[Feature Request][design]: DA error handling

[tuxcanfly]

Implementation ideas

This is a design discussion to coordinate error handling between celestia core/app and rollup clients like rollkit.

Some issues with the current model

• core/app accepts transactions which may not be mined see: mempool transaction validation doesn't account for padding shares celestiaorg/celestia-app#3081
• tx which are not mined return timed out waiting for tx to be included in block
  • should this be handled by core/node? see: [Feature Request][da]: gracefully handle submit mempool errors rollkit#1522
  • should core/node autoretry with higher fee
• if a tx is accepted by the mempool (e.g account being reused), rollup cannot do anything until it gets dropped
  • can core allow a faster ttl? or allow replace-by-fee?
• testing DA submission issues is dependent on mempool state (e.g. can be congested) which makes it hard to reproduce issues like Rollup stops posting to DA layer after a while rollkit#1520

Overall this makes handling DA submission challenging because of the lack of congest control flow and lack of dynamic fee market.

[zmanian]

These issues seem broadly related the general "comet BFT mempool does not accommodate the application specific requirements" issue.

I surfaced these issues to the comet BFT mempool working group.

1. The mempool doesn't support "contracts" like replace by fee. It's just a dumb queue.

2. Ante handlers are ignorant of "whole block" construction factors and can get stuck in head of the line blocking issues where a tx can't be proposed but also blocks the queue. Block construction needs be able to generate a more complex view into the mempool than ReapMaxBytes allows.

Binary Builders has indicated they are building capabilities into server v2 that would enable these kinds of features but would be good to align on an overall technical approach.

[zmanian]

Relevant

cosmos/cosmos-sdk#19329