Bruce Momjian <maillist@candle.pha.pa.us> writes:
> Tom, did we address this. I forgot.
No, it's still an open issue as far as I'm concerned. I was hoping to
hear something from Vadim about how pg_upgrade could work safely under
MVCC...
regards, tom lane
>> Bruce Momjian <maillist@candle.pha.pa.us> writes:
>>>>> BTW, it seems to me that it is a good idea to kill and restart the
>>>>> postmaster immediately after pg_upgrade finishes. Otherwise there might
>>>>> be buffers in shared memory that do not reflect the actual contents of
>>>>> the corresponding pages of the relation files (now that pg_upgrade
>>>>> overwrote the files with other data).
>>
>>>> Your issue with buffer cache is a major one. Clearly, this would be a
>>>> problem. However, it is my understanding that the buffer cache after
>>>> initdb would only contain system table info, so if they pg_upgrade after
>>>> that, there is no way they have bad stuf in the cache, right?
>>
>> Cached copies of system tables obviously are no problem, since
>> pg_upgrade doesn't overwrite those. I'm concerned whether there can
>> be cached copies of pages from user tables or indexes. Since we've
>> just done a bunch of CREATE INDEXes (and a VACUUM, if my latest hack
>> is right), it seems at least possible that this would happen.
>>
>> Now all those user tables will be empty (zero-length files), so there is
>> nothing to cache. But the user indexes are *not* zero-length --- it looks
>> like they are at least 2 pages long even when empty. So there seems
>> to be a real risk of having a cached copy of one of the pages of a user
>> index while pg_upgrade is overwriting the index file with new data...