in my previous life I was part of data modeling team. For hive I used Embaracadero data architect for hive physical data modeling. your logical model should not be impacted by the physical implementation. This is what the gods of LDM tell you. And I tend to agree. You should continue to do version control of your LDM just like you do today as again, the LDM has no baring on PDM. Now version control PDM is similar to how you do on RDMBS. Store all your DDLs in sequential order or invest in source control. Again no different then what you do today fro RDBMS. I believe from a modeling perspective the challenge is how to apply modeling principles to hive. using 3NF is not the right approach. Using the kimball approach (data marts) apply in this space. Those that do not want to invest time into thinking through a data model will say to hyper denormalize in hive. That means they generally don't have a clue about what a data model is and the importance of one in a enterprise. With hive LLAP & tool such at AtScale allow us to model in a more "native" format. Important: for nosql data models, it is best to use domain driven design practice. This is the closest one will get to a PDM. nosql does not follow relational theory, and hence modeling such using PDM core values is a nugatory exercise. Now Apache Phoenix which is a sql skin on top of hbase changes that. You can apply some relational theory PDM rules there.
... View more
