Thread by @krzKaczor

I just saw A LOT of excitement about fast confirmations on newly deployed @Uniswap on @optimismPBC , but how does it really work, and can users trust them? Finally, isn't a single sequencer that provides such confirmations a threat to decentralization? Let's talk it through👇👇👇

2/ First, things first — sequencers are privileged actors in many rollup systems (@optimismPBC , @arbitrum, @StarkWareLtd , @zksync ). They receive transactions from users, order them and submit them in batches to L1.

3/ The main reason why they exist is the simplicity and efficiency of having a single coordinator. At this stage, there is usually a single sequencer per rollup, and it's run by the creators of a rollup.

4/ A sequencer is a single actor responsible for ordering txs. Thanks to this, after receiving a user's tx, a sequencer can instantly mine it and send an instant confirmation back to the user. This dramatically improves UX.

5/ If you're worried that sequencers can do arbitrary MEV then you're absolutely right but I leave this topic to a separate discussion (we are gonna touch this at the end)... Don't tell @EdgarArout 😏

6/ It's all great when a "good guy" sequencer is an honest actor but what if the sequencer goes rogue, lying to the users and trying to break the network? Let's dive deeper🤿

7/ Most important question: can sequencer steal users' funds? Short answer: NO. Validity of state transition is ensured by a rollup architecture (fraud proofs in optimistic rollups or validity proofs in zkrollups).

8/ Can it censor user txs? - Yes, actually it can. A sequencer is usually a JSON RPC node. Similar to Infura, it can even lie about the state of the network or censor users tx.

9/ Luckily censorship is not a huge problem as all rollup systems have a way to post L2 tx via uncensorable L1. The protocol then forces the sequencer to include the user's tx into a rollup within few minutes.

10/ If sequencer lies about the state, users would need to run their own nodes recreating the rollup state based on batches posted to L1. This might sound bad but it's actually the same situation as on L1 🤷‍♂️

11/ Finally, can sequencer lie about instant confirmation? Yes, as mentioned before, the sequencer can lie about the current network state and about the inclusion of the user's tx.

12/ For example, sequencer could reply back to the user that their trade was successful but in a reality it reverted (or the other way around). The user would only realize this after recreating the rollup state from L1.

13/ A rollup tx is not finalized until it's posted on L1. That's why web3 libs for rollups usually allow devs to easily build UIs so they wait until tx is mined on L1.

14/ Future solutions to this problem can involve sequencer signing a confirmation of receiving user's tx. If tx wouldn't end up in a rollup user could slash the sequencer. This could be automated by watchtower-like services.

15/ You see this is what's really excites me about this -- it's only the beginning for sequencer tech. In the near future, we will see more sophisticated designs that solve a lot of problems that I've just mentioned.

16/ Instead of running a single permissioned sequencer, we can run a whole permissionless PoS network of sequencers. Each L1 batch would be created by a randomized sequencer from a bigger set. This would make resilience and censorship resistance much better.

17/ Of course, each sequencer would need to post some kind of bond so it can be slashed when it misbehaves.

18/ Other projects like @arbitrum, experiment with a fair protocol that tries to discover the “true” ordering of txs. eprint.iacr.org/2020/269.pdf

h/t @sgoldfed

19/ Instead of fighting MEV, it can be embraced by MEV auctions where parties bid against each other to be able to run a sequencer for some time (but this idea has some problems).

20/ You can read a little bit more about design space for rollups in @VitalikButerin's article on rollups: vitalik.ca/general/2021/0… ("Who can submit a batch?" section)

21/ To sum up, IMO sequencers strike the right balance between decentralization and speed. What we see now is basically just an MVP of sequencer tech and a lot of smart people work hard on improving it 💡 Future looks bright!