bug: Flashloans: wrong solver address saved to database

[sunce86]

Problem

The example order having flashloan was executed on Sepolia: https://explorer.cow.fi/sepolia/orders/0xe5e140e989a8ab8a66a8fe70d3dbd1095935d6a9355a21752c7f7a50d9e3e49b35ed9a9d1122a1544e031cc92fcc7ea599e28d9c67dc304c

Backend expects that proper solver (submission address) is saved to the database table settlements::solver. Instead, when order contains a flashloan, flashloan router address is saved instead of solver submission address.

The reason for this is settlement event containing flashloan router address as solver.

One function that depends on this is find_settlement_transaction: https://github.com/cowprotocol/services/blob/main/crates/database/src/settlements.rs#L6-L22
Edit: another dependency: https://github.com/cowprotocol/services/blob/main/crates/database/src/solver_competition.rs#L202

There are at least two ways to handle this:

1. Order onchain events before inserting into database (specifically, make sure flashloan event comes after the settlement event), then when handling flashloan event, take solver address and overwrite settlement event entry that is already processed. This requires flashloan event to contain tx hash and solver address(only solver address, since tx_hash is retrievable for every onchain event), because we could have more than one settlement in the same block.

2. Create a new database table to store flashloan events, then the backend code has to be aware that this indirection is not automatic and needs to handle it properly.

I don't have a strong opinion which solution is better. (1) makes settlement::solver easier to use across other sql queries, however it does introduce a difference between onchain event and what is stored in the database, which may or may not turn out as a bad decision in the future.

[MartinquaXD]

Another approach would be to analyze the callstack of the settle() call to determine which address is actually a solver. We could start from tx.from and go down to the settle() call. For each caller address we check authenticator.isSolver(caller). The actual solver that triggered the transaction will be the first authenticated address in that chain. If we think about blame it was either this solver that initiated a tx or this solver who that got exploited. In both cases it seems reasonable to punish or reward that address.
IMO this is the nicest solution because it is a very local change and does not depend on us indexing any new events or think about ordering of said events.
The found address would then be used to overwrite the settlements::solver column when the Observer indexes a settlement transaction.