mirror of
https://github.com/netbox-community/netbox.git
synced 2025-12-28 23:57:10 +00:00
fb004bb94e
* Fixes #20327: Device queries are now faster when including ConfidContexts Move .distinct() from main queryset to tag subquery to eliminate performance bottleneck when querying devices with config contexts. The .distinct() call on the main device queryset was causing PostgreSQL to sort all devices before pagination, resulting in 15x slower API responses for large installations (10k+ devices, 100+ config contexts). Moving .distinct() to the tag subquery eliminates duplicates at their source (GenericForeignKey tag relationships) while preserving the fix for issues #5314 and #5387 without impacting overall query performance. * Add performance regression test for config context annotation The test verifies that: - Main device queries do not use expensive DISTINCT operations - Tag subqueries properly use DISTINCT to prevent duplicates from issue #5387 This ensures the optimization from issue #20327 (moving .distinct() from maintaining query to tag subquery) cannot be accidentally reverted while maintaining the correctness guarantees for issues #5314 and #5387. * Address PR feedback, clean up new regression test The new regression test now avoids casting the query to a string and inspecting the string, which was brittle at best. The new approach asserts directly against `queryset.distinct` for the main query and then finds the subquery that we expect to have distinct set and verifies that is in fact the case. I also realized that the use of `connection.query_log` was problematic, in that it didn't seem to return any queries as expected. This meant that the test was actually not making any assertions since none of the code inside of the for loop over `device_queries` ever ran.