[haardikk21]

Stacked on #31; the diff is the last commit only.

Summary

BestTransactions::pop_best removed each yielded transaction from all. all is an imbl::OrdMap cloned from the pending pool, so its nodes are shared with the pool. The first write to a shared node copies that node, including a clone of every PendingTransaction (and its Arc) in it. Transaction ids from different senders are spread over the whole map, so almost every pop copied a fresh leaf.

Popped transactions now stay in all. A popped: FxHashSet<TransactionId> records them instead:

• pop_best skips keys whose id is in popped, then marks the id it yields.
• ancestor ignores popped ids, so a live transaction whose ancestor was already yielded still becomes independent, as before.
• Inserting a live transaction removes its id from popped, so a replacement of an already yielded transaction is yielded again, as before.
• size_hint subtracts popped.len(). popped only holds ids in all, so the bound is exact, as before.

Yield order and outcomes are unchanged. The new test test_best_live_transactions_after_their_ancestor_or_original_was_popped covers both live-update cases. On the parent commit its yield assertions pass too; only its new check that popped transactions remain in all fails there.

Benchmarks

benches/best_transactions.rs (added in #31), ABBA on a Linux devbox (AMD Ryzen AI MAX+ 395, 32 threads, Linux 7.2, rustc 1.96.0). Each arm was built once and the binaries ran in the order A, B, B, A, pinned to cores 4–15, with 2s warm-up and 5s measurement. A = #31 (e69a3ac09), B = this change. Times are means of the A runs and of the B runs, in µs.

Scenario	A	B	Δ	A runs	B runs
take_512, 1k senders	71.1	64.7	−8.9%	70.9 / 71.2	64.1 / 65.4
take_512, 10k senders	100.4	85.7	−14.7%	100.4 / 100.4	85.8 / 85.6
take_512, 40k senders	125.6	106.7	−15.1%	125.5 / 125.7	105.4 / 108.0
take_512, 10k senders × 4	130.5	114.3	−12.4%	130.7 / 130.3	114.0 / 114.7
drain, 1k senders	129.7	123.5	−4.8%	130.2 / 129.2	121.4 / 125.6
drain, 10k senders	1,824.2	1,777.9	−2.5%	1,822.1 / 1,826.4	1,764.2 / 1,791.6
drain, 40k senders	9,051.6	9,130.1	+0.9%	9,082.6 / 9,020.7	9,113.5 / 9,146.7
drain, 10k senders × 4	9,470.2	9,230.0	−2.5%	9,467.1 / 9,473.4	9,225.2 / 9,234.8
snapshot, 40k senders	0.070	0.072	+2.1%	0.070 / 0.070	0.071 / 0.072
add_128_with_live_snapshot, 40k senders	135.4	136.3	+0.7%	137.7 / 133.0	134.1 / 138.5

take_512, which matches how payload builders use the iterator (take a prefix, then drop it), saves about 30–40 ns per yielded transaction. A full drain eventually copies most of the shared map whichever way it is mutated, so it gains less.

Pool writes while a snapshot is live (add_128_with_live_snapshot) are unchanged.

cargo bench -p reth-transaction-pool --features test-utils --bench best_transactions

Validation

• cargo test -p reth-transaction-pool --all-features: 280 unit, 15 integration and 3 doc tests pass.
• cargo clippy -p reth-transaction-pool --all-features --all-targets -- -D warnings
• cargo +nightly fmt -p reth-transaction-pool -- --check

[haardikk21]

CI failures here are not from this diff and match #31 / the releases/v2.7.0 base:

• clippy, clippy binaries: missing-backticks lint in crates/storage/storage-api/src/state.rs:197,199 (untouched here).
• Ethereum state tests: job cancelled after the timeout (unit success depends on it).
• docs, cargo deny (RUSTSEC advisories), Scan GitHub Actions (Step Security 403): also failing on perf(txpool): share the pending pool's independent set with best transactions snapshots #31 and other PRs against this base.

reth-transaction-pool tests, clippy (-D warnings, all targets/features) and fmt pass locally.