On 16 May 2015, at 04:39, Anton Brazhnyk <[email protected]<mailto:[email protected]>> wrote:
For me it wouldn’t help I guess, because those newer classes would still be loaded by different classloader. What did work for me with 1.3.1 – removing of those classes from Spark’s jar completely, so they get loaded from external Guava (the version I prefer) and by the classloader I expect. Note that Hadoop <= 2.6.0 wont' work with Guava >= 0.17; see: HADOOP-11032 FWIW Guava is a version nightmare across the hadoop stack; almost as bad as protobuf.jar. With Hadoop 2.7+, Hadoop will run on later versions, it'll just continue to ship an older one to avoid breaking apps that expect it.
