Bruce Momjian <pgman@candle.pha.pa.us> writes:
>> Done and done. I also realized that pg_upgrade had another nasty bug
>> in it: the VACUUMs were not necessarily executed as superuser, but as
>> whichever user happened to own the item dumped last by pg_dump in each
>> database. That would result in VACUUM skipping over tables it thought
>> it didn't have permission to vacuum --- like, say, all the system
>> tables. Perhaps this explains the failures that some people have
>> reported.
>>
>> Another day, another bug swatted ...
> Yes, good eye. Thanks.
BTW, I believe that this was actually a bug I created --- in prior
releases VACUUM would happily vacuum anything in sight, whether you
had privileges to lock the target tables or not. So pg_upgrade didn't
have to worry before. But it does now...
regards, tom lane