Back to Details

phargus-verify

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Phargus Verify

Phargus Verify

Determine whether the requested change achieved its runtime goal. Read verification and data semantics. Use tools for capture, summarize and compare.
  1. Establish the goal and baseline before editing code or running the after measurement. Resolve this skill's absolute directory; invoke its scripts/phargus.mjs there.
  2. Record workload, concurrency, requests, warmupMs, durationMs, nodeVersion, machine, intervalMs and windowMs in capture conditions. Record application/config revisions.
  3. Reproduce matching conditions after the change. Keep both data files and manifests.
  4. Run compare with both manifests. Different conditions, missing data, truncation or unknown loss yield an inconclusive result; obtain a controlled raw capture to resolve it. Live capture stops at 10,000 events by default: for a high-rate service pass --max-events or a shorter --duration-ms. compare reads files without an event-count cap.
  5. For comparable measurements, assess deltas against the goal and inspect collateral heap/RSS, GC and backpressure changes. Repeat if normal variation can change the verdict.
Return before/after values, conditions, quality and a justified verdict with remaining uncertainty. A smaller number under lower load does not establish improvement. Read compatibility before requesting unsupported profiling.
Example request: "Verify that replacing this synchronous handler improved runtime latency."
确定所请求的变更是否实现了其运行时目标。 阅读验证和数据语义。 使用工具进行捕获、汇总和比较。
  1. 在编辑代码或运行变更后测量之前，明确目标并建立基线。 解析此 skill 的绝对目录；在其中调用 scripts/phargus.mjs。
  2. 在捕获条件中记录工作负载、并发数、请求数、warmupMs、durationMs、nodeVersion、machine、intervalMs 和 windowMs。记录应用程序和配置的修订版本。
  3. 变更后复现相同的测量条件。保留两个数据文件及其 manifest。
  4. 使用两个 manifest 运行 compare。测量条件不同、数据缺失、数据截断或存在未知数据丢失时，结果均不确定；应获取受控的原始捕获数据来解决这些问题。实时捕获默认在 10,000 个事件处停止：对于高速率服务，请传入 --max-events 或使用更短的 --duration-ms。compare 读取文件时不受事件数量上限限制。
  5. 对于可比较的测量结果，根据目标评估各项变化，并检查相关的 heap/RSS、GC 和 backpressure 变化。如果正常波动可能改变结论，则重复测量。
返回变更前后的数值、测量条件、数据质量和有依据的结论，并说明剩余的不确定性。较低负载下数值更小并不能证明性能得到改善。请求不支持的 profiling 前，请阅读兼容性说明。
示例请求：“验证用此同步处理器替换原处理器后是否改善了运行时延迟。”