Thread: count(*) vs count(id)

On 2/1/21 4:53 PM, Matt Zagrabelny wrote:
> Greetings,
> 
> Is there a preferred method of counting rows?
> 
> count(*) vs count(field_name)
> 
> I have used count(*) for a long time and am hitting an inconsistency. At 
> least it is an inconsistency to me (I'm sure there is an explanation)...
> 
> INCORRECT OUTPUT (not desired results)
> $ select calendar.entry, count(*)  from 
> generate_series('2020-01-01'::date, '2021-01-10'::date, '1 
> day'::interval) as calendar(entry) left join call_records on 
> calendar.entry = call_records.timestamp::date group by calendar.entry;
> [...]
>   2020-08-30 00:00:00-05 │     1
>   2020-08-31 00:00:00-05 │     1
> [...]
> 
> CORRECT OUTPUT (desired results)
> $ select calendar.entry, count(id)  from 
> generate_series('2020-01-01'::date, '2021-01-10'::date, '1 
> day'::interval) as calendar(entry) left join call_records on 
> calendar.entry = call_records.timestamp::date group by calendar.entry;
> [...]
>   2020-08-30 00:00:00-05 │     0
>   2020-08-31 00:00:00-05 │     0
> [...]
> 
> What am I missing between count(*) and count(id)?
> 
> Thanks for any help!
> 
> -m

You got one null from count(*) likely.

Matt Zagrabelny <mzagrabe@d.umn.edu> writes:
> On Mon, Feb 1, 2021 at 5:57 PM Rob Sargent <robjsargent@gmail.com> wrote:
>> You got one null from count(*) likely.

> What is count(*) counting then? I thought it was rows.

Yeah, but count(id) only counts rows where id isn't null.

            regards, tom lane

select count(id) from call_records where id is null;

 count

═══════

     0

(1 row)

Time: 0.673 ms

Which field is count(*) counting if it is counting nulls?

-m

On Mon, 2021-02-01 at 19:14 -0600, Matt Zagrabelny wrote:
> > > What is count(*) counting then? I thought it was rows.
> > 
> > Yeah, but count(id) only counts rows where id isn't null.
> 
> I guess I'm still not understanding it...
> 
> I don't have any rows where id is null:

Then the *result* of count(*) and count(id) will be the same.

The asterisk in count(*) is misleading.  Different from any other
programming language that I know, the SQL standard has decided that
you cannot have an aggregate function without arguments.  You have
to use the asterisk in that case.

So count(*) really is count(), that is, it counts one for every
row that it finds, no matter what the row contains.

But count(id) includes a check: if "id IS NULL", it is not counted.

If that condition is satisfied for all "id"s, you end up with
the same count.  But count(id) is more expensive, because it
will perform this unnecessary NULLness check for each row.

In short: use count(*) if you want to count rows, and use
count(x) if you want to count all rows where x IS NOT NULL.

Yours,
Laurenz Albe
-- 
Cybertec | https://www.cybertec-postgresql.com

Am Wed, Feb 03, 2021 at 01:43:14AM -0500 schrieb Cherio:

> I just ran a few practical tests on large (~14mil rows) tables that have
> multiple indexes.
>
> SELECT COUNT(id) forces PostgreSQL to use the primary key index.
> SELECT COUNT(*) allows PostgreSQL to chose an index to use and it seems to
> be choosing one of smaller size which leads to less IO and hence returns
> the result faster.

Would you mind throwing in a test for

    select count(1) ...

?

Karsten
--
GPG  40BE 5B0E C98E 1713 AFA6  5BC0 3BEA AC80 7D4F C89B