Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
None
Description
A successful attempt should fail only when its outputs are reported as failed. We could also pro-actively do that for nodes that YARN marks as lost/decommissioned but that technically needs to consider if those outputs are really needed (job may have progresses far ahead) and also on whether the output class actually stored the outputs on that node.