queryset.count() optimized

[somitmittal]

Sorry this is my first time contributing to open source,

In this PR, I have made the changes related to how count of queryset is evaluated.
SO currently, what we do is :
list_length = iterable.count()
This leads to a SQL query of the form : SELECT COUNT(*) as __count which is time consuming at SQL level because it does a count on the all records,

Solution:
Since we are only interested in the queryset count or number of records for pagination purposes, I feel the optimized or faster way to do this would be to run a SQL Query like : SELECT COUNT('pk') as pk__count where pk refers to the primary key in the db table. This query would definitely consume lesser time than SELECT COUNT(*) as it will do a count on primary key.
TO do that we can change the line of code from
list_length = iterable.count()
TO :
list_length = iterable.aggregate(Count('pk'))['pk__count'] // this will run a query like SELECT COUNT('pk') as pk__count

I feel this change can bring some speedup in graphql queries as we are now only fetching count of primary key rather than count(*) which can give some speed advantages.

Please correct me if I am wrong. Thanks :)

[ulgens]

Do we have any documentation or benchmark about the performance increase provided by this change?

[keithhackbarth]

https://dba.stackexchange.com/questions/41090/postgres-count-vs-countid
https://stackoverflow.com/questions/2710621/count-vs-count1-vs-countpk-which-is-better

I'm pretty sure COUNT(*) is as fast if not faster than COUNT(id) on most databases. And even if it wasn't I think it would have to be significantly faster to warrant the non-standard syntax. I would suggest closing this PR.