[Feature]: Add an endpoint to check for nullifiers given block bounds and prefixes

[tomyrd]

If we were to have an endpoint to validate whether a list of notes have been consumed / nullifiers are present in the nullifier SMT, we could add a single step after both the normal sync and the ignored notes sync where we see if any of those got consumed.

We do have a CheckNullifiers endpoint in the node - thought, it requires sending full nullifiers and returns back authentication paths for the nullifiers found in the node. So, this is a bit more heaving then needed and also reduced privacy. Maybe it is worth adding an endpoint to the node to return a list of nullifiers based on some prefix and block bounds.

Originally posted by @bobbinth in 0xMiden/miden-client#405 (comment)

[bobbinth]

One option is to modify the current CheckNullifiers endpoint to return only the found nullifiers and make returning of the authentication path optional. The request could look something like this:

message CheckNullifiersRequest {
    // List of nullifiers to check.
    repeated digest.Digest nullifiers = 1;
    // Whether or not to return authentication paths for each found nullifier.
    optional bool include_proof = 2;
}

The main concern with this approach is that the user would tell the network exactly which nullifiers they are interested in.

Another option is to introduce a new endpoint - e.g., something like CheckNullifiersByPrefix. The request for this endpoint could look like so:

message CheckNullifiersByPrefixRequest {
    // List of nullifiers to check. Each nullifier is specified by its 16-bit prefix.
    repeated uint32 nullifiers = 1;
    // Block number from which to start checking nullifiers.
    fixed32 start_block_num = 2;
}

The endpoint would return nullifiers which match any of the specified prefixes and were created after request.start_block_num. The response could look simply as:

message CheckNullifiersByPrefixResponse {
    // List of nullifiers created between `request.start_block_num` and the last block in the chain.
    repeated NullifierUpdate nullifiers = 1;
}

The main concern with this endpoint is that it may return too much data (especially when the network operates at high TPS). For example, if 10K notes are consumed per second (using this as a proxy for 10K TPS), a request covering 1 week would return about 3MB per nullifier prefix.

[Mirko-von-Leipzig]

The main concern with this endpoint is that it may return too much data (especially when the network operates at high TPS). For example, if 10K notes are consumed per second (using this as a proxy for 10K TPS), a request covering 1 week would return about 3MB per nullifier prefix.

One approach to handle this is pagination and rate-limiting. Both request and response include an optional pagination token. In the simplest implementation its just an integer indicating how many records were already processed in previous requests. However this probably won't work well in this instance as for subsequent requests nullifiers have changed status.

We could instead have the token represent the nullifiers already processed + the block number. The next request would then recheck that range nullifiers for any new blocks/nullifiers (presumably with very little new data), and then fill up any remaining response space with more nullifiers.

Another issue with pagination is that a user might stop early once they've received the nullifier(s) they're actually interested in. So that leaks more info to the network. Additionally, users now have to aggregate responses i.e. nullifiers can change status over subsequent calls, and so will proofs maybe? I don't know enough about how the latter is implemented.

Quite a bit more complicated than a once-off query though. Much simpler to instead throw an error if the caller requested too much data.

[bobbinth]

However this probably won't work well in this instance as for subsequent requests nullifiers have changed status.

I think this depends on how we decide to paginate. For example, if we do it based on block numbers, and return nullifiers that were created between block x and y, the request to get nullifiers between blocks y and z would return any new nullifiers which might have been created after the request for (x, y) range was served.

One approach I was thinking about more generally is to support "natural" pagination based on "epochs" (not just for this endpoint, but for other endpoints as well, and this endpoint may be a good one to think through the general idea). This would work like so:

First, we define an epoch as $2^{16}$ blocks. Basically, the 16 most significant bits of block number define its epoch. Depending on where we end up with block times, this will be between 2 and 6 days per epoch.

Then, we change the response message to look like so:

message CheckNullifiersByPrefixResponse {
    // List of nullifiers created between `request.start_block_num` and the last block of the epoch
    // defined by `request.start_block_num`.
    repeated NullifierUpdate nullifiers = 1;
    // Number of the latest block in the chain.
    fixed32 chain_tip = 3;
}

The core idea here is that the response will contain data for nullifiers created between request.start_block_num and last_block_of_epoch(request.start_block_num / 2^16). If there is no relevant data in this range, the response will come back empty.

The user, then, will keep making requests setting start_block_num to the start of the next epoch until they get to the chain tip. The pseudo-code for this could look something like this:

let mut start_block_num = <initial value for the first request>
let nullifiers = vec![<list of nullifier prefixes>];

loop {
    let response = check_nullifiers_by_prefix(nullifiers, start_block_num);
    todo!("process the response");

    let epoch = epoch_of(start_block_num);
    if epoch == epoch_of(response.chain_tip) {
        // if we got to the end of the chain, break out of the loop
        break;
    } else {
        block_num = first_block_of(epoch + 1);
    }   
}

fn epoch_of(block_num: u32) -> u16 {
    (block_num >> 16) as u16
}

fn first_block_of(epoch: u16) -> u32 {
    (epoch as u32) << 16
}

The main benefit of this approach is that it should be pretty simple to implement. It also naturally supports sharding of the underlying data by epoch (this is not something we need now, but I anticipate that we probably will need in the future).

The main drawback is that we don't fix the amount of data per response: some may be empty while others may still contain quite a bit of data. However, for this specific endpoint, I don't think this will be an issue as nullifiers should be randomly distributed and thus, variability across responses should be relatively low.

We can also implement the approach suggested in #174 and stream all messages to the user to eliminate the need for many round trips (basically, the node would run the loop internally and stream messages to the user until chain tip is reached).

[bobbinth]

Based on some more thinking described in 0xMiden/miden-client#405 (comment), it seems like we may not need a time-bounded endpoint. Instead, checking nullifiers "across all time" may be a better option - but we would do it by prefix rather than for a specific nullifier.

So, CheckNullifierByPrefix endpoint could look something like this:

message CheckNullifiersByPrefixRequest {
    // List of nullifiers to check. Each nullifier is specified by its 16-bit prefix.
    repeated uint32 nullifiers = 1;
}

message CheckNullifiersByPrefixResponse {
    // List of nullifiers matching the 16-bit prefixes specified in the request.
    repeated NullifierUpdate nullifiers = 1;
}

We could put a pretty aggressive limit on the number of nullifier prefixes that can be submitted with the request to avoid dealing with pagination (we could even allow just a single prefix per request to simplify things). The response sizes should be manageable until we get to about 100 TPS, and after that we can bump up the prefix size to 20 or 24 bits.

[bobbinth]

One other thought on this: maybe the request should look like:

message CheckNullifiersByPrefixRequest {
    // Number of bits used for nullifier prefix.
    unit32 prefix_len = 1;
    // List of nullifiers to check. Each nullifier is specified by its prefix with length equal
    // to prefix_len
    repeated uint32 nullifiers = 2;
}

Right now, the only valid value for prefix_len would be 16, but this way, we could update this in the future without introducing breaking changes.

[tomyrd]

How should the prefix_len be used inside the database query? Currently only 16-bit prefixes are stored.