feat(mempool): Add TxTracker support for handling temporary rejections #646

vladjdk · 2025-09-22T15:37:37Z

Description

This PR adds the tracking of priority transactions via TxTracker, a local queue. This is populated by checking for temporary rejections (such as in the case of nonce gaps or underpriced tx), and adding it to the TxTracker if that is the case. This is a separate state and worker that runs in a goroutine and periodically (initially after 10s) reevaluates its contents. No longer valid txs are dropped, newly valid txs are queued, and the rest remain in the local queue.

If the home directory is configured and not blank, then the contents of the TxTracker are journaled and persisted to disk (by default in ~/.evmd/data/txpool/transactions.rlp). This is picked up between node restarts so that those local txs are not lost.

This PR also adds the 4 new fields (locals, no-locals, journal, rejournal) that is used by TxTracker to toml.go and server flags under evm.mempool. The configureEVMMempool/GetLegacyPoolConfig helpers have also been updated to make use of the toml and/or cli flag values.

Closes: #500

Author Checklist

I have...

tackled an existing issue or discussed with a team member
left instructions on how to review the changes
targeted the main branch

mempool/txpool/locals/journal.go

+	for _, txs := range all {
+		for _, tx := range txs {
+			if err = rlp.Encode(replacement, tx); err != nil {
+				replacement.Close()
+				return err
+			}
+		}
+		journaled += len(txs)
+	}


mempool/txpool/locals/journal.go

+	for _, txs := range all {
+		for _, tx := range txs {
+			if err = rlp.Encode(replacement, tx); err != nil {
+				replacement.Close()


To correctly handle potential data loss or errors that occur upon closing a writable file handle, we must check and handle the error returned from replacement.Close() in all places it is called. Specifically:

In the loop on line 158, after a failed rlp.Encode(replacement, tx), the code calls replacement.Close() but ignores the error. We should close the file, check the return value, and if an error occurs, wrap it (or ensure it is reported). If both Encode and Close fail, we should not lose either error—often wrapping or combining is best.

Similarly, after the write loop on line 164, the call to replacement.Close() is made, but its error is ignored; we should check this and return the error if it happens.

The cleanest Go idiom is to set up a named return value, and when closing after an error, either wrap or combine the errors (e.g., use fmt.Errorf("...: %w", err)). In the normal path, if replacement.Close() fails, return its error.

No new imports are required, as fmt.Errorf is already available.

All changes occur in mempool/txpool/locals/journal.go, in the rotate function.

mempool/txpool/locals/journal.go

+		}
+		journaled += len(txs)
+	}
+	replacement.Close()


The best way to fix this issue is to explicitly check the error returned by replacement.Close() after all write operations in the journal rotation process. This involves modifying the code region around line 164 to assign the result of Close() to an error variable, and handle it appropriately (return it if it occurs and there’s no prior error). Since this code is currently after a loop of writes, we must ensure that a failure in Close is treated as a failed journal rotation. The fix should only change this portion of code, add no new functionality, retain all code structure, and avoid modifying file-level logic elsewhere.

This does not require new imports, but does require updating error assignment/handling in the rotate method.