Right choice is (a) chukwa supports two different reliability strategies
To explain I would say: The first, default strategy, is as follows: collectors write data to HDFS, and as soon as the HDFS write call returns success, report success to the agent, which advances its checkpoint state.