Open DanilaOleynik opened 6 years ago
Yoda should not be hard synchronized with Harvester or any other edge service. All processed events should be declared for stage-out. Yoda should not keep meta info about processed events in the internal queue somehow long.
A test job was submitted with 3 nodes(1Yoda+2Droids, 16 cores per node, walltime 30 minutes), Yoda loaded 20000 events. 2 Droids processed 189 events but But the final event_status.dump.json has only 12 events