Hello, We are experiencing this old issue in our current installation:
https://issues.apache.org/jira/browse/MAPREDUCE-5537 All our data is LZO compressed and indexed; the case is 100% reproducible on our CDH 5.2.0 cluster (using MR2 and Yarn). Do you know if we might be missing a patch or if maybe this particular problem found a way back into the code? Best regards, Nathalie Blais B.I. Developer - Ubisoft Montreal