Changed the handler returns to OrderInfoResponse & Added Signed Effects

[gdanezis]

• Now the order and certificate handlers in authority return an OrderInfoResponse struct, containing the signed order, certificate and signed effects.
• A structure SignedOrderEffects binds the TransactionDigest with the mutated and deleted ObjectRefs. Since an ObjectRef contains the ObjectDigest a quorum of these could be used for light clients.
• We also store the SignedOrderEffects to return them if a certificate is submitted again.
• Ensure the handlers are idempotent.
• Review the error return conditions from handlers.

Future PR / out of scope:

• Review any additional info to return to help clients sync other lagging authorities (such as the parent TxDigests of Objects needed?).

Perf impact: bench 9100 tx/s -> 8500 tx/s.

[sblackshear]

I think gas used and status code would also make sense here (more details in #120)

[gdanezis]

Ok, I am now including a Result<(), FastPayError> returned by move_call or module_publish.

[sblackshear]

Somewhat tangential, but I'm wondering how signed_orders and signed_effects work w.r.t state sync. The other components of AuthorityStore should be the same on every authority, but signed_orders and signed_effects are a bit odd in that they're inherently specific to the single authority providing the signature. So specifically:

• Does a full node need to pick a particular authority to sync all of its signed_orders and signed_effects from? Or can it not serve queries that require reading from these stores?
• Same question, but for a new authority joining the network

[gdanezis]

I think that this depend on the degree of re-validation an authority that syncs wishes to perform.

• At the extreme end it wants to re-validate everything: in this case the only set it syncs are the certificates, and then re-executes them all, therefore deriving the effects and objects on the way.
• Somewhere in the middle and authority wants to trust a quorum but not re-execute. In this case it downloads the effects from f+1 authorities, and if they agree it just applies them. This ensures that one authority at least was honest.
• At the lower end of assurance an authority can just pick a random set of other authorities (sample 3-5 by stake) and apply the effects that all agree on.
• Then finally an authority can fully trust another authority: and just copy the effects.

Thinking about it: is there a reason to sync the full history of effects? I guess only to serve authenticated reads on the events down the line when we have them?

[gdanezis]

PS: SignedOrder may be different between authorities, but SignedEffects should not be (if processing is determnistic), besides the signature being from a different authority.

[sblackshear]

I guess only to serve authenticated reads on the events down the line when we have them?

Yes, authenticated reads of both events and other order effects (e.g., "did this order create object O"?)

PS: SignedOrder may be different between authorities, but SignedEffects should not be (if processing is determnistic), besides the signature being from a different authority.

Ah, interesting... Is SignedOrder only different (besides name/signature) between authorities when a client equivocates?

[huitseeker]

Without at minimum arguing that some actors in the system (which may be authorities, or "fishermen" incentivized for this explicit purpose) download both certificates & effects, re-validate everything, and are able to submit a fraud proof when they disagree with the effects they receive, can we argue that the execution of orders in FastNFT has any sort of computational integrity?

Note: the standard I'm suggesting above is the weakest available, that of optimistic rollups. I don't suggest we adopt it.

[oxade]

@huitseeker Any feedback on this PR?
Hoping to merge it soon to unblock some client changes for multi-object orders and Move calls.

[huitseeker]

The code LGTM. The question of SignedOrderEffects being part of state sync deserves a separate discussion, but as Sam mentioned, it's tangential to the point of this PR.

[gdanezis]

Ok merging!