performance-optimization

Performance Optimization — Aprender Sistema

Measure before optimizing. Profile first, identify the actual bottleneck, fix it, measure again. Don't scatter React.memo or select_related on a hunch.

AS v2 has active perf epics: #777 (grid phases), #866 (ACID). PR #1160 cut monthly grid p95 by -69% (phase 2).

Workflow

1. MEASURE  → Establish baseline (numbers, not feelings)
2. IDENTIFY → Find the actual bottleneck (profile, don't guess)
3. FIX      → Address that specific bottleneck
4. VERIFY   → Measure again, confirm improvement vs baseline
5. GUARD    → Add a test/monitor to prevent regression

Done when: before/after numbers are recorded, the targets below are met, and a guard exists.

Where to look

• Backend (query profiling, N+1, indexes, aggregations, Celery, Redis cache): reference/backend.md
• Frontend (Lighthouse, bundle analysis, re-renders, Antd, code splitting, images, caching): reference/frontend.md

Targets (CI enforced)

Backend (Django/DRF)

Metric	Target	Where
API p95 latency	≤ 200ms	gunicorn logs
DB queries per request	≤ 10	django-debug-toolbar (dev)
Cache hit rate (Redis)	≥ 70%	availability queries
Monthly grid p95	≤ baseline −69%	after #777 phase 2

Frontend (React/Vite), from v2/frontend/lighthouserc.cjs

Metric	Target	Severity
Lighthouse Performance	≥ 70	error (fails CI)
Lighthouse Accessibility	≥ 90	error (fails CI)
LCP	≤ 2.5s	error
CLS	≤ 0.1	error
Total Blocking Time	≤ 300ms	error
Initial JS bundle (gzipped)	≤ 250KB	Antd is the main weight
manualChunks in vite.config	none	breaks dep ordering (prod crash)

Current Lighthouse: Desktop 95, Mobile 79 (mobile bottleneck = antd bundle).

Common Rationalizations

Rationalization	Reality
"It's fast on my machine"	Profile on prod-like conditions (docker + realistic data volume)
"Users won't notice"	RUM from real users. 100ms matters.
"Framework handles perf"	Django/React prevent some issues, not N+1 or bundle bloat
"We'll optimize later"	Perf debt compounds. Fix N+1 and bundle issues now.
"The DB is fast enough"	Check query count and duration per request

Verification

After a perf change:

• [ ] Before/after measurements with specific numbers
• [ ] Bottleneck identified (not guessed)
• [ ] Core Web Vitals in target range
• [ ] Bundle size change reviewed
• [ ] No new N+1 queries
• [ ] Lighthouse CI passes
• [ ] Existing tests still pass

References

• Epic #777: Monthly grid optimization · Epic #866: ACID/deadlock handling
• PR #1160: Phase 2 grid p95 -69%
• Memory: Lighthouse scores · feedback_no_manual_chunks.md · feedback_react18_fetchpriority.md