Fix intermittent failure on add_parent test
Closes #1511 (closed)
I cannot reproduce this at all locally, so it took me an hour to reproduce it in CI with some extra print()
. I tested various methods of ordering the parent paths or making the query more deterministic, and with a bunch of EXPLAIN
I found that adding ORDER BY path
(not id
or element_id
or ordering
or something else) caused the query plan to not change one bit, likely because the index is already internally ordered by path
, so this should not waste the performance boosts that we got with the rewrite of add_parent
.
@vrigal Since you seem to be able to reproduce more easily, can you check whether or not the failure seems to have been fixed on this branch?