CG <cgg...@yahoo.com> writes:
> I have the fallback position for the Percona ETL strategy. But I feel like
> I'm REALLY close with pg_repack and I just don't understand enough about the
> system internals to nudge it to correctness and need some expert assistance
> to tap it in the hole.
The only "system metadata" that would need changing is
pg_class.relhasoids. However, it's certainly necessary to rewrite
all the physical table rows (and hence all the table's indexes).
I don't know a lot about pg_repack, but I'm a bit skeptical that
it can really make that transparent. I'm even more skeptical that
it could almost support it except this obvious use-case never
occurred to the developers. I suspect that either the support
already exists but you missed it, or else there's some fundamental
stumbling block. Maybe go discuss this with the pg_repack
developers? (Maybe they read pgsql-general, but I wouldn't count
on it.)
regards, tom lane
