Speed up mirror row count queries
We can discard records from before 2020-03-28 as they are almost certainly on free plans. See gitlab-org/gitlab#216252 (comment 334514544) for more details.
Using the same script from before (https://gitlab.com/gitlab-org/gitlab-exporter/-/merge_requests/110)
| Query | Old runtime | New runtime | Old result | New result |
|---|---|---|---|---|
| mirrors_ready_to_sync | 2044ms | 236ms | N/A | N/A |
| mirrors_not_updated_recently | 329ms | 173ms | 9 | 9 |
| mirrors_updated_very_recently | 80ms | 67ms | N/A | N/A |
| mirrors_behind_schedule | 2223ms | 441ms | N/A | N/A |
| mirrors_scheduled_or_started | 105ms | 112ms | N/A | N/A |
| mirrors_scheduled | 66ms | 57ms | N/A | N/A |
| mirrors_started | 42ms | 45ms | N/A | N/A |
We can't compare the results of any query looking at fresh data because they are constantly changing as mirrors are processed, so those columns of the table turned out to not be very helpful
However, this does stop the two slowest queries being so dramatically slow. I should have investigated this further when @ahmadsherif found that before