One issue with the current structures is that they allow partitioning and indexing to be back-defined in the past. This is proper for when systems do not have data but it will fail in production.
I propose two pieces to this:
The settings should also be transmitted out to the cluster management piece and then on to the data nodes
Our tooling should refuse to start if the production setting is not set to 1
If the setting is missing or set to 0, only then can we back-date partitioning.
By default indexes should have the same logic but perhaps after 1.0 we should explore ways to backfill indexes.
One issue with the current structures is that they allow partitioning and indexing to be back-defined in the past. This is proper for when systems do not have data but it will fail in production.
I propose two pieces to this:
production
setting is not set to1