Thanks for making me elaborate, because I'm probably wrong. I'll step
through what my thinking was though.
'.annotate(local=F('related__field'))' will join to the related table. If
`related` is a nullable foreign key, then a join would do an implicit
filter, removing the row from the result set if it had no relation, and
reducing the count.
But Django models a nullable foreign key join as a LEFT JOIN which wouldn't
filter the row out and wouldn't reduce the count.
Multivalue relationships (reverse foreign key and m2m joins) would need
some care though I think. I'm not sure if the same annotation above works
on m2m joins or not.
On Sunday, 20 August 2017 09:41:43 UTC+10, Tom Forbes wrote:
>
> Thanks for your reply Josh. Can you elaborate on why optimizing out '
> .annotate(local=F('related__field'))' would not be safe?
>
> On 20 Aug 2017 00:10, "Josh Smeaton" <josh.s...@gmail.com <javascript:>>
> wrote:
>
> I'd like to see this provided all bases were covered. I'll just list below
> the cases where I think it wouldn't be safe.
>
> - Filtered annotations
> - Annotations that join to anything other than a non-null foreign key:
> .annotate(local=F('related__field'))
> - Annotations that have a GROUP BY on fields that are not just the PK or
> entire field set (with .values())
>
> The group by one I'm unsure of, because I forget if a count is called with
> an aggregate query as a subquery or if it's an error.
>
> I think an optimisation is available if:
>
> 1. No filters ref an annotation
> 2. values() has not been called
> 3. There are no joins at all
>
> Further optimisation scenarios may be available with stricter rules (such
> as join types), but we could look at doing that separately if it was
> difficult.
>
> On Sunday, 20 August 2017 03:11:13 UTC+10, Tom Forbes wrote:
>>
>> Hello,
>> I think there is potential room for improvement with how Django handles
>> count() queries, specifically relating to annotations. Currently any
>> annotations that are added to a queryset are included in the SQL statement
>> that is generated in the count() query, including all joins and SQL
>> function calls - I've personally been bitten by this with DRF and a query
>> with a particularly complex set of annotations where the count query took
>> longer than fetching the data itself, but was semantically equivalent to a
>> much faster, plain Model.objects.count() call.
>>
>> I think in most cases annotations can be stripped, for example in the two
>> queries below the annotation is not filtered on so it can be safely removed
>> from the query without affecting the result:
>>
>> one = Book.objects.annotate(Count('chapters')).count()
>> two = Book.objects.count()
>>
>> I wanted to gather some feedback from the group as to whether this always
>> holds true: if an annotation is not filtered (or ordered on) it can always
>> be safely removed from a count query? I wouldn't be surprised if there is a
>> case where this isn't safe, but I cannot think of one.
>>
>> I've made an initial merge request that removes annotations from
>> querysets that have no filter or ordering that depends on annotations here:
>> https://github.com/django/django/pull/8928. It seems like Django also
>> has all the information to optimize out annotations that are not directly
>> or indirectly used by filters, but I wanted to gather some feedback before
>> I attempted this.
>>
>> Tom
>>
> --
> You received this message because you are subscribed to the Google Groups
> "Django developers (Contributions to Django itself)" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to django-develop...@googlegroups.com <javascript:>.
> To post to this group, send email to django-d...@googlegroups.com
> <javascript:>.
> Visit this group at https://groups.google.com/group/django-developers.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/django-developers/506d0f2c-51e8-400f-a4fb-66c2b8597aeb%40googlegroups.com
>
> <https://groups.google.com/d/msgid/django-developers/506d0f2c-51e8-400f-a4fb-66c2b8597aeb%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
>
>
--
You received this message because you are subscribed to the Google Groups
"Django developers (Contributions to Django itself)" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to django-developers+unsubscr...@googlegroups.com.
To post to this group, send email to django-developers@googlegroups.com.
Visit this group at https://groups.google.com/group/django-developers.
To view this discussion on the web visit
https://groups.google.com/d/msgid/django-developers/fb0c0183-2a39-4b2e-84cd-86e76d233487%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
