Summary

• Collapses Push() from a two-lock pattern (RLock for isFull() → Lock for insert) into a single write-lock with inline capacity check
• Eliminates a TOCTOU race: under the old pattern, the queue could become full between the RLock check and the Lock acquisition, leading to unbounded growth past maxSize
• Removes the now-obsolete isFull() helper

Motivation

The original Push() called isFull() which acquired an RLock, released it, then acquired a full Lock for the actual insert. Between the two locks, another goroutine could insert, meaning:

1. The isFull() check could pass, but by the time Lock is acquired the queue is already full — silently exceeding maxSize
2. Conversely, isFull() could reject when there was actually space — unnecessary rejections under contention

The fix is a single Lock that checks capacity and inserts atomically.

Benchmark Results (benchstat, 5 runs, Apple M4 Pro)

This change affects the Insert path, not SelectBy directly. However, the iterator benchmark shows indirect improvement from reduced lock contention between the insert goroutine and the background queue drainer:

                                      │   main   │ queue-lock │
                                      │  sec/op  │ sec/op  vs │
EVMMempoolIterator/EVM_2500/Cosmos_0     4958µ     3362µ  -32%
EVMMempoolIterator/EVM_10000/Cosmos_0   20541µ    17910µ  -13%
EVMMempoolIterator/EVM_0/Cosmos_50        97µ       73µ   -24%
EVMMempoolIterator/EVM_0/Cosmos_500      224µ      150µ   -33%
EVMMempoolIterator/EVM_0/Cosmos_2500    1212µ      655µ   -46%
EVMMempoolIterator/EVM_0/Cosmos_5000    2196µ     1462µ   -33%

Geomean latency:  -22%

Files Changed

Test Plan

• go test -tags=test -race ./mempool/... — 88 tests pass with race detection
• 5 queue-specific tests pass
• Correctness: eliminates TOCTOU race on capacity check

🤖 Generated with Claude Code