[chain filters] should not be using inactive fields for joins

[piranha]

Resolves #39618

[replay-io]

Status	Complete ↗︎
Commit	7cfe353
Results	⚠️ 4 Flaky ✅ 2337 Passed

[crisptrutski]

Makes sense to me. A pity about the manual reset, but doesn't seem worth rabbit holing for a small bug fix.

[calherries]

This PR seems to be fixing a symptom of the issue, not the root cause.

The fact that a field referenced by fk_target_field_id is inactive is real issue here. That state should be impossible. I thought I fixed that in Remove a retired field's FK target references during sync. It's possible that there was some edge case that I missed. If so we'll need to do another migration like this.

This fix is unfortunately too narrow because it fixes one symptom of the bug but it means there's probably other ways this becomes an issue out there. Having to check that a field is active every time we use fk_target_field_id is error prone and complicates the code everywhere we use this information.

[piranha]

@calherries hmm... I'm not understanding semantics completely, but is a field being inactive an error at all times?

[calherries]

No, but a foreign key reference to an inactive field is definitely a bug, because the constraint doesn't exist in the database. This is a special case of the bug I raised yesterday #39687 (total coincidence though)

[calherries]

Opening this again because I feel we should just land this rather than talking about solutions more.

[calherries]

Looks good

[github-actions]

@piranha Did you forget to add a milestone to the issue for this PR? When and where should I add a milestone?

[piranha]

@metabase-bot backport release-x.48.x