It's not my concern about this, it's why I think the rewrite procedure does what it does currently. What I mean is that I don't think the initial implementation didn't rewrite files across different partition specs because the manifest files themselves would have a different schema for the partition tuple. That makes passing the data around a bit harder if you want to use data frames.
On Thu, Dec 7, 2023 at 1:12 PM Pucheng Yang <py...@pinterest.com.invalid> wrote: > Ryan, got it. Can you speak more for your first concern? If we are to > rewrite manifests of the same spec at a time, how will that lead to > inconsistent schema? > > As of now, we have a PR ready to be merged > https://github.com/apache/iceberg/pull/9242 please comment if you suggest > this to be put on hold, thanks > > On Thu, Dec 7, 2023 at 1:00 PM Ryan Blue <b...@tabular.io> wrote: > >> It sounds like a feature that we could add. I think there are two >> concerns. First, we don't want to process the files for multiple specs at >> the same time because we probably want a consistent schema. Second, there's >> probably some confusion over how to select a spec since we don't like users >> to need to work with IDs in the format directly. >> >> On Wed, Dec 6, 2023 at 10:28 PM Pucheng Yang <py...@pinterest.com.invalid> >> wrote: >> >>> Based on what I understand, it seems there is no particular reason, and >>> seems like a feature to be added on. >>> >>> On Wed, Dec 6, 2023 at 8:06 PM Pucheng Yang <py...@pinterest.com> wrote: >>> >>>> Hi community, >>>> >>>> May I know why manifest rewrite will only touch files that have the >>>> latest spec id? What will be the suggestion if we want to rewrite manifest >>>> files that belong to non current spec id? >>>> >>>> Manifest selection logic: >>>> https://github.com/apache/iceberg/blob/6a9d3c77977baff4295ee2dde0150d73c8c46af1/spark/v3.5/spark/src/main/java/org/apache/iceberg/spark/actions/RewriteManifestsSparkAction.java#L295 >>>> >>>> Best, >>>> Pucheng >>>> >>> >> >> -- >> Ryan Blue >> Tabular >> > -- Ryan Blue Tabular