Performance review

Code Review - Performance Review

Identify performance bottlenecks, caching issues, and optimization opportunities. This skill is Phase 4 of the 6-phase code review methodology.

When to Use This Skill

Use this skill when you need to:

• Identify performance bottlenecks
• Optimize critical code paths
• Reduce resource consumption
• Improve response times
• Address scalability concerns
• Profile memory and CPU usage
• Evaluate caching strategies

Trigger phrases: "performance review", "bottleneck", "slow code", "optimize", "profiling", "latency", "throughput", "memory usage", "caching"

What This Skill Does

Performance Dimensions

Dimension	Metrics
Time	Response time, latency, throughput
Memory	Heap usage, allocations, leaks
CPU	Utilization, hot paths
I/O	Database queries, network calls
Concurrency	Threading, async efficiency
Caching	Hit rate, TTL, invalidation

Severity Classification

Level	Alias	Description
P0	CRITICAL	Production outages, severe degradation
P1	HIGH	Significant performance impact
P2	MEDIUM	Notable inefficiency
P3	LOW	Minor optimization opportunity

Instructions

Step 1: Profile Application

# Python
python -m cProfile -s cumtime script.py
py-spy record -o profile.svg -- python script.py

# JavaScript/Node.js
node --prof app.js
clinic doctor -- node app.js

# Java
java -XX:+FlightRecorder -XX:StartFlightRecording=duration=60s,filename=app.jfr App

Step 2: Identify Hot Paths

1. CPU Profiling

   • Functions with highest cumulative time
   • Frequent function calls
   • Complex algorithms

2. Memory Analysis

   • Large object allocations
   • Memory leaks
   • Garbage collection pressure
   • Event listener leaks (registered but never removed)

3. I/O Analysis

   • N+1 query patterns
   • Unoptimized database queries
   • Unnecessary network calls

Step 3: Common Anti-Patterns

Reference: references/code-quality-checklist.md (Performance & Caching section)

Anti-Pattern	Issue	Solution
N+1 Queries	Loop database calls	Batch queries, joins
Large Payloads	Excessive data transfer	Pagination, field selection
No Caching	Repeated computations	Add caching layer
Sync I/O	Blocking operations	Async/await
String Concatenation	Memory allocation in loops	StringBuilder/join
Missing Memoization	Same pure function called repeatedly	Add memoization
Over-fetching	SELECT * when only 2 columns needed	Select specific columns
No Pagination	Loading entire tables	LIMIT/OFFSET or cursor pagination

Step 4: Caching Strategy Analysis

Issue	Risk	Diagnostic
Missing cache	Repeated expensive computations	"Is this expensive operation called more than once with the same inputs?"
Cache without TTL	Stale data served indefinitely	"How long is cached data valid?"
No invalidation strategy	Cache and database drift	"When the source data changes, how is the cache updated?"
Key collisions	Different data overwriting each other	"Could two different inputs produce the same cache key?"
User data cached globally	Data leaks between users	"Does the cache key include user/tenant identity?"
Cache stampede	All caches expire simultaneously	"What happens when the cache expires under load?"

Step 5: Boundary Conditions Affecting Performance

Reference: references/code-quality-checklist.md (Boundary Conditions section)

Condition	Performance Impact
Unbounded collections	Lists/maps growing without limit, OOM risk
Large file loading	Reading entire files into memory instead of streaming
String concatenation in loops	O(n^2) memory allocation
Empty collection edge cases	.reduce() without initial value, .sort() on empty arrays

Step 6: Diagnostic Questions

Apply these questions to each module under review:

1. "What is the most expensive operation in the critical path? Can it be cached, batched, or deferred?"
2. "Are there any N+1 query patterns? (Loop that issues a query per iteration)"
3. "Is there any unbounded data structure that grows with input size?"
4. "For cached data, what is the TTL and invalidation strategy?"

Step 7: Document Findings

## Performance Finding

**File**: [path/to/file.py:42]
**Severity**: P1 (HIGH)
**Impact**: 500ms added latency per request
**Category**: Database Query

### Issue
N+1 query pattern in user loading

### Current Code
```python
users = User.query.all()
for user in users:
    orders = Order.query.filter_by(user_id=user.id).all()

Optimized Code

users = User.query.options(
    joinedload(User.orders)
).all()

Expected Improvement

• Latency: 500ms -> 50ms
• Database queries: N+1 -> 1

## Language-Specific Tools

### Python
- cProfile, py-spy, memory_profiler
- line_profiler, tracemalloc

### JavaScript
- Chrome DevTools, Node.js profiler
- clinic.js, 0x

### Java
- JFR, VisualVM, async-profiler
- JMH for benchmarks

### Go
- pprof, trace
- benchmarks

### C# / .NET
- dotTrace, dotMemory
- BenchmarkDotNet

## Quality Checklist

- [ ] Application profiled
- [ ] Hot paths identified
- [ ] Database queries analyzed (N+1, missing indexes, over-fetching)
- [ ] Memory usage reviewed (leaks, unbounded collections)
- [ ] Caching strategy evaluated (TTL, invalidation, stampede)
- [ ] Boundary conditions checked
- [ ] Diagnostic questions applied to each module
- [ ] Findings documented with metrics and severity (P0-P3)

## Related Skills

- `context-analysis` - Context understanding (Phase 1)
- `code-quality` - Code quality + SOLID review (Phase 2)
- `security-review` - Security analysis (Phase 3)
- `performance-testing` - Load testing
- `testing-review` - Test assessment (Phase 5)
- `final-report` - Consolidated report (Phase 6)

---

**Version**: 2.0.0
**Last Updated**: February 2026
**Based on**: DevAI-Hub code review methodology + code-review-expert


### Iterative Refinement Strategy
This skill is optimized for an iterative approach:
1. **Execute**: Perform the core steps defined above.
2. **Review**: Critically analyze the output (coverage, quality, completeness).
3. **Refine**: If targets aren't met, repeat the specific implementation steps with improved context.
4. **Loop**: Continue until the definition of done is satisfied.