feat: a nonce-based transaction confirmation strategy for web3.js

[steveluscher]

Usage

const tx = new Transaction({nonceInfo});
const signature = connection.sendTransaction(tx, [/* ... */]);
await connection.confirmTransaction({
  minContextSlot, // A slot where the nonce was known to be the candidate value.
  nonceAccountPubkey,  // The account where we can expect to find the latest nonce value.
  nonceValue: nonceInfo.nonce,  // Whatever nonce value the `tx` contained when it was signed.
  signature,
});

Closes #25303.

[codecov]

Codecov Report

Merging #25839 (b03c38f) into master (04789ca) will decrease coverage by 0.2%.
The diff coverage is 69.7%.

@@            Coverage Diff            @@
##           master   #25839     +/-   ##
=========================================
- Coverage    76.8%    76.5%   -0.3%     
=========================================
  Files          55       55             
  Lines        2986     3084     +98     
  Branches      426      454     +28     
=========================================
+ Hits         2294     2362     +68     
- Misses        544      564     +20     
- Partials      148      158     +10     

[steveluscher]

Alright! This is ready for you again @jstarry. All the new commits at the bottom are part of #26296.

[jstarry]

Can you PR the non-confirmation changes separately so that we can get those in more quickly and break up this large PR?

[jstarry]

Could this be refactored to while (!done) and then each loop we either sleep or get the current nonce value? That way we don't need so many done checks.

[steveluscher]

There’s a done check after each async operation, because it’s at that point we need to check to make sure a confirmation didn’t come in while we were waiting for the async op to finish.

For instance, we want to check for done after the sleep, so that we don’t waste time on another nonce fetch for no reason. Putting the done check at the loop level would make this impossible.

[jstarry]

Putting the done check at the loop level would make this impossible

My suggestion was to make the loop only run one async operation per loop so that done is still called immediately after. So it's possible, but I suppose it's less obvious so this isn't a big deal.

        let readyToPoll = false;
        let currentBlockHeight = await checkBlockHeight();
        while (!done && currentBlockHeight <= lastValidBlockHeight) {
          if (readyToPoll) {
            currentBlockHeight = await checkBlockHeight();
          } else {
            await sleep(1000);
          }
        }

        if (!done) {
          resolve({ __type: TransactionStatus.BLOCKHEIGHT_EXCEEDED });
        }

[steveluscher]

I’ll massage this further when I get back to a computer!

[steveluscher]

Alright, done! How about this, instead of separate PRs, I've reduced this PR to 5 commits that each do exactly one thing. You can review each as though it was it's own PR.

Because it's 2022, and GitHub still doesn't have stacked PRs.

[steveluscher]

Updated!

• Rebased
• Updated/added code comments
• Renamed NonceBasedTransactionConfirmationStrategy to DurableNonceTransactionConfirmationStrategy
• Fixed spelling of Exceedance
• Fixed spelling of BlockHeight
• Removed signature from… method signature.
• Created object return datatype for getTransactionConfirmationPromise
• Eliminated INVALID_NONCE_ERROR_SENTINEL
• Added comment pointing out exhaustive switch.

[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.

[jstarry]

Apologies for dropping off on the review of this one. The main points to address are:

1. The logic for double checking that a nonce transaction hasn't been processed isn't quite right: feat: a nonce-based transaction confirmation strategy for web3.js #25839 (comment) and feat: a nonce-based transaction confirmation strategy for web3.js #25839 (comment)
2. Looks like this branch should check for the existence of minContextSlot before referencing it: feat: a nonce-based transaction confirmation strategy for web3.js #25839 (comment)

[steveluscher]

Alright; rebased and shipping this, at long last. If it ends up sucking we can revert it. If we find there are improvements to make, we can follow up with those improvements!