Problem
Currently when a binding is deleted, we don't delete checkpoints and therefore if you re-add the binding, the materialization re-starts from where it left off. We do delete the actual underlying data (like the table) most of the time though, so this leads to only partial data being re-materialized.
Proposal
We should very carefully delete checkpoints when bindings are deleted. I say very carefully, cause removing checkpoints is extremely dangerous if we do it incorrectly.
Problem Currently when a binding is deleted, we don't delete checkpoints and therefore if you re-add the binding, the materialization re-starts from where it left off. We do delete the actual underlying data (like the table) most of the time though, so this leads to only partial data being re-materialized.
Proposal We should very carefully delete checkpoints when bindings are deleted. I say very carefully, cause removing checkpoints is extremely dangerous if we do it incorrectly.