Integration tests should be exercised with all potential features flags and configuration toggles enabled
Many times a new feature and its tests are implemented behind a feature flag. Once the implementer is confident of a feature, the flag is removed and enabled by default. Before the flag is removed, other features will be tested without the feature flag. Any undesired behavior that happens from a disabled feature won't be caught in our tests.
For example, while working on removing a feature flag for info/ref cache, it was discovered that the partial clone tests did not account for caching (!2038 (comment 322020978)). This was a simple fix, but it could have been caught much sooner and prevented unexpected results in production had the feature flag been enabled in tests.
This also happens with configuration toggles. We have a number of Praefect configuration toggles for features (e.g. SQL leader election). We won't discover problems with different combinations of features enabled until a production incident or the feature is enabled by default.