Error on serialization/deserialization of total time

[wandering-tales]

I have stumbled upon this weird error lately affecting the adjust_total_time method in the Iteration module:

def adjust_total_time
  self.total_time += (Time.now.utc.to_f - start_time.to_f).round(6)
end

Sometimes this addition fails with the following error:

TypeError: no implicit conversion of Float into String

We use Resque as queuing backend for ActiveJob and, browsing the Failures section of its web interface, I found out that the total_time job argument was sometimes saved as a string instead of a float (I'll omit other job arguments with potentially sensitive information):

# good

job_class: ********
job_id: 87dcc377-1b18-4f2c-ba58-f2645b6b83a7
provider_job_id:
queue_name: default
priority:
arguments:
- ...
  _aj_ruby2_keywords:
  - ...
executions: 1
exception_executions: {}
locale: en
timezone: UTC
enqueued_at: '2023-06-13T14:19:09Z'
cursor_position:
- 2
- f502bdad-2efd-4d71-89e0-d2077e59b59e
times_interrupted: 4
total_time: 3605.106918

# bad

job_class: ********
job_id: 15a254eb-218b-4c7d-8669-00c6e1342e75
provider_job_id:
queue_name: default
priority:
arguments:
- ...
  _aj_ruby2_keywords:
  - ...
executions: 1
exception_executions: {}
locale: en
timezone: UTC
enqueued_at: '2023-06-13T19:20:17Z'
cursor_position:
- 19
-
times_interrupted: 8
total_time: '7205.1291820000005'

My guess is a problem occurring either during the serialization or deserialization of job-iteration extra arguments, but I cannot figure out which.