Dimitri Fontaine <dfontaine@hi-media.com> writes:
> Well at bootstrap time I guess noone is able to disturb the system by
> placing a concurrent CLUSTER pg_class; call. Once started, do those rels
> still need to have a special behavior?
It doesn't matter, if you fail to get past bootstrap because you
couldn't find pg_class. The existing design for this is absolutely
dependent on the fact that pg_class has a fixed relfilenode = fixed
physical file name. MVCC has nothing to do with it.
See my followon message for a sketch of a possible solution. Basically
it's pushing the fixed file name over to another place ...
regards, tom lane