+1 (binding) On Thu, Oct 22, 2026 at 9:56 AM Anoop Johnson <[email protected]> wrote:
> Hi, everyone - > > Following the discussion on v4 stats tightness [1], I'd like to call a > vote on the spec change that replaces the per-column `tight_bounds` field > with a single file-level `tight_stats` flag. > > Summary of the change: > > - Stats tightness is tracked once per file with a `tight_stats` flag > on the tracked file instead of a `tight_bounds` field on every > column's stats struct. > - The tightness flag covers all statistics (bounds and null, NaN, and > value counts), not just bounds, so metadata-only queries can trust the > counts as well. > - When the `tight_stats` flag is absent, readers must assume stats are > not tight. > - A deletion vector no longer implicitly makes stats wide; > `tight_stats` is the authoritative indicator, and engines are allowed > to retighten stats without rewriting data files. > > References: > > 1. Discussion thread: > https://lists.apache.org/thread.html/tv7bvqctdfh5qng1sx5rp8r4r00nsv60 > 2. Design options doc: > https://docs.google.com/document/d/13DbwnwKQXjvY3vRcFJo-qVAQWk-GB4zLzzYmFdPntZg/ > > 3. Spec PR: https://github.com/apache/iceberg/pull/18585 > > Please vote within the next 72 hours. > > [ ] +1 Adopt the tight_stats spec change > [ ] +0 > [ ] -1 Do not adopt, with reason > > Thanks, > Anoop >
