[General] Utilizing Decentralized Sequencers

[khalidzahra]

Team or Project

No response

ZK chain

Era

Environment

Testnet

L2 block number

No response

Provide a brief description of the functionality you're trying to implement and the issue you are running into.

Hello, I have been looking into implementing a decentralized sequencer for research purposes using zkSync. So far, I have seen that progress has been made by implementing a custom consensus (ChonkyBFT) in the era-consensus repo, but I cannot find instructions for running it. Also, it is not entirely clear how the decentralized sequencer would integrate with the current zksync-era architecture.

So I was wondering if there is some type of interface that the sequencer must implement (whether it be centralized or decentralized) so that it can interact with the rest of the components regardless of its architecture. If so, where can I find it? If not, how can one integrate a decentralized sequencer into the current zksync-era architecture?

Thank you!

Repo Link (Optional)

No response

Additional Details

No response

[brunoffranca]

Hi, @khalidzahra. I'm the consensus lead, I can help you.
First, era-consensus is already integrated on zksync-era and running on Era. Indeed I have not yet written instructions on how to enable it (very busy here!), but it's not complicated.
So, are you interested in running consensus on your own chain or are you interested in understanding how era-consensus integrates with zksync-era? If it's the first, I can write and publish the instructions. If it's the second, I can point you to the integration code.

[brunoffranca]

By the way, it's also possible to run era-consensus in stand-alone mode (basically it uses dummy blocks). Not sure if that's interesting for your research.

[cytadela8]

Please mark as unanswered when you have questions so we are notified

[khalidzahra]

Hey @brunoffranca, thank you for taking the time to get back to me! I am interested in understanding how era-consensus integrates with zksync-era in its current form. Would love it if you give me some pointers!

[brunoffranca]

Sure. I'll assume you have some familiarity with era-consensus and focus just on the integration.
So, era-consensus exposes 3 traits that must be implemented for any codebase that uses it. They are the payload manager, the block store and the replica store.
Payload manager is used by replicas to get block proposals and verify them, that entire logic is handled by the execution layer (in our case zksync-era), for era-consensus blocks are basically just binary blobs. Replica store is used to persist some replica state if the replica crashes or restarts, needed to maintain safety of the BFT algorithm. Block store is used by all nodes, not only replicas, and is mainly used to store blocks and sync with other nodes. If you have a codebase that imports era-consensus as a library and implements these 3 traits, you can run it. That's exactly how we run load tests, we have a dummy implementation of the 3 traits.
In zksync-era the consensus code is here. This is where you'll find the implementation of the traits and the integration with the rest of the codebase.
One thing that is worth mentioning is that currently the consensus component in zksync-era runs asynchronously. Meaning that the main node just produces blocks like usual and the consensus just runs behind trying to finalize each block. This was implemented like this initially because 1) it made the integration much easier and 2) if consensus failed it wouldn't stop the chain. Evidently this doesn't support leader rotation and so the integration will need to be changed to be sync (i.e. execution waits for consensus). This will be done when we are confident that our consensus code works flawlessly.

[brunoffranca]

Hopefully this helps! Feel free to ask if you have any doubts.