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).
소스 정보
- 저장소
- districtr/districtr-v2
- 최근 소스 활동
- 2026년 9월 18일 14:42
- 감지된 SKILL.md 언어
- 영어
- 스타
- 6
- 포크
- 3
설치 방법
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
소스 파일 검토
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
SKILL.md 표시 중
SKILL.md
소스 지침 · 읽기 전용 미리보기- 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.
GitHub에서 보기