Closed jakhani closed 7 years ago
How should we handle the case where user provided Spark version is not installed? Should we throw certain exceptions?
The existing mechanism will fall back to using the default version. Throwing an exception might be better here to make sure expectations are met. It would be better if the exception message also includes valid version Strings so users know which ones are valid.
@zhe-thoughts & @Victsm Yes throwing an exception would be better in this case. So if provided version is not installed then it won't go by default installation. Do we follow same case for other types of jobs? For e.g. Pig or Hive?
Other job types right now do not use the version string approach, but rather require users to specify the full path of Pig/Hive installation.
Ok then throwing an exception would work and won't cause any inconsistency. I will make that change.
It would be even better if we can infer valid version Strings and include these in the exception message.
There should be a way to add new Spark binary without dependency on providing Spark home for that binary. This change covers a way to follow directory to keep all spark binaries and naming pattern using which we can easily add new binary and don't need to add property for it's path.