Phase 3 of Offramp prototype Pendulum -> Spacewalk -> Stellar Mykobo

[prayagd]

This ticket describes a webapp based version of the offramping prototype that uses Spacewalk and a SEP-24 based offramp on Stellar in order to offramp EURC.s on Pendulum.

Goal

The goal of this phase is remove the stellar secret key input field too and allow the user to offramp just by connecting their Pendulum wallet

Todo

This is a follow-up ticket to #19

• Remove the stellar secret key input field
• Build a backend server that:
  • has access to the secret of the Stellar funding account (supplied via env variable)
  • exposes a POST endpoint that receives the to-be-cosigned transactions (offramp and merge-account) of the ephemeral account. The server then checks the content/operations contained in the transactions, signs the transactions and sends them back to the client. The client will then try to submit them as before.
  • Note that we could also make it two different POST endpoints, one for the offramping transaction and one for the merge-account transaction, but I'm not sure if it's worth distinguishing.
  • Make the client send the pre-signed Stellar transactions to the backend before calling the submitOfframpTransaction() and cleanupStellarEphemeral() functions
• The app should make sure the stellar part of the transaction is also signed
• The UI should only have the Begin offramp button

[prayagd]

@gianfra-t @ebma The signing part of the stellar txn and tech details of it are not very clear to me, hence just mentioned as a line item, Would you be able to add the tech details there?

Here is the notion ticket, where Torsten has explained it.

[prayagd]

Hey team! Please add your planning poker estimate with Zenhub @b-yap @bogdanS98 @ebma @gianfra-t @TorstenStueber

[ebma]

@prayagd done.

@TorstenStueber do you already have something in mind for the backend server? In the past we learned that AWS Lambda function and similar are more costly than spinning up a simple service in our infrastructure so I guess we should do that. Should we add the code to this repository as well or put it somewhere else?

[prayagd]

one question

1. Would we need a account were this temporary stellar account be funded from?

[ebma]

Yes we do. Something similar to the account you shared previously. To be on the safe side, we can merge the funds from that account into a new one so that we have a new account with a secret that was not exposed in the browser text field. I would only do this once the implementation of this is almost finished.

[TorstenStueber]

@ebma Lambda was more costly? I actually expect that we are able to use it for free. Do you remember where he had to pay for it?

Anyway, it's best to leave this to @zoveress to find a way to deploy it and we can just write a normal nodejs backend.

Particularly as this is a (possibly throw away) prototype, I would put the backend code in the same repository.

[ebma]

I think we wanted to use Lambda for our token stats API but identified that it's more costly. Though maybe just because we wanted to add an extra cache and rate limiting or something, I don't remember the details.

I would put the backend code in the same repository.

Alright, then let's add it to the same repository. We can again draw inspiration from the implementation of our existing services, like the faucet.

[TorstenStueber]

About the data flow between frontend and backend

There are four transactions that require that both parties (funding account and ephemeral account) sign:

• the transaction to create and set up the ephemeral account
• the transaction to add the trustline
• the offramping transaction
• the merge account transaction

I think that we can combine the first two transactions into a single one. Each of these transactions needs to be signed by both key pairs (so once in the browser and once in the backend).

It is important that the backend does not blindly sign a transaction. Out of experience it is usually easiest and safest to:

• for each transaction create the transaction twice, once in the frontend and once in the backend with shared code, only transmit the parameters that are required to create the transaction
• only send signatures between frontend and backend and not the transaction

I suggest two endpoints:

• one to create a signature for the first two transactions (that we can combine)
• one to create two signatures for the last two transactions

[TorstenStueber]

Some decisions we took today:

• Node.js backend (instead of Rust) – allows to reuse code for Stellar transactions
• no database required for this prototype

[gianfra-t]

Regarding the data flow, we could make the back-end send the finalized transaction and only respond with a success or error, and delegate stellar error handling to the back-end.

Or alternatively, we can send back the transaction signed by the back-end to the client again, and maintain the error handling in the client.
Any preference regarding these alternatives @TorstenStueber @ebma ?

[ebma]

Or alternatively, we can send back the transaction signed by the back-end to the client again, and maintain the error handling in the client.

I think it's better to keep the back-end lean and leave the transaction submission and error handling to the client.

[gianfra-t]

Yes I thought the same too, in fact it would be the easiest to create the tx on the backend, sign it, and then send it to the user for final co-sign.

Now, I have another doubt that is more in lines with security or potential exploits. If we at any point allow the user control of the ephemeral secret, we risk an exploit in which the user purposely shuts down the process before the clean up (could be as soon as the creation of the ephemeral account is submitted).
So a malicious could do this and we would end up with many funded but irrecoverable ephemeral accounts.

[TorstenStueber]

In the future it would make sense to let the backend store the transactions and also submit them as it should not rely on the user to submit the transactions once Spacewalk transferred the money to the ephemeral accounts (otherwise the funding account can be drained). But this goes beyond (this version of) the prototype.

For the data transport I again recommend that not transactions but only signatures are sent. That is from our experience that it is much cleaner and safer code to recreate transactions on both sides than validating a transaction (there are a lot of many little details that need to be validated in order for this to be safe and this code is brittle if we need to change transaction details).

Transaction Envelopes

Here are some details how to do this. The TransactionBuilder does actually not create a transaction, but a transaction envelope, two different things. The envelope consists of a) a transaction and b) an array of signatures for this transaction (confusingly the TypeScript type for the envelope in the stellar-sdk is just Transaction).

This allows to attach multiple signatures to the envelope. The envelope is what is ultimately submitted to the Stellar horizon (RPC) server, not the blank transaction.

For our envelopes we need to attach two signatures, the one from the ephemeral keypair and the one from the funding account keypair. The TransactionBuilder always creates an envelope with an empty array of signatures.

The sign method of the envelope (TypeScript type Transaction) does two things: create a signature and attach it to the envelope.

We need two new operations now:

• create a signature for an envelope and just return the signature (no attachment required)
  • done via the method getKeypairSignature: envelope.getKeypairSignature(keypair)
• attach a given signature to an envelope
  • done via the method envelope.addSignature(accountId, signature) – this additionally requires the account id (public key) of the signing key pair