This tutorial is an outline for how to set up your own L2 roller for Urbit. Some familiarity with how L2 works in general, as well as the role of the roller, is expected. See Layer 2 Overview for a technical overview of L2, Azimuth Data Flow to gain an understanding of how Azimuth data handling and processing is done on Urbit, and Rollers for a short summary of what the different Gall agents involved for rollers are. This tutorial is focused on setting up a roller to work on the main Ethereum network, but only minor changes are needed to utilize the Ropsten Ethereum network. We also explain how to set up a front end (Bridge) from which transactions to be batched by the roller are sent, but use of a front end is not mandatory.

Note that this process involves giving the private key of an Ethereum wallet to the ship running the roller so that it may spend ETH to submit transactions. If you do not fully understand what this entails, we recommend against running your own roller on the Ethereum mainnet. See below for more information on this.

Henceforth, by "roller" we are referring to the ship used as the transaction aggregator and batch submitter, %roller refers to the Gall agent running on the roller that performs these actions, and "front-end" refers to the optional web GUI most users will use to interact with your roller (probably Bridge).

There are three main steps involved with setting up a roller:

• ensuring that the data in %azimuth is up to date,
• starting and configuring %roller,
• aiming your front-end at the roller

1. Make sure %azimuth state is up to date

If you are using an ordinary live ship on the network as the roller, you should already have the latest %azimuth state and this step should not be necessary and you may move to step 2.

If your roller is a fakezod, you will need to configure %azimuth to pull the latest Azimuth data from Ethereum. The most common way to do this is via an Infura node, but you could instead use e.g. your own Ethereum node. A free tier Infura node ought to be sufficient for most people and is easy to set up using their tutorials.

To set up your fakezod to automatically receive Azimuth updates from the Infura node, enter the following commands in dojo: :azimuth %resub followed by :azimuth|watch 'https://MAINNET_INFURA_URL' %default. Here, the url can be found under the Setting page for the node on infura.io listed under ENDPOINTS.

If you do not perform this step, you'll later see an error "roller not ready" when the first roller batch is about to be submitted.

2. Starting and configuring %roller

This step must be performed whether you're using a fakezod or a live ship.

First we need to start %roller and %roller-rpc (the agent used to send commands to %roller via HTTP). If you are not making use of a front-end, and instead will be accepting transactions to batch entirely from within Urbit, %roller-rpc is not necessary. We assume you will be using Bridge as a front-end.

To start the agents, enter the following command in dojo: |rein %base [& %roller] [& %roller-rpc]. You should see the following response:

gall: installing %roller-rpc
>   'init'
gall: installing %roller
>   %received-azimuth-state

Note that %received-azimuth-state only indicates that %roller successfully subscribed to %azimuth; it does not indicate that you successfully completed step 1.

Next we need to configure %roller. %roller has several parameters that can be tweaked, such as how many transactions it will accept from a given ship in a given time period, how long the time period is, how much gas to use, exceptions to these rules for particular ships, etc. We will cover these in a moment, but first there are two mandatory settings that must be configured: the Ethereum node to submit batches to, and the private key for the wallet that will submit the transactions.

First let's set the Ethereum node to which batches will be submitted, which we will presume is an Infura node. This does not need to be the same node given in step 1, but it is fine for them to be the same (assuming it is configured to accept transactions - some Infura nodes are set up only to make blockchain data available but do not participate in propagating transactions). Enter :roller|endpoint 'https://MAINNET_INFURA_URL' %mainnet into dojo.

Next we set the private key for the Ethereum wallet from which batches built by %roller will be submitted. We strongly emphasize that this step has serious security implications. %roller will be able to spend ETH in this wallet. This is mandatory, of course, as the primary function of %roller is to submit batches of transactions to Ethereum, which requires ETH. If you downloaded %roller from someone other than Tlon and have not personally audited their code, you are putting the assets in this wallet at risk. Also keep in mind that while Urbit has undergone several security audits, some components such as the runtime have never been audited by a third party. We maintain a list of these audits here, as well as an FAQ on the state of security in Urbit. Thus we recommend a fresh wallet generated specifically to be used by %roller and to only put the amount of ETH in it that you expect to need in the near future. We also recommend that you only use this address for sending L2 batches, since otherwise the nonce will get out of sync. %roller can detect this and handle it, but it could cause a delay in batch submission.

With that being said, the command to set the private key for the wallet to be utilized by the roller is :roller|setkey '0x1234567890abcdef', where 0x... is the private key.

The :roller|endpoint and :roller|setkey commands are the only mandatory commands to set up the roller. At this point, if your roller is live on the network (as opposed to a fakezod) and everything has been performed correctly, it would be possible for others to submit transactions to it manually. All other settings are sensible defaults at time of writing. However, to utilize a front-end, one more command is needed: |cors-approve 'https://YOUR_FRONTEND', where the url is the address of the front end you're using - probably a modified Bridge that you're hosting somewhere which you'll set up in the next step. fakezods can only accept transactions from other ships it can talk to, namely other fake ships running on the same machine, so you will need to set it up to work with a front-end if you want to use it on livenet.

We cover the additional settings for %roller at the end.

3. Aiming Bridge at the roller

The last step is to set up the web interface by which users can submit transactions to be batched by the roller, which we refer to as the front-end. We presume here that you'll be using Bridge, by which we mean you'll be hosting a custom version of Bridge modified to use the roller you set up above instead of Tlon's roller.

First you'll need to make the urbit running %roller accessible to the web. This is outside the scope of the tutorial, but we suggest using a tool like caddy or nginx. You'll want to set up a URL such as https://myroller.sampel-pal.net and use the port you'd ordinarily use to access Landscape, probably 80 or 8080. To be clear, this is the same process you'd use if you set up a custom URL to access your urbit hosted in the cloud.

To point Bridge at your roller we launch Bridge with a custom command that modifies an environment variable called REACT_APP_ROLLER_HOST:

REACT_APP_ROLLER_HOST=https://myroller.sampel-pal.net npm run pilot-mainnet

This will launch a server running Bridge that utilizes the mainnet roller you set up at https://myroller.sampel-pal.net.

Additional %roller commmands

%roller has a few other settings and commands for managing things like the rate at which transactions are submitted and manually submitting batches. These can be modified using the following generators: