TypeError raised in get_results_for_interval

[iaindillingham]

analysis/measures_test_hi.py (here) passed locally but failed in production with a TypeError (here). Here

ehrql/ehrql/measures/calculate.py

Lines 131 to 133 in f7f8e39

	# Denominator cannot be None because population only includes rows where
	# denominator is non-empty
	totals[1] += denominator

totals[1] was of type int, but denominator was of type NoneType.

Thanks for reporting this, @alschaffer.

[iaindillingham]

@alschaffer reports that this passed in production on second run.

[inglesp]

Maybe the Python interpreter noticed this comment on its second run?

            # Denominator cannot be None because population only includes rows where
            # denominator is non-empty
            totals[1] += denominator

[evansd]

@alschaffer reports that this passed in production on second run.

Ah, crap. This is going to be a lot harder to debug if we can't reproduce.

[evansd]

Ah ha! The timing of this error (17:04) is suspiciously close to when Slack reported entering DB maintenance (17:40). And the assumption of logical consistency with respect to denominator not being NULL only holds if the data doesn't change between running the population query and running the final query.

There's a wider question here, which this ticket probably isn't the best place to discuss.

[evansd]

Closing in favour of above job-runner ticket and this isn't an ehrQL-specific issue.