v0.2.4

Released: October 2026

v0.2.4 introduces major low-level architectural optimizations across the core key-value engine, network pipelining, and parsing layers. Pipelined GET achieves an all-time peak throughput record of 53.38M ops/sec (187ms for 10M keys), Mixed SET/GET drops to 44ms (45.87M ops/sec), and Hot Key contention latency drops by over 2.4× to 265ms (37.68M ops/sec).

---

Highlights

• Pipelined GET Breakthrough (53.38M ops/sec):

  • 187ms for 10 million GET operations under loopback TCP with 100 concurrent clients.
  • Zero-copy cache hit path in dispatch_get_run: CompactKey::eq_bytes validates raw bytes directly on prefetched Entry pointers, completely bypassing std::str::from_utf8 validation and function dispatch layers on cache hits.
  • RESP bulk format strings are synthesized directly into connection output write buffers.

• In-Place Atomic CAS SET Updates (39.62M ops/sec):

  • Pipelined SET achieves 39.62M ops/sec (252ms for 10 million operations).
  • In-place mutation via SmallStr::overwrite_bytes under per-key spinlock with sequence bumping and release ordering, eliminating heap allocations and string clones on update paths.

• Hot Key Contention Slashed by 2.4× (37.68M ops/sec):

  • Single-key contention latency plummeted from 647ms down to 265ms (over 2.4× faster) for 10 million operations on a single contended key across 100 concurrent clients.
  • Direct atomic CAS in-place updates eliminate contention bottlenecks on hot entries.

• Mixed SET/GET Sub-45ms (45.87M ops/sec):

  • Mixed SET/GET (50/50) execution time dropped from 69ms to 44ms (45.87M ops/sec).
  • Redundant prefetch bypass in prefetch_ahead: detects incoming pipelined GET (*2\r\n$3\r\nGET) and SET (*3\r\n$3\r\nSET) runs and yields directly to specialized run engines, eliminating 32 duplicate hash calculations and slot probes per pipeline batch.

• Sub-100ms Latency Across Complex Workloads:

  • Pub/Sub delivery: 88ms elapsed, 113.98M msg/sec fan-out throughput.
  • Pub/Sub publish: 87ms elapsed, 2.29M ops/sec.
  • INCR (atomic counters): 18ms elapsed, 108.56M ops/sec.
  • LPUSH/RPOP (queue): 30ms elapsed, 66.93M ops/sec.
  • HSET/HGET (sessions): 45ms elapsed, 44.44M ops/sec.
  • SADD (sets): 50ms elapsed, 40.38M ops/sec.
  • JSON.GET (documents): 76ms elapsed, 26.21M ops/sec.
  • XADD (streams): 84ms elapsed, 23.76M ops/sec via WriteOp::Xadd and Store::xadd_batch varint streaming.

• Wire Parser Branch Reduction:

  • Replaced split 32-bit and 16-bit comparisons in try_fast_get_at, try_fast_get_key_at, try_fast_set_at, and try_fast_set_key_val_at with single unaligned 64-bit word reads (u64::from_ne_bytes), reducing branch count across all fast-path parsers.

• Geospatial & Sorted Set Enhancements:

  • Replaced 52-step division loop in src/storage/geo.rs with 5-instruction Morton bit-interleaving (spread26), dropping GEOADD/GEODIST elapsed time under 199ms.
  • Direct output buffer writing in zadd_added_flags_into and right-append optimization in ZSetData::place with CHUNK_MAX = 64.

• Strict Memory Safety:

  • Maintained 24-byte SmallStr footprint and 64-byte Entry size in mimalloc.
  • Peak RSS is 876.28 MB, strictly under the 900MB memory ceiling.
  • Zero clippy warnings under cargo clippy --workspace -- -D warnings.

---

Full Benchmark Summary (v0.2.4)

Single FyroDB node on a 6-core Intel i5-11400H (12 hardware threads), loopback TCP, 100 concurrent clients:

Benchmark	v0.2.4 Throughput	v0.2.4 Elapsed	Redis Cluster (12 nodes)	Speedup vs Redis
Pipeline-64 SET	15.41M ops/sec	649ms	5.00M ops/sec	3.08×
Pipelined SET	39.62M ops/sec	252ms	6.53M ops/sec	6.07×
Pipelined GET	53.38M ops/sec	187ms	11.62M ops/sec	4.59×
Mixed SET/GET (50/50)	45.87M ops/sec	44ms	5.82M ops/sec	7.88×
INCR (atomic counters)	108.56M ops/sec	18ms	6.54M ops/sec	16.6×
HSET/HGET (sessions)	44.44M ops/sec	45ms	6.63M ops/sec	6.70×
LPUSH/RPOP (queue)	66.93M ops/sec	30ms	7.17M ops/sec	9.33×
SADD (per-client sets)	40.38M ops/sec	50ms	6.27M ops/sec	6.44×
ZADD (per-client zsets)	17.83M ops/sec	112ms	4.78M ops/sec	3.73×
JSON.GET (documents)	26.21M ops/sec	76ms	3.23M ops/sec	8.11×
JSON.SET (documents)	16.28M ops/sec	123ms	3.23M ops/sec	5.04×
SET+EXPIRE (cache TTL)	15.14M ops/sec	132ms	2.41M ops/sec	6.28×
Hot Key (contention)	37.68M ops/sec	265ms	1.61M ops/sec	23.4×
Producer/Consumer (50+50)	40.61M ops/sec	49ms	1.70M ops/sec	23.9×
Pub/Sub publish	2.29M ops/sec	87ms	161.9K ops/sec	14.1×
Pub/Sub delivery	113.98M ops/sec	88ms	8.09M ops/sec	14.1×
XADD (streams)	23.76M ops/sec	84ms	4.27M ops/sec	5.56×
SETBIT/BITCOUNT (bitmap)	38.11M ops/sec	52ms	5.94M ops/sec	6.42×
PFADD/PFCOUNT (HyperLogLog)	63.43M ops/sec	32ms	6.03M ops/sec	10.5×
GEOADD/GEODIST (geospatial)	10.06M ops/sec	199ms	3.07M ops/sec	3.28×
Vector HSET load	100.2k ops/sec	499ms	66.4K ops/sec	1.51×
Vector HNSW KNN-10	4.9k ops/sec	410ms	1.1K ops/sec	4.45×
Text term search	34.3k ops/sec	29ms	1.6K ops/sec	21.4×