Model.counter_culture_fix_counts does nothing after the initial run

[kkerley]

I've been manually testing the front end of my Rails 6.1 API application after upgrading from 5.2.8 earlier today. All my rspec tests pass but I found a few weird things happening when actually using the site through the front end. As I was testing this stuff, I discovered that one of my counter caches for a has_many through: relationship wasn't working properly because I didn't set it up properly.

I changed the configuration in the appropriate model file and after adding a few new records, the counter cache started updating properly after adding additional new records, but it never took the existing records into account. When I went into the Rails console and ran the appropriate command to manually update the count, I just get this output:

irb(main):007:0> PlayedSong.counter_culture_fix_counts
  Artist Load (15.7ms)  SELECT "artists".id, "artists".id, COUNT("played_songs".id)*1 AS count, MAX("artists".played_songs_count) AS played_songs_count FROM "artists" LEFT JOIN "played_songs" AS played_songs ON "artists".id = played_songs.artist_id GROUP BY "artists"."id" ORDER BY "artists"."id" ASC LIMIT $1  [["LIMIT", 1000]]

  Performance Load (6.7ms)  SELECT "performances".id, "performances".id, COUNT("played_songs".id)*1 AS count, MAX("performances".played_songs_count) AS played_songs_count FROM "performances" LEFT JOIN "played_songs" AS played_songs ON "performances".id = played_songs.performance_id GROUP BY "performances"."id" ORDER BY "performances"."id" ASC LIMIT $1  [["LIMIT", 1000]]

  Performance Load (0.8ms)  SELECT "performances".id, "performances".id, COUNT("played_songs".id)*1 AS count, MAX("performances".played_songs_count) AS played_songs_count FROM "performances" LEFT JOIN "played_songs" AS played_songs ON "performances".id = played_songs.performance_id WHERE "performances"."id" > $1 GROUP BY "performances"."id" ORDER BY "performances"."id" ASC LIMIT $2  [["id", 1000], ["LIMIT", 1000]]

  Song Load (5.9ms)  SELECT "songs".id, "songs".id, COUNT("played_songs".id)*1 AS count, MAX("songs".played_songs_count) AS played_songs_count FROM "songs" LEFT JOIN "played_songs" AS played_songs ON "songs".id = played_songs.song_id GROUP BY "songs"."id" ORDER BY "songs"."id" ASC LIMIT $1  [["LIMIT", 1000]]

=> []

And here's the PlayedSong model's relevant code:

belongs_to :artist, autosave: true
belongs_to :performance, autosave: true
belongs_to :song, autosave: true
counter_culture :artist
counter_culture [:performance, :tour]
counter_culture [:performance, :tour_leg]
counter_culture [:performance]
counter_culture :song

So, using the above, it was the counter cache on both Tour and TourLeg that wasn't populating. Once I fixed those counter_culture lines to use the array format, new performances added started updating the appropriate PlayedSong counter caches expected.

I tried running the counter_culture_fix_counts command for all the models/associations I'm using counter_culture on and it's the same behavior for each. Finishes almost instantly and just has an empty array for output with similar SQL output. If I wipe out my development database and re-import from production, everything works great (I made a migration that contains the all the necessary Model.counter_culture_fix_counts which runs after the counter fields are added to each model).

It says right in the description of the gem that you run these manual functions weekly so I assume it works, but I can't seem to figure out how to get it to work. I've scoured the documentation and the issues so if I'm missing something obvious, I apologize. And the most advanced functionality I'm using with counter_culture counter caches is handling a couple HMT relationships.

Is there something more I need to do to get counter_culture_fix_counts to work after it's already been run on the database initially? I figured new records would allow it to see something was up, but no. Same with adjusting the hierarchy on the HMT relationships.

[magnusvk]

I'm sorry but I cannot provide support here. If you can come up with a failing test case for me I'm happy to take a look at that.