Gregory Stark <stark@enterprisedb.com> writes:
> Should reindex be doing an in-place update?
Not if you'd like it to be crash-safe.
> Alternatively, why does the planner need access to the pg_class entry and not
> just the pg_index record?
For one thing, to find out how big the index is ... though if we could
get around that problem, it might indeed be possible to treat the
pg_index records as property of the parent table not the index itself,
which would give us license to read them without locking the index.
regards, tom lane