We should wait until 0.9.0 is release and we can point to the non-snapshot version before merging this.

If we disagree with the change in behavior, this PR gives us a chance to discuss before 0.9.0 is released.

You don't just want to set the config to maintain the old behavior? It seems odd that the output of the dbscan function is based on a config a user could set for a transitive dependency they don't even know is used for dbscan

I really dislike the way the spark configs are global configurations and so modifying a spark config changes the state of the whole cluster.

I'm thinking of adding a setter to graphframes 0.9.0 so I can set this value on the connectedComponents instance without changing spark configs.

Then the output schema of DBSCAN is not based on the user's setting of the graphframe spark config (one way or the other).

Yeah it's definitely not the cleanest thing, and I was thinking the same thing a setting on the connected components object would be much cleaner