Hi all,

I opened PR #17164 about a month ago and it hasn't picked up a reviewer yet.
The stale bot flagged it this week, so I'm following its suggestion to raise
it here.

  https://github.com/apache/iceberg/pull/17164

The change adds an optional "metadata-file-size-bytes" metric to
CommitMetricsResult, so MetricsReporter implementations can observe how
large
the metadata.json produced by a commit is. Today there is no way to see that
through the metrics reporting API, which makes metadata bloat hard to track
--
and some catalogs enforce a hard cap on metadata.json size (BigLake, for
example, limits it to 1 MiB), so operators have no signal before they hit
it.

The size is captured at write time from the PositionOutputStream rather
than by
re-reading the file after the commit, to avoid adding a metadata read to the
commit path.


I'm happy to rework the approach if there's a preferred way to surface this.
Would anyone be willing to take a look?

Thanks,
Amir Mor

Reply via email to