You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When a multihop transfers is generated, intermediate transfers are scheduled without enforcing any source replica. If the transfer fails
during submission, and the intermediate transfer remain in "queued" state in the database, the intermediate transfers will be later be submitted on their own, but they don't have any source replica expression set, so they will end using whatever source is judged best.
An example of multihop which resulted in this case :
conveyor-submitter[67/181]: e6c76217fb284b5baf5c9795cc58654e: Cannot pick transfertool, or create intermediate requests
conveyor-submitter[67/181]: e6c76217fb284b5baf5c9795cc58654e: Multihop : A request already exists for the transfer between src_rse=CERN-PROD_RAW and dst_rse=BNL-OSG2_DATADISK. Will cancel all the parent requests
conveyor-submitter[67/181]: e6c76217fb284b5baf5c9795cc58654e: Best path is multihop: CERN-PROD_RAW--8f1682d7464e4404b3074384426461da->CERN-PROD_DATADISK--e6c76217fb284b5baf5c9795cc58654e->BNL-OSG2_DATADISK
conveyor-submitter[67/181]: e6c76217fb284b5baf5c9795cc58654e(data17_13TeV:data17_13TeV.00324781.physics_Main.daq.RAW._lb0229._SFO-8._0003.data): Ordered sources: multihop: CERN-PROD_RAW:0:52
conveyor-submitter[67/181]: e6c76217fb284b5baf5c9795cc58654e: 1/2 sources left after filtering. Dropped: IN2P3-CC_DATATAPE
conveyor-submitter[67/181]: e6c76217fb284b5baf5c9795cc58654e(data17_13TeV:data17_13TeV.00324781.physics_Main.daq.RAW._lb0229._SFO-8._0003.data): Found 2 sources
and the associated intermediate replicas living their own lives (which shouldn't normally happen) :
The fact that transfers remain in queued state in the database is not normal, they should be cleaned up correctly, but it's not the first time we have issues with this cleanup: there are too many things which can go wrong. To reduce impact of such unexpected corner cases, it may be worth enforcing source_replica_expression=rse_name for all intermediate hops.
The text was updated successfully, but these errors were encountered:
bari12
changed the title
Transfers: source replica expression not always correctly respected on multihops
source replica expression not always correctly respected on multihops
Feb 1, 2022
piperov
pushed a commit
to piperov/rucio
that referenced
this issue
Feb 25, 2022
Motivation
When a multihop transfers is generated, intermediate transfers are scheduled without enforcing any source replica. If the transfer fails
during submission, and the intermediate transfer remain in "queued" state in the database, the intermediate transfers will be later be submitted on their own, but they don't have any source replica expression set, so they will end using whatever source is judged best.
An example of multihop which resulted in this case :
and the associated intermediate replicas living their own lives (which shouldn't normally happen) :
Modification
The fact that transfers remain in queued state in the database is not normal, they should be cleaned up correctly, but it's not the first time we have issues with this cleanup: there are too many things which can go wrong. To reduce impact of such unexpected corner cases, it may be worth enforcing source_replica_expression=rse_name for all intermediate hops.
The text was updated successfully, but these errors were encountered: