Prevent sync lookups from reverting to awaiting block

[dapplion]

Issue Addressed

Sync lookups have an "oracle" into the da_checker state. Blob or custody column requests are only progressed when the block is already downloaded. This allows deterministic requests after we are sure the block has data.

However, the current implementation has an interesting flaw. If the "oracle" is buggy, the sync lookup can consider that:

• t_0: block is not downloaded, halt custody column downloads
• t_1: block is downloaded, continue column downloads
• t_2: block is not downloaded, halt custody column downloads (again)

This sequence of states should never happen but the type models allow it to happen.

Proposed Changes

Instead, we can model SingleBlockLookup such that once the block is seen, it can never be unseen. This new enum:

enum ComponentRequests<E: EthSpec> {
    WaitingForBlock,
    ActiveBlobRequest(BlobRequestState<E>, usize /* expected blob count */),
    ActiveCustodyRequest(CustodyRequestState<E>),
    NotNeeded(&'static str),
}

and moving the "oracle" logic from the network context to the lookup logic achieve this effect.

I'm leaving the PR in draft for those familiar with the topic to thumbs up this design. The overall goal is to reduce the chance of lookups getting stuck if we change the da_checker logic.

[jimmygchen]

This looks great to me! I think moving the logic up to SingleBlockLookup is a lot cleaner, and avoids the possibility of the scenario you described.

[jimmygchen]

@mergify queue

[mergify]

queue

✅ The pull request has been merged automatically

The pull request has been merged automatically at a0a62ea