any engine is sound. It would be a necessary first step towards integration with other engines (e.g presto/trino). For the most part it looks like the Parquet file reader is mostly independent of Spark. The ColumnReaders and CometVector itself could be made more generic using only Arrow types and Arrow vectors while the adaptation to operate as a Spark columnar batch may be moved into comet-spark. This may involve a fair amount of difficult refactoring, but imho, would be worth it. -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
huaxingao commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2078208013 also cc @andygrove @parthchandra @snmvaughan @kazuyukitanimura -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
Re: [I] Making Comet Common Module Engine Independent [datafusion-comet]
viirya commented on issue #329: URL: https://github.com/apache/datafusion-comet/issues/329#issuecomment-2078204505 cc @sunchao -- 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. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org