performance-memory
Investigating a slow endpoint or high memory, or making a change that could affect backend memory usage, request latency, or where a heavy computation runs (database, server process, or browser worker).
来源信息
- 仓库
- districtr/districtr-v2
- 最近来源活动
- 2026年9月18日 14:42
- 检测到的 SKILL.md 语言
- 英语
- 星标
- 6
- 分支
- 3
安装方式
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
检查来源文件
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
正在显示 SKILL.md
SKILL.md
来源说明 · 只读预览- name
- performance-memory
- description
- Investigating a slow endpoint or high memory, or making a change that could affect backend memory usage, request latency, or where a heavy computation runs (database, server process, or browser worker).
- user-invocable
- false
# Performance & Memory
## Constraints
- **A district graph is the expensive resource in this system.** Every metric that needs adjacency (contiguity, compactness, cut edges, `assigned_units`) loads a full block/VTD-level graph for the relevant state. Anything that changes how many copies of a graph exist per process, or how many processes exist, multiplies memory straight through. Backend task sizing (`backendMemory` in `infra/config.ts`) and the graph LRU cache (`_GRAPH_CACHE_MAX_SIZE`, `backend/app/evaluation/graph_loader.py`) are coupled — check both before resizing either.
## Where the rest lives
The incident history behind this constraint — the ~7GB cache-growth fix (PR #540), the mmap-shared graph direction (PR #721), the assignments departitioning under lock convoy (PR #625), and the computation-placement fixes (PR #550, #470) — is in `docs/decisions.md`, each entry dated and PR-anchored.
在 GitHub 查看