openinx commented on pull request #2354: URL: https://github.com/apache/iceberg/pull/2354#issuecomment-809869373
> basically the row key information at that point of time is already in the delete file written at that time. @jackye1995 @rdblue Yes, I can understand that we've maintained the equality field ids inside each equality delete files and iterating records from an old snapshot should be correct because we've considered [applying process](https://github.com/apache/iceberg/blob/90225d6c9413016d611e2ce5eff37db1bc1b4fc5/data/src/main/java/org/apache/iceberg/data/DeleteFilter.java#L126) for different equality fields ids. But in this [comment](https://github.com/apache/iceberg/pull/2354#issuecomment-809033279), I'm not discussing the data correctness (the data correctness has no problem). I mean people should be able to read the old `RowKey` from an older snapshot. In this issue https://github.com/apache/iceberg/issues/1029, we are proposing to attach the `schema` in snapshot so that users could create a complete table based on an existing snapshot by using CTAS or RTAS, in this scenarios I think we also need to use the old `PartitionSpec`, `SortOrder`, `RowKey` in the newly created table. Then at the iceberg table level, people could see the consistent table metadata. Another case is rollback to an older snapshot by replacing the latest snapshot, though we currently do not support replacing schema/partition-spec/sort-order with the old one, but in my mind I think we'd better to because providing an uniformed view of data files, schema and other table metadata at the old timestamp `t3` matches the expectations from end users. That's why I recommended to track the multiple versions of `RowKey` in iceberg table metadata. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
