pinker <pinker(at)onet(dot)eu> writes:
> I was analysing shared buffers content and noticed that exactly the same
> disk block appears there many times with different or the same usagecount.
Postgres would be completely broken if that were true, because
modifications made to one copy would fail to propagate to other copies.
I don't know where your data came from, but it can't be an accurate
representation of the instantaneous state of the buffer cache.
... actually, after looking at your query, I wonder whether the issue
is that you're failing to include database and tablespace in the
grouping key. relfilenode isn't guaranteed unique across directories.
The fork number can matter, too.
regards, tom lane