性能套件

工程开发 社区

技能解读

当前原文的自动摘译,尚未人工审校。

工作流程

确定性能表面。将纯构建器/读取模型与瘦请求/视图包装器分开。首先优先对未缓存的服务器端工作进行基准测试。对于 Django UI 工作,在考虑浏览器渲染之前先覆盖只读 GET 表面。创建单独的表演通道。

设计规则

针对 PostgreSQL 进行衡量。 SQLite 计时对于 Django 性能护栏没有用。将时机和正确性视为单独的问题。快速错误的结果仍然是回归。将查询上限与时间预算配对。查询计数通常是最清晰的预警信号。尽可能对两个层进行基准测试:用于诊断的构建器/读取模型成本

实施模式

构建套件时,创建以下部分: 用于性能运行的专用 Django 设置模块。一个tests/performance/包。确定性场景播种者。消除不稳定场的结果标准化器。

应避免什么

当性能测试需要大量设置时,请勿将性能测试混合到默认的 pdm run test 或 pytest 路径中。不要只对微小的装置进行基准测试并假设结果具有普遍性。当可以通过查询上限确定性地捕获大型 ORM 回归时,不要仅依赖于时间。如果原始 HTML 或完整上下文包含会不断变化的不稳定值,请勿保留它们的快照。不要在每次运行后默默地更新预算。这会破坏回归信号。

产出预期

当用户要求此类工作时,生成: 单独的 pytest 通道 确定性场景,大小适合产品表面快照和查询计数保护 计时预算和工件报告

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
工程开发
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@tomevault-io · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
需简单配置
是否需要外部接口密钥
不需要
兼容的系统
未声明(默认跨平台)
运行环境要求
未检出明确要求
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
  • 命令执行
网络行为
允许外网请求
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# Django Pytest Performance Suite

## Overview

Build trustworthy Django performance regression coverage for server-side surfaces. Prefer this skill when the goal is not a one-off benchmark, but a repeatable lane that can detect correctness drift, query regressions, and timing regressions over time.

## Workflow

1. Identify the performance surface.
   - Separate pure builders/read models from thin request/view wrappers.
   - Prefer benchmarking uncached server-side work first.
   - For Django UI work, cover read-only GET surfaces before thinking about browser rendering.
2. Create a separate performance lane.
   - Keep it out of the default unit-test path.
   - Use a dedicated settings module.
   - Require PostgreSQL instead of SQLite.
   - Use explicit commands such as `perf-test`, `perf-test-strict`, `perf-refresh-snapshots`, and `perf-accept-baseline`.
3. Make the run deterministic.
   - Seed fixed timestamps, UUIDs, slugs, and RNG.
   - Block background dispatch and outbound integration behavior.
   - Keep Celery eager if task code may be touched indirectly.
   - Use reusable databases when setup cost is large.
4. Validate correctness before timing.
   - Run an untimed pass first.
   - Normalize the result into stable JSON.
   - Compare it to a checked-in known-good artifact.
   - Count queries and assert a cap.
   - Only then run the benchmark timing pass.
5. Record and enforce budgets.
   - Keep checked-in timing budgets per surface and scenario.
   - Keep checked-in query caps per surface and scenario.
   - Generate machine-readable and human-readable reports for each run.
6. Protect coverage from drifting.
   - Keep a registry of read-only GET surfaces.
   - Add a structural test that fails when a new GET surface is unregistered.

## Design Rules

- Measure against PostgreSQL. SQLite timings are not useful for Django performance guardrails.
- Treat timing and correctness as separate concerns. A fast wrong result is still a regression.
- Pair query caps with timing budgets. Query counts are often the clearest early-warning signal.
- Benchmark both layers when possible:
  - builder/read-model cost for diagnosis
  - request/view wrapper cost for user-facing surfaces
- Prefer RequestFactory for request-surface measurements unless middleware behavior is the thing being tested.
- Keep large datasets realistic enough to trigger ORM and template-shaping costs that small fixtures hide.
- Store large-case correctness as summaries plus a payload hash instead of enormous snapshots.
- Keep baseline changes explicit. Refresh snapshots only for intentional output changes. Accept timing baselines only for intentional steady-state changes.

## Implementation Pattern

When building the suite, create these pieces:

- A dedicated Django settings module for performance runs.
- A `tests/performance/` package.
- Deterministic scenario seeders.
- Result normalizers that remove unstable fields.
- Snapshot assertions for correctness.
- Query-count capture helpers.
- A checked-in budget table.
- A report writer for latest results.
- A manual CI workflow that runs the strict lane and uploads artifacts.

For a detailed implementation checklist and the non-obvious stability techniques, read [references/patterns.md](references/patterns.md).

For a compact example of the expected plan/report shape, read [examples/performance-suite-plan.md](examples/performance-suite-plan.md).

## What To Avoid

- Do not mix performance tests into the default `pdm run test` or `pytest` path when they require heavy setup.
- Do not benchmark only tiny fixtures and assume the result generalizes.
- Do not rely on timing alone when large ORM regressions can be caught deterministically with query caps.
- Do not keep snapshots of raw HTML or full contexts if they include unstable values that will churn constantly.
- Do not silently update budgets after every run. That destroys the regression signal.
- Do not let new GET surfaces appear without performance-suite registration.

## Output Expectations

When the user asks for this kind of work, produce:

- the separate pytest lane
- deterministic scenarios sized to the product surface
- snapshot and query-count guards
- timing budgets and artifact reports
- run commands for local and CI usage
- documentation for refresh and baseline-accept workflows

If the repository already has ad hoc benchmarks, prefer folding them into the same lane rather than leaving multiple incompatible performance workflows in place.

---
> Source: [btfranklin/skills](https://github.com/btfranklin/skills) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-22 -->

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。