Fetching by a cached index produces two queries

While debugging N+1 queries, I found an unusual behaviour in IDC: when fetching records using a cached attribute, IDC produces two queries instead of one. The first query uses the cached attribute, while the second uses the primary key.

After some investigation and some very helpful tips from coworkers, I managed to boil the issue down to this snippet:

# migration file
class CreateSampleRecords < ActiveRecord::Migration[5.2]
  mark_as_podded_migration!

  def up
    create_table :sample_records do |t|
      t.string :key
      t.timestamps
    end
  end

  def down
    drop_table :sample_records
  end
end

# ActiveRecord model
class SampleRecord < ApplicationRecord
  include IdentityCache
  cache_index :key
end

Then, on rails console:

[1] pry(main)> SampleRecord.create(key: '123')
   (1.6ms)  BEGIN
  SampleRecord Create (1.0ms)  INSERT INTO `sample_records` (`key`) VALUES ('123')
   (2.3ms)  COMMIT
=> #<SampleRecord:0x00007fa11061a158 id: 1, key: "123">
[2] pry(main)> SampleRecord.fetch_by_key('123')
   (1.4ms)  SELECT `sample_records`.`id` FROM `sample_records` WHERE `sample_records`.`key` = '123'
  SampleRecord Load (0.9ms)  SELECT `sample_records`.* FROM `sample_records` WHERE `sample_records`.`id` = 2
=> [#<SampleRecord:0x00007fa0f5169a70 id: 1, key: "123">]

I also tried with unique: true, it produces the same result.

I might be wrong, but I would expect only one query to be produced by this snippet. The first query already fetches the desired record, I see no reason to search for the object again using the primary key.

Thoughts?

Reproduce the behavior in a Rails console with the SampleRecord model, cache_index :key, and SampleRecord.fetch_by_key('123'). Then read the fetch_by_key and cache_index entry points to determine why the primary-key lookup follows the cached-index query. Done means the reproduction performs one query and has regression coverage for the cached-index fetch.