Robert Haas <robertmhaas@gmail.com> writes:
> On Sun, Jul 12, 2015 at 12:02 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> As best I can tell (evidence below), the SQL standard requires that if a
>> single query reads a table with a TABLESAMPLE clause multiple times (say,
>> because it's on the inside of a nestloop), then the exact same set of
>> sampled rows are returned each time.
> Hmm, I tend to agree that it would be good if it behaved that way.
> Otherwise, it seems like the behavior could be quite surprising.
Yeah. As a concrete example, consider
select * from t1, t2 tablesample ... where t1.x = t2.x
and suppose that there are multiple occurences of x = 10 in both tables.
As things stand, if the join is done as a nestloop then a particular t2
row with x = 10 might appear in the output joined with some of the t1 rows
with x = 10 but not with others. On the other hand, the results of a hash
join would not be inconsistent in that way, since t2 would be read only
once.
regards, tom lane