count estimation on querysets

[Denis Cornehl]

on bigger tables a normal count() gets slow quickly, though quite often we only need a rough estimation (for pagination for example, see #26451 ).

I want to add a new estimate_count method on QuerySets to allow users to access this count estimation.

For Postgres and other databases this is possible by using the query planner statistics to do this estimation.

[Simon Charette]

This has already be mentioned in #11287 and it was deemed hard to implement such a function returning reliable results. There's multiple way of achieving this on ​PostgreSQL alone which makes me feel this is somewhat application specific.

I'm -1 for adding this only for PostgreSQL and -0 if we come up with a reliable method on all backends.

[Simon Charette]

This is especially hard to do as soon as your queryset as any filters applied to it.

[Denis Cornehl]

@Simon thanks for the article, with the method I want to use (explain output based) I don't see any big problem there.

It's an estimation, and will never be exact, and often outdated.

What would be a valid definition for "reliable enough" we could follow here?

Adding this to all backends is definitely quite some work, since all of them have statistics, but I do believe it's not so simple to translate them into this count estimation, as it is for postgres