Heikki Linnakangas <heikki(at)enterprisedb(dot)com> writes:
> But I did notice that we're not fsyncing the newly written relation like
> we should.
Good point, but doesn't the analogy to copy_relation_data say that we
should sync if not rd_istemp? (This is my fault BTW; your original
patch kept the data in shared buffers, so it wasn't subject to the
problem.)
> BTW: In tablecmds.c the new relation is fsynced with smgrimmedsync, not
> heap_sync.
That's okay since that routine is just copying the one table. TOAST is
handled via recursion of ATExecSetTableSpace.
regards, tom lane