In this ticket we need to rethink how we're currently doing the data model and try to remove any knowledge of keys in memory except during verification. For workloads that require knowledge of keys, they should search for the key.
To get a random key there are two possibilities:
- We could use a random cursor here
- Or given our current key generation model we should know what the min and max key is and can simply use a random linear distribution and generate a valid key. (this could be faster than random cursors?)