now that the owner of the building is implemented, there is already one task instance for each record.
first of all, let's make two definitions:

1. task, is the celery method you define, for example:

@celery.task
def celery_task():
  pass

1. task instance, which is the actual task to be run

task_instance = celery_task.delay()

Task 1, query; Task 2, traversal and update.
so the design of the landlord himself is:
two task, million (if enough data) task instances (that is, one task has been created for each eligible data).

since it is not very convenient to answer the landlord's question in the comments, I will answer the questions in the comments here.
Plan 1:
increase the number of consumers of celery and increase the number of worker.
is not recommended, because there are many uncontrollable factors and may not achieve the desired results.
Plan 2 (personal suggestion, which can be modified according to the situation):
add the judgment flag bit on your own.
does not know how the landlord uses celery, so it is assumed that the publish and subscribe tasks completed through redis have been operated.

-sharp 
@celery.task
def query_from_db():
  results = db.query
  for result in results:
    if redis.get(result.id):
      continue
    -sharp 
    -sharp updatequery_from_db
    -sharp updateredisresult.id
    redis.set(result.id, 'something', two_hours)
    update_result.delay(result.id)
    
@celery.task
def update_result(result_id):
  result = db.query.get(result_id)
  rv = requests.get(.....)
  result.update(rv.json())
  redis.delete(result_id)