add conditional check before adding counts

[matthewhegarty]

With large datasets, avoiding the count check results in a large performance gain (see #122).

This PR adds a conditional check to avoid adding counts. It depends on a new REST_FRAMEWORK_DATATABLES dict in settings.py

The default is to include the count so it won't break any existing implementations.

David - if you are ok with this approach I will add tests and documentation.

[codecov-commenter]

Codecov Report

Merging #138 (4d13d14) into master (433457f) will not change coverage. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##            master      #138    +/-   ##
==========================================
  Coverage   100.00%   100.00%            
==========================================
  Files            8         6     -2     
  Lines          417       307   -110     
==========================================
- Hits           417       307   -110     

Impacted Files	Coverage Δ
...st_framework_datatables/django_filters/backends.py	100.00% <100.00%> (ø)
rest_framework_datatables/__init__.py
rest_framework_datatables/filters.py

:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more

[matthewhegarty]

I've been testing this afternoon and I'm going to close the PR because I find that optimising the count has only a very small effect.

• Using Example app
• Loaded 55m albums into a postgres db (using the create_albums.py helper script)
• Started example app
• Used 'Example using column filters'
• Note: You have to create a new album after the server starts (use create_albums.py)
  • this is so that the queryset cache is invalidated, and a new count() call is made
  • If this is not done, then there is no performance improvement because the queryset cache is used
• With DISABLE_COUNT=False: Filtered on 'Miles Davis' - DT response time ~11 seconds
• With DISABLE_COUNT=True: Filtered on 'Miles Davis' - DT response time ~10 seconds

The optimisation is to prevent the initial call to count(). The effectiveness of enabling the flag will depend on how expensive that call is. If it takes several seconds then it is worth enabling the flag if you don't need the 'filtered from N total entries' text. However in the majority of cases I don't think this will be of benefit.

disable_count_tests.txt