Hiroshi Inoue <Inoue@tpf.co.jp> writes:
> Tom Lane wrote:
>> Why? We are able to construct all the non-nailed relcache entries
>> from scratch during backend startup. That seems a sufficient
>> proof that we can reconstruct any or all of them on demand.
> Hmm,why is it sufficent ?
> At backend startup there are no rel cache except
> some nailed rels. When 'reset system cache' message
> arrives,there would be many rel cache entries and
> some of them may be in use.
Doesn't bother me. The ones that are in use will get rebuilt.
That might trigger recursive rebuilding of system-table relcache
entries, and consequently recursive syscache lookups, but so what?
That already happens during backend startup: some relcache entries
are loaded as a byproduct of attempts to build other ones.
> In addtion there could be some inconsitency of db
> in the middle of the transaction. Is it safe to recon
> struct rel cache under the inconsistency ?
No worse than trying to start up while other transactions are
running. We don't support on-the-fly modification of schemas
for system catalogs anyway, so I don't see the issue.
regards, tom lane
