Round 1:

    100k:  15 minutes 20 seconds
    1k:    0 minutes 7 seconds

        %timeit {c['_type']: c['count'] for c in repo_version.content.values('_type').annotate(count=Count('_type'))}
        >20 minutes (cut off)

        %timeit list(Content.objects.filter(pk__in=repo_version.content))
        4.52 s ± 156 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)


Round 2 (101,000 pre-existing units):

    100k:  10 minutes 8 seconds
    1k:    0 minutes 17 seconds

        %timeit {c['_type']: c['count'] for c in repo_version.content.values('_type').annotate(count=Count('_type'))}
        >20 minutes (cut off)

        %timeit list(Content.objects.filter(pk__in=repo_version.content))
        6.18 s ± 30.4 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)


Round 3 (202,000 pre-existing units):

    100k:  11 minutes 27 seconds
    1k:    0 minutes 22 seconds

        %timeit {c['_type']: c['count'] for c in repo_version.content.values('_type').annotate(count=Count('_type'))}
        >20 minutes (cut off)

        %timeit list(Content.objects.filter(pk__in=repo_version.content))
        10.4 s ± 66.3 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)


Round 4 (303,000 pre-existing units):

    100k:  22 minutes 5 seconds
    1k:    0 minutes 26 seconds

        %timeit {c['_type']: c['count'] for c in repo_version.content.values('_type').annotate(count=Count('_type'))}
        >20 minutes (cut off)

        %timeit list(Content.objects.filter(pk__in=repo_version.content))
        16.4 s ± 360 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)


Round 5 (fresh repo, background 404,000 pre-existing units):

    100k:   <skipped>
    1k:     <skipped>

        %timeit {c['_type']: c['count'] for c in repo_version.content.values('_type').annotate(count=Count('_type'))}
        <skipped>

        %timeit list(Content.objects.filter(pk__in=repo_version.content))
        <skipped>

Final database size: <skipped>