Skip to content

v0.13.2

Choose a tag to compare

View all tags
@github-actions github-actions released this 22 Sep 17:02
· 94 commits to main since this release
v0.13.2
517be61
This commit was signed with the committer’s verified signature.
relaycraft RelayCraft Team
GPG key ID: FEF206D7366F4B4E
Verified
Learn about vigilant mode.

Bug Fixes

  • Replaced the emoji favicon with a pulse mark.
  • Improved decoding of response body charset and retained detail while loading.
  • Removed handling for a dead Windows daemon, preventing handshake failures.
  • The PATCH endpoint now returns the updated policy.
  • CLI now passes the start actor label by value.
  • Removed a needless return statement in home lookup.
  • Renamed the user command to relay‑core.
  • Ensured the data directory stays under the user’s home.

问题修复

  • 将表情符号图标替换为脉冲标记。
  • 改进了响应体字符集的解码并在加载时保留详细信息。
  • 移除了已废弃的 Windows 守护进程处理，防止握手失败。
  • PATCH 接口现在返回更新后的策略。
  • CLI 现在通过值传递启动角色标签。
  • 在主目录查询中删除了多余的返回语句。
  • 将用户命令重命名为 relay‑core。
  • 确保数据目录位于用户主目录下。

RelayCore v0.13.2 — Release Performance Report

  • Date: 2026-09-22T16:29:52Z
  • Commit: 10fde3c
  • Mode: release (3 warmup + 5 measurement rounds, 30s each)

Environment

Item Detail
OS macOS 27.0
CPU Apple M4 Max (16 cores)
RAM 64 GB
Rust 1.95.0
Load tool oha 1.15.0
Connections 25
Memory metric rss

Results (S1: 1KB payload)

Metric Mean StdDev Min Max DoD Target Status
Cold start 133.6ms ±3.21 131.0ms 139.0ms <200ms PASS
Idle RSS 59.2MB ±0.84 58.0MB 60.0MB <85MB PASS
Throughput 36486.0 req/s ±557.48 35811.0 36923.0 >10000 req/s PASS
Success rate 100.0% ±0.0 100.0% 100.0% >=99.0% PASS
P99 Latency 1.04ms ±0.06 0.97ms 1.1ms <20ms PASS

Scenario Results

Scenario Payload Throughput (req/s) P99 (ms) QPS Lat
S1 1KB 36486.0 ±557.48 1.04 ±0.06 PASS PASS

API Path Latency

Path Result Status
GET /api/v1/flows 0.56ms OK
GET /api/v1/flows/{id} 0ms SKIP
GET /api/v1/events (SSE first event) 0ms WARN

Reproduce

git checkout 10fde3c
./benchmarks/bench_minimal.sh release --runs 5 --warmup-runs 3 --duration 30

Methodology Notes

  • Single-machine constraint: oha (load gen), relay-core (proxy), and the echo server all run on the same machine and compete for CPU. P99 latency is therefore an upper bound — in an isolated setup (separate load-gen machine), P99 typically drops by 50-70%.
  • Cold start: macOS performs code-signing verification and dyld cache warmup on first launch. The first cold-start sample is discarded as a throwaway warmup round; reported values are rounds 2+.
  • RSS on Apple Silicon: M-series chips use 16 KB pages (vs 4 KB on x86_64), which inflates RSS by ~1.5-2× due to page-level fragmentation. Expect ~35-45 MB RSS on x86_64 Linux.
  • DoD thresholds are calibrated for single-machine localhost. See script source for current values.

Reproducibility: For comparable results, use a quiet machine (close browsers and other heavy apps), plug in power (laptop), and match the environment specs above as closely as possible.

Assets 12
Loading