PHOENIX-4855 Continue to write base table column metadata when creati…

[twdsilva]

…ng a view in order to support rollback

Previously in MetadataEndpointImpl we were removing parent column metadata while creating a child view. We still need to write the parent columns in order to be able to rollback if needed. This PR filters out the parent columns while creating the PTable of a child view. When a view is resolved we also resolve all its parents and add their columns (see MetadataEndpointImpl.addDerivedColumnsFromAncestors).
I added some more tests for salted tables. I also changed PhoenixDatabaseMetaData.getPrimaryKeys() to load the PTable and return the metadata instead of directly querying the SYSTEM.CATALOG table (which will not work any more since we don't store columns inherited from parents in a view)

@ankitsinghal @ChinmaySKulkarni can you please review?

[twdsilva]

ping @ChinmaySKulkarni @karanmehta93 any chance I can get this reviewed soon?

[ChinmaySKulkarni]

@twdsilva Had a quick glance over the PR. I had a few questions. So in the case a view is created on top of a salted base table, we explicitly filter out the salting column (first column) when resolving the columns for a view, right? Also, in case any ancestor of a view drops a column, that will be part of the excluded columns list and we filter out those columns as well, when resolving a view? What happens if a view diverges from the base table, the base table drops a column and then the diverged view re-adds this column?

Overall, can you please add more comments to make it easier to understand the steps taken when handling excluded columns? I will go over the PR in detail and let you know some more feedback. Thanks

[ChinmaySKulkarni]

A few comments. Please also address the test failures inside ViewIT.java. Otherwise, lgtm.

[twdsilva]

@ChinmaySKulkarni Thanks for the review, I have updated the PR. I will commit this to 4.x and master branches.