ebyhr commented on code in PR #11041: URL: https://github.com/apache/iceberg/pull/11041#discussion_r4624793478
########## format/view-spec.md: ########## @@ -42,12 +42,28 @@ An atomic swap of one view metadata file for another provides the basis for maki Writers create view metadata files optimistically, assuming that the current metadata location will not be changed before the writer's commit. Once a writer has created an update, it commits by swapping the view's metadata file pointer from the base location to the new location. +### Materialized Views + +Materialized views are a type of view with precomputed results from the view query stored as a table. +When queried, engines may return the precomputed data for the materialized views, shifting the cost of query execution to the precomputation step. + +Iceberg materialized views are implemented as a combination of an Iceberg view and an underlying Iceberg table, the "storage-table", which stores the precomputed data. +Materialized View metadata is a superset of View metadata with an additional pointer to the storage table. The storage table is an Iceberg table with additional materialized view refresh state metadata. +This addition is backward compatible: a reader that is unaware of materialized views sees the unknown `storage-table` field, ignores it, and evaluates the view query directly. Review Comment: One concern about the compatibility claim. @InvisibleProgrammer asked about bumping `format-version` in https://github.com/apache/iceberg/pull/11041#issuecomment-3709796824, and @stevenzwu replied in https://github.com/apache/iceberg/pull/11041#issuecomment-3873040325 that it isn't needed since there are no breaking changes. That's true for readers, but there's a problem on the writer side. An MV-unaware writer (e.g. an older Iceberg Java client running `ALTER VIEW ... SET TBLPROPERTIES` against a non-REST catalog) would silently remove `storage-table`, since it drops unknown fields when rewriting the metadata. The MV would become a plain view and the storage table would be orphaned. I don't think we can prevent this with spec text, because old libraries can't tell an MV from a view. What do you think about revisiting the bump? Old engines would then fail on MVs instead of silently dropping them. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
