feat(solana): rate limited solana provider

[Reinis-FRP]

This implements the same rate limited provider logic as src/providers/rateLimitedProvider.ts, but for Solana using V2 of @solana/web3.js library that allows creating RPC clients with composable transports.

Note that this @solana/web3.js library produces build errors and CI node version would need to be updated.

It would be quite a lift to implement mocked Solana RPC node for tests, but the expected behavior was tested with:

import { createDefaultRpcTransport, createSolanaRpcFromTransport } from "@solana/web3.js";
import winston from "winston";
import { RateLimitedSolanaRpcFactory } from "@across-protocol/sdk";

const maxConcurrency = 1;
const pctRpcCallsLogged = 10;
const logger = winston.createLogger({
  transports: [new winston.transports.Console()],
  level: "debug",
});
const chainId = 34268394551451; // Last 6 bytes of keccak256("solana-mainnet")
const url = "https://api.mainnet-beta.solana.com";
const rateLimitedRpcClient = new RateLimitedSolanaRpcFactory(maxConcurrency, pctRpcCallsLogged, logger, chainId, {
  url,
}).createRpcClient();

const rateLimitedRequests = Array.from({ length: 20 }, () => rateLimitedRpcClient.getSlot().send());
const rateLimitedResults = await Promise.all(rateLimitedRequests);
console.log("Rate limited results:", rateLimitedResults);

const regularRpc = createSolanaRpcFromTransport(createDefaultRpcTransport({ url }));
const regularRequests = Array.from({ length: 20 }, () => regularRpc.getSlot().send());
const regularResults = await Promise.all(regularRequests);
console.log("Regular results:", regularResults);

This produces following logs showing that new Solana slots get mined during rate limited requests (with concurrency set to 1) while normally these concurrent requests would resolve to the same slot:

{"at":"SolanaProviderUtils","chainId":34268394551451,"datadog":true,"level":"debug","message":"Solana provider response sample","method":"getSlot","params":[{"commitment":"confirmed"}],"provider":"https://api.mainnet-beta.solana.com","success":true,"timeElapsed":0.03452311199999895}
Rate limited results: [
  315643630n, 315643630n,
  315643630n, 315643630n,
  315643631n, 315643631n,
  315643631n, 315643631n,
  315643631n, 315643631n,
  315643631n, 315643631n,
  315643631n, 315643631n,
  315643631n, 315643631n,
  315643631n, 315643631n,
  315643631n, 315643632n
]
Regular results: [
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n,
  315643632n, 315643632n
]

[nicholaspai]

This queuing logic looks almost identical to the EVM version, is there any way to refactor and re-use code? We might be able to unit test any refactored code more easily

[nicholaspai]

I also think it could be useful to refactor and re-use some of this elapsedTime logic. It would also be a bit easier to unit test if we abstract the hypothetical re-used function well.

[nicholaspai]

This is really well written code! I think we can possibly try to reuse some code from the evm provider and that might make unit testing with mocks even easier!

[md0x]

Is it safe to send the params in the payload to Winston without validating or serializing them first?

[Reinis-FRP]

I think it should be safe as we expect these requests to originate via Solana web3.js library. In addition, we use createDefaultRpcTransport here to create the underlying transport that internally type checks the request here.

[md0x]

Looks good to me! Just added a question.