Support solana-web3.js batching

[epicfaace]

Summary of Changes

Fixes #23627 by adding a config option autoBatch to Connection that, if set to true, batches all JSON RPC requests so they are sent every 50ms.

[t-nelson]

we can't break API like this. i don't think we want this to be the default behavior anyway (imposed batching delay). can you make it a separate function?

[epicfaace]

we can't break API like this. i don't think we want this to be the default behavior anyway (imposed batching delay). can you make it a separate function?

Hi @t-nelson -- this isn't a breaking change, the default behavior is still the same as before. The batching behavior is only triggered when you pass in autoBatch: true to the connection options. Does that work?

[stale]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.

[epicfaace]

this isn't a breaking change, the default behavior is still the same as before. The batching behavior is only triggered when you pass in autoBatch: true to the connection options. Does that work?

Hi @t-nelson -- see my comment above. Does this work for you? Let me know if you'd like me to make any additional changes.

[steveluscher]

Neat! I like batching, in principle. If we do it, we should do it with a global lock that people can opt into in userspace.

For a long time, this is how React itself performed batching.

React.unstable_batchedUpdates(() => {
  setSomeState(true);
  callSomeMethodThatMaybeSetsSomeState(123);
  if (something) {
    setSomething(something);
  }
});

Essentially, the implementation was something like:

function unstable_batchedUpdates(cb) {
  _batch = true;
  try {
    cb();
  } finally {
    _batch = false;
  }
}

…and that _batch variable was accessible by whatever needed to make the batch-or-not-to-batch decision.

Ours might look something like this:

connection.batchRequests(() => {
  connection.requestAirdrop();
  methodThatMightMakeARequest();
  if (needBlockTime) {
    connection.getBlockTime();
  }
});

From there, there would be a ton of work to actually make that work.

Things that would need to be resolved:

• Right now _rpcBatchRequest is written in such a way that one error nukes the entire batch. It doesn't understand that the return value of a batch is an array of the successes and and an array of the errors. If even one request in a batch errors, it nukes the whole promise: https://github.com/steveluscher/solana/blob/5eae7b708c45979cc3573e6be2b932e87bd9ec38/web3.js/src/connection.ts#L1061-L1067. That will need to be fixed. Mental note for me that this sort of means getParsedTransactions is broken, for some definition of broken.
• The API I've suggested above is really confusing if you're used to request method being async. How would you write the above if methodThatMightMakeARequest depended on requestAirdrop finishing first? Is that two batches now? Is it even possible to build a dependency tree of which calls depend on which results, so that you can prepare exactly the number of batches necessary?

[stale]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.

[epicfaace]

Still working on this

[epicfaace]

If we do it, we should do it with a global lock that people can opt into in userspace.

@steveluscher thanks for the feedback. Isn't that (global lock people can opt into in userspace) what this PR enables? Or did you have something else in mind?

[steveluscher]

This PR can be implemented as a customFetch on Connection! You should repurpose it as such, and open source it.

More detail, here: #23627 (comment)