Thread: Plan rows - 1 or many

Hi,

I have an application where I would really like to be able to look at en
SQL query and answer the question:

"Is this query capable of returning more than 1 row?"

So basically, queries are divided into 2 categories. Those that look up
a single row (if it exists) and those who return a (possible empty) set
of rows. (which could also be a set of 1 due to current database content.)

Example:
It's obvious to see that this query can never return more than one row,
no matter what's in the DB:
SELECT * FROM mytable WHERE uniquecolumn = 17;

But only because we know the column has a unique constraint.
Now, unique constraints might be more complex. It could be composite
(A,B) where A is a foreign key to another table (Ta) where we can
determine that both row referenced by A is unique wrt. the query, so
filtering on B and a unique column in Ta would still only ever result in
1 row.

SELECT * FROM Ta,Tb WHERE Ta.id = Tb.A AND Tb.B = 17 AND Ta.uniquecolumn
= 42;

This seems to be about the same as the query planners "plan_rows" try to
do. I'm aware that plan_rows is only an estimate and can't answer the
question of exactly how many rows in general a query returns, but the
question is whether it can be trusted to answer the above more narrow
question?

I tried look at the source, but could quite figure out if plan_rows were
intended to be strict about when it concluded that there was only one
row. (like when it evaluated a GROUP BY with only 1 aggregate).

Can I conclude that when plan_rows is 1 then there will never be more
than 1 row returned by the query?

/Peter

Peter Mogensen wrote:
> I have an application where I would really like to be able to look at en
> SQL query and answer the question:
> 
> "Is this query capable of returning more than 1 row?"

> Can I conclude that when plan_rows is 1 then there will never be more
> than 1 row returned by the query?

I would say no.

If the planner estimates one row, that means that it guesses that
it will be one or less.

And a query that can potentially return many rows can certainly
also return only one row or none at all.

Yours,
Laurenz Albe

On 2014-03-12 09:28, Albe Laurenz wrote:
> Peter Mogensen wrote:
>> I have an application where I would really like to be able to look at en
>> SQL query and answer the question:
>>
>> "Is this query capable of returning more than 1 row?"
>
>> Can I conclude that when plan_rows is 1 then there will never be more
>> than 1 row returned by the query?
>
> I would say no.
>
> If the planner estimates one row, that means that it guesses that
> it will be one or less.

I think "guesses" is the operative word here. Because, if the planner
can actually promise that for the example queries a gave there would be
"one or less" rows, then I'm happy.

> And a query that can potentially return many rows can certainly
> also return only one row or none at all.

But that's fine... I have no problem with an answer telling me that this
query can return any number of rows.

Hmm.. I think my question is answered by posing it in another way:

Could "plan rows" tell me whether the result of a given query could be
changed by later INSERT statements?

And the answer becomes "no", because there's a difference between
queries of the type I gave as example and queries using aggregates and
GROUP BY. ... for which the planner would also return "plan_rows = 1".
However... GROUP BY queries could have their result changed by later
INSERT statements.

/Peter