Oops, I've forgotten to tell I'm using the 0.7-rc2 branch with some patches that has nothing to do with hadoop.
When I start the CFInputFormat to read a CF in a keyspace of RF=3 on a 4-node cluster:
- If all the nodes are all up, everything works fine and I don't have any problems walking through the all data in the CF, however
- If there's a node down, the hadoop job does not even start, just dies without any errors or exceptions.
Is this a bug or feature or a misuse?
I'm just using the same directives in the sample code.