ICS4: Receive, Acknowledge, and Timeout Packet should be structured to indicate already relayed messages

[colin-axner]

Currently, if N relayers relay on the same channel end, N-1 relayers will likely waste fees per packet relayed. This can be even worse if relayers attempt to retry relay attempt. Then it is N-1 * avgRetryAttempts wasted fees per packet

I am planning on adjusting the ibc-go code to indicate well known error codes for already relayed packets. This can allow retry attempts to be dropped. In the SDK, it may eventually be possible to use this mechanism to drop already relayed messages from the mempool

I'd be great if the spec could structure the receive, acknowledge, timeout packet checks in the same way to indicate the exact point at which a packet can be deemed as already relayed.

For acknowledge packet, this might be the place where the next send sequence > packet.sequence && packet commitment doesn't exist

[colin-axner]

I've started work on this on ibc-go, ran into a question.

@cwgoes is it intentional that only one packet can be timed out on an ordered channel? ie if 10 packets are unreceived and one packet times out, none of the other packets can be timed out

[colin-axner]

Here is the ibc-go pr. The only case unaccounted for is already timed out packets on ordered channels (since the channel closes, it is hacky to add an error code for)

[cwgoes]

@cwgoes is it intentional that only one packet can be timed out on an ordered channel? ie if 10 packets are unreceived and one packet times out, none of the other packets can be timed out

Hmm, if a packet times out which was sent after other packets? The issue is that always-increasing expiration times aren't enforced.

[colin-axner]

Hmm, if a packet times out which was sent after other packets?

Or a packet times out which was sent before other packets. My point is that there's an inherent restriction on ordered channels that they must not have per packet revert logic in the timeout handler since only one packet could ever be timed out. It seems to me we should make this more explicit somehow or enable all ordered packets to be potentially timed out