How will clients connect to Tableland when there is more than one Validator?

[joewagner]

I have the sense that this is a complex question. As a means to spur discussion I'll propose a small set of terms, and some questions and considerations. Everyone should feel free to correct the definitions of the terms, or propose we use others. In regard to the questions, in some cases depending on the answers we choose new questions will probably arise. Definitely fire away with thoughts or ideas here!

Terminology: (open to changes)

• Validator: All of the codebase inside go-tableland, minus the go client that lives inside /pkg
• Gateway: I don't have a good sense of what a Gateway is specifically responsible for in Tableland. See questions below.
• Client: The code that lives in js-tableland as well as the go client mentioned above. In general these will make http requests to a Validator (maybe via a Gateway?). They also make requests to an EVM chain network via a provider like Infura or Alchemy.
• EVM chain: Some examples of this are Ethereum Mainnet, Goerli, Polygon Mumbai, etc...
• [UPDATE] Tableland Network: a Tableland Network is a combination of a the TablelandTables.sol contract deployed on a specific EVM chain and a Validator that is listening to that contract for events and serving http requests. Of note: there is no reason that there can't be multiple Tableland Networks on the same EVM chain, and a Validator can and will participate in many Tableland Networks.

Questions and Considerations:

• What is the relationship between Validators and Gateways?
• When there is more than one Validator, will each Validator need it's own DNS entry?
• We have been using the terminology that we are "connecting to Validators", should we be "connecting to Gateways"?
• If each Validator does not run it's own Gateway will we have multiple Gateways for a single Tableland Network? In the case of multiple Validators per Gateway how do Gateways choose which Validator to direct requests to?
• Do we need to separate the go-tableland repo into two? One for Validator, one for Gateway.
• Do clients need to specify an RFC-3986 URL where the Validator can be reached? People are most often going to use whatever default value the client specifies. If there are many Validators and one default we could end up in a situation where even though we have a network of Validators, only a single Validator is handling all the requests. Will this have an affect on how much incentives a Validator receives for running their instance? It will be worth understanding how The Graph planning to handle this problem.
• There is an outstanding issue concerning using the concept of an "approved address" to enable Validators to act on behalf of an EOA for Registry contract interaction. We are planning to build this out, and when we do the client is going to need a way to ensure the Validator they are approving and connecting to are the same.

Related issue here
There are related commets in this PR

[joewagner]

Here's a first attempt at a proposal to alleviate this tension: https://github.com/orgs/tablelandnetwork/discussions/242