Using older hadoop (for example hadoop-2.7 which is included in the spark distributions) is not possible any more even with using the isolated class loader (introduced in
Fortunately it can be fixed:
- We can support null in storageStatistics field with checking everywhere before call it.
- We can create a new constructor of OzoneClientAdapterImpl without using OzoneFSStorageStatistics): If OzoneFSStorageStatistics is not in the method/constructor signature we don't need to load it.
- We can check the availability of
HADOOP-13065and if the classes are not in the classpath we can skip the initialization of the OzoneFSStorageStatistics