feat(txpool): add configurable batch timeout to transaction pool insertions #20661

keanji-x · 2025-12-29T04:05:53Z

Introduced a mechanism to forcefully batch transactions for a specified duration using --txpool.batch-timeout.

The previous implementation relied on greedy consumption (poll_recv_many), which processes whatever is available immediately. Under low-to-medium load, this often results in small batches (or single transactions), failing to effectively amortize the cost of acquiring the transaction pool lock.

By introducing a timeout, we can enforce buffering up to max_batch_size or batch_timeout, significantly improving lock contention and throughput by ensuring larger persistent batches.

Implementation Details

Used Optiontokio::time::Interval in BatchTxProcessor to maintain zero-cost overhead for the default immediate mode (timeout=0).
Refactored the poll loop to support both immediate consumptions and time-buffered batching.
Added config propagation and integration tests.

feat: add `batch_timeout` configuration for transaction pool insertions. refactor: batcher poll method to continuously process available batches and clarify flush conditions.

codspeed-hq · 2025-12-29T04:25:37Z

CodSpeed Performance Report

Merging #20661 will not alter performance

_{Comparing keanji-x:fix_timeout (7408999) with main (3d4efdb)}

Summary

✅ 118 untouched
⏩ 7 skipped¹

7 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

mattsse

this makes sense and I#ve thought about this as well because most of the time this ends up just batching 1 single request, but even waiting a bit should be very beneficial here.

left some suggestions and a q re the poll logic

mattsse · 2025-12-29T10:21:01Z

crates/rpc/rpc/src/eth/builder.rs

    max_batch_size: usize,
+    batch_timeout: Duration,


nit: could we introude a dedicated Batchconfig type?

mattsse · 2025-12-29T10:23:29Z

crates/transaction-pool/src/batcher.rs

+        let interval = if batch_timeout.is_zero() {
+            None


can we make this an option then instead?

mattsse · 2025-12-29T10:26:27Z

crates/transaction-pool/src/batcher.rs

+        match this.interval.as_mut().as_pin_mut() {
+            // Immediate mode (timeout = 0): zero-cost path, original behavior
+            None => loop {
+                match this.request_rx.as_mut().poll_recv_many(cx, this.buf, *this.max_batch_size) {
+                    Poll::Ready(0) => return Poll::Ready(()), // Channel closed
+                    Poll::Ready(_) => {
+                        Self::spawn_batch(this.pool, this.buf);
+                        this.buf.reserve(*this.max_batch_size);
+                        // continue to check for more requests
+                    }
+                    Poll::Pending => return Poll::Pending,
+                }
+            },


all of this looks a bit too complex

how should this behave exactly?

can we not use just the interval to yield pending if it's not ready yet?

Fixed, thanks for review

…ration and batching modes.

…e `BatchConfig` struct.

mattsse

I'd like to restructure the poll logic a bit so that this behaves like:

poll_recv many
if batch empty -> return early
if batch full -> spawn + continue
if batch not full and interval is set: poll interval

I'd also like to see some manual poll tests for this

mattsse · 2025-12-31T09:29:05Z

crates/transaction-pool/src/batcher.rs

-                continue;
+            // 1. Collect available requests (non-blocking)
+            while this.buf.len() < *this.max_batch_size {
+                match this.request_rx.as_mut().poll_recv(cx) {


I think this should keep using poll_recv_many

mattsse · 2025-12-31T09:30:36Z

crates/transaction-pool/src/batcher.rs

+
+            if batch_full || timeout_ready {
+                Self::spawn_batch(this.pool, this.buf);
+                this.buf.reserve(*this.max_batch_size);


this is wasteful, because this will most likely over allocate

mattsse · 2025-12-31T09:32:34Z

crates/transaction-pool/src/batcher.rs

+            // 2. Check flush conditions
+            let batch_full = this.buf.len() >= *this.max_batch_size;


if the batch is full we can spawn it immediately and then only need to reset the interval and can skip polling it

mattsse · 2025-12-31T09:36:19Z

crates/transaction-pool/src/batcher.rs

+                !this.buf.is_empty()
+            };
+
+            if batch_full || timeout_ready {


I find timeout_ready a bit confusing here because if no timeout is set, then this represents something else

…icient request collection and enhance batching logic with new tests.

keanji-x requested review from Rjected, klkvr, mattsse and yongkangc as code owners December 29, 2025 04:05

github-project-automation bot added this to Reth Tracker Dec 29, 2025

github-project-automation bot moved this to Backlog in Reth Tracker Dec 29, 2025

keanji-x changed the title ~~feat: add configurable batch timeout to transaction pool insertions~~ feat(txpool): add configurable batch timeout to transaction pool insertions Dec 29, 2025

feat: add configurable batch timeout to transaction pool insertions

16065ff

feat: add `batch_timeout` configuration for transaction pool insertions. refactor: batcher poll method to continuously process available batches and clarify flush conditions.

keanji-x force-pushed the fix_timeout branch from a211735 to 16065ff Compare December 29, 2025 04:14

fix clippy

6669802

keanji-x requested a review from gakonst as a code owner December 29, 2025 04:29

fix

edc2c11

keanji-x force-pushed the fix_timeout branch from 714cd63 to edc2c11 Compare December 29, 2025 04:33

mattsse requested changes Dec 29, 2025

View reviewed changes

github-project-automation bot moved this from Backlog to In Progress in Reth Tracker Dec 29, 2025

mattsse added M-changelog This change should be included in the changelog A-rpc Related to the RPC implementation A-cli Related to the reth CLI labels Dec 29, 2025

refactor: introduce BatchConfig to unify BatchTxProcessor configu…

ebaa386

…ration and batching modes.

keanji-x force-pushed the fix_timeout branch 3 times, most recently from a892bef to 5579b0b Compare December 29, 2025 12:11

refactor: Consolidate transaction batching configuration into a singl…

f324ebf

…e `BatchConfig` struct.

keanji-x force-pushed the fix_timeout branch from 5579b0b to f324ebf Compare December 29, 2025 12:16

build: enable serde feature for reth-transaction-pool dependency

b45035c

keanji-x requested a review from mattsse December 29, 2025 12:37

mattsse requested changes Dec 31, 2025

View reviewed changes

keanji-x added 2 commits January 5, 2026 10:27

refactor: Refactor BatchTxProcessor to use poll_recv_many for eff…

d6e022a

…icient request collection and enhance batching logic with new tests.

fix clippy

7408999

keanji-x requested a review from mattsse January 5, 2026 03:24

		// 2. Check flush conditions
		let batch_full = this.buf.len() >= *this.max_batch_size;

feat(txpool): add configurable batch timeout to transaction pool insertions #20661

Are you sure you want to change the base?

feat(txpool): add configurable batch timeout to transaction pool insertions #20661

Uh oh!

Conversation

keanji-x commented Dec 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Introduced a mechanism to forcefully batch transactions for a specified duration using --txpool.batch-timeout.

Implementation Details

Uh oh!

codspeed-hq bot commented Dec 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging #20661 will not alter performance

Summary

Footnotes

Uh oh!

mattsse left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mattsse left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

keanji-x commented Dec 29, 2025 •

edited

Loading

codspeed-hq bot commented Dec 29, 2025 •

edited

Loading