Hello,
Thanks for the review. I have addressed the comments in the latest draft:
https://docs.google.com/document/d/1REmpQ2r5W12OvW2CU9tNjDWjpkFJegiTZnE1IZbBH_U/edit?tab=t.0

  1.
The ambiguous delete-encoding validation was removed. Format-version 
compatibility is checked before table creation, while delete semantics are 
handled by the Iceberg reader during bootstrap.
  2.
Nested identifier fields are rejected during Coordinator-side validation. Only 
top-level identifier fields are supported.
  3.
Two bucket-mapping examples were added, covering both day(ts), bucket(16, id) 
and day(ts) without an Iceberg bucket transform.

Thanks,
Liting

________________________________
发件人: Muhammet Orazov via dev <[email protected]>
发送时间: 2026年7月1日 21:21
收件人: [email protected] <[email protected]>
抄送: Muhammet Orazov <[email protected]>
主题: Re: 回复: [DISCUSS] Draft design: Support upgrading existing Iceberg tables 
to Fluss tables

Hello Liting,

Thanks for the design draft!

I have couple questions/clarifications on top of the latest addressed
version.

1)

> If the selected Iceberg reader cannot guarantee correct
> snapshot-visible rows for a table that contains equality deletes,
> position deletes, deletion vectors, or any other delete encoding,
> create_from_lake rejects the table with a clear validation error.

For me this is not clear, if above check happens on validation phase or
discovered once the selected reader is running.

2)

> Iceberg tables with non-empty identifier fields are mapped to Fluss
> primary-key tables and then follow the same bootstrap pipeline as
> FIP-25.

The Iceberg identifier fields may be nested in structs, how would we map
these identifier fields? I assume it will be rejected.

3)

For the Bucket Configuration section, it would help to add some
examples.

Iceberg with combined partitioning: `day(ts), bucket(16, id)` and
identifier field `id`, how this will be transformed into Fluss PK table?
Maybe good idea to add some examples, given iceberg partitioning
combinations to the Fluss tables mapping.



Reply via email to