Guesstimate partition boundaries from extrema rather than using expensive offsets

Using order by and offset makes the partitioning a n^2 operation on the number of entries in the table, rather than an instant operation when using min/max.

This assumes the indexable column is more or less uniform, which is not exactly true but not the worst approximation either.

Test Plan

tox

---

Migrated from D1267 (view on Phabricator)

Build is green See https://jenkins.softwareheritage.org/job/DLS/job/tox/64/ for more details.

Should we be worried that tests are green after this diff?

Merge request was accepted

approved this merge request

! In !57 (closed), @douardda wrote: Should we be worried that tests are green after this diff?

We only test that the partitioning works (i.e. that we can generate partitions) and, I think, that all partitions yield all the entries.

In any case, one of the bitbucket lister tests was red before I did the ugly ugly string -> datetime bodge.

mentioned in commit dd148fc6

Some references in the commit message have been migrated:

• D1267 is now !57 (closed)

mentioned in commit c574897e

Merge request was merged

closed