Refactor APIs to receive keys instead of reading them #7954
Assignees
Comments
nventuro
added a commit
that referenced
this issue
Aug 14, 2024
This is a tentative redesign of the key registry contract, which lets us read all 4 keys in a single merkle inclusion proof, while also performing fewer public storage writes. It relies on `PublicMutable` with custom deadlines instead of `SharedMutable` as the key registry must be lax during reads to prevent abuse from accounts performing frequent rotation. This is meant to be a proof of concept and reference for discussion, so there are no tests yet. If we were to adopt this contract it should be relatively straightforward to replace the current getters with `get_current_public_keys`. --- Update: we've now decided to move forward with this design, even if whether we actually do have full key rotation is up for debate. This is a good middle ground in terms of the code being performant and easy to either add or remove rotation in the future. What this PR does is introduce the new registry and switch all old contracts to use it. They'll be using historical mode, i.e. reading keys at a specific block in the past instead of reading the current keys, resulting in no max block number constraints. This is a departure from the current behavior, but is only temporary until we switch over from the old API to the new one (#7953) so that we can then benefit from reduce gate counts (#7954). I've also left the original contract, libraries, etc., unmodified so as to later delete them cleanly instead of making this PR too messy (#7955). --------- Co-authored-by: Jan Beneš <[email protected]>
This was referenced Aug 14, 2024
AztecBot
pushed a commit
to AztecProtocol/aztec-nr
that referenced
this issue
Aug 15, 2024
This is a tentative redesign of the key registry contract, which lets us read all 4 keys in a single merkle inclusion proof, while also performing fewer public storage writes. It relies on `PublicMutable` with custom deadlines instead of `SharedMutable` as the key registry must be lax during reads to prevent abuse from accounts performing frequent rotation. This is meant to be a proof of concept and reference for discussion, so there are no tests yet. If we were to adopt this contract it should be relatively straightforward to replace the current getters with `get_current_public_keys`. --- Update: we've now decided to move forward with this design, even if whether we actually do have full key rotation is up for debate. This is a good middle ground in terms of the code being performant and easy to either add or remove rotation in the future. What this PR does is introduce the new registry and switch all old contracts to use it. They'll be using historical mode, i.e. reading keys at a specific block in the past instead of reading the current keys, resulting in no max block number constraints. This is a departure from the current behavior, but is only temporary until we switch over from the old API to the new one (AztecProtocol/aztec-packages#7953) so that we can then benefit from reduce gate counts (AztecProtocol/aztec-packages#7954). I've also left the original contract, libraries, etc., unmodified so as to later delete them cleanly instead of making this PR too messy (AztecProtocol/aztec-packages#7955). --------- Co-authored-by: Jan Beneš <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Once #7953 is implemented, we'll be reading all keys at once but discarding most results. We should change most APIs so that they can receive keys instead of reading them on demand, as that allows for reuse of the historical read proof, which is quite expensive. A simple example is
BalancesMap::add, which fetches the npk instead of receiving it.Perhaps in the future we might think of some global cache in which we can store expensive results to later reference them, but we'd need to at the very least settle on having key rotation before we explore such a thing.
The text was updated successfully, but these errors were encountered: