Don't resubmit intermediate transfers

[rcarpa]

Motivation

When a multi-hop transfer fails, each of its hops is marked as failed. The issue is that the finisher will then re-schedule each of the hops separately. This can result in some undesired behaviors:

• deleting the rule which triggered the initial multihop transfer will not stop rucio from attempting the intermediate hops. Things will solve by themselves after 3 failed attempts for the first hop (other hops have the same issue, but they usually will fail fast without actually submitting anything). This can be expensive when this are 3 attempts to stage-in from a tape.
• re-scheduling the initial multihop will fail until those 3 attempts are actually executed because of

  rucio/lib/rucio/core/transfer.py

  Line 1554 in 30f2a77

  logger(logging.WARNING, 'Multihop : A request already exists for the transfer between %s and %s. Will cancel all the parent requests', transfer_path[0].src, transfer_path[-1].dst)

Modification

Fail fast the intermediate transfers. It's probably enough to

• return false for intermediate request at:
  

  rucio/lib/rucio/core/request.py

  Line 83 in 74186dd

  def should_retry_request(req, retry_protocol_mismatches):

• check if an replica already exists when trying to create one at
  

  rucio/lib/rucio/core/transfer.py

  Line 1525 in 30f2a77

  add_replicas(rse_id=rws.dest_rse.id,

  
  and set it to "C" state if it exists.