agentsclimarketplace

Caching strategies

Skill pluginagentmarketplace/custom-plugin-system-design/skills/caching-strategies

Production-grade caching strategies skill for Redis patterns, CDN configuration, cache invalidation, and performance optimizationFrom its SKILL.md

Install
npx -y skills add pluginagentmarketplace/custom-plugin-system-design --skill caching-strategies

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

9.9 KB, as published. Nobody here has run it

Caching Strategies Skill

Purpose: Atomic skill for caching architecture with comprehensive invalidation patterns and performance metrics.

Skill Identity

AttributeValue
ScopeRedis, CDN, Cache Invalidation
ResponsibilitySingle: Caching layer design and optimization
InvocationSkill("caching-strategies")

Parameter Schema

Input Validation

parameters:
  caching_context:
    type: object
    required: true
    properties:
      use_case:
        type: string
        enum: [session, api_response, database, static_assets, compute]
        required: true
      data_profile:
        type: object
        required: true
        properties:
          size_per_item: { type: string, pattern: "^\\d+[KMGB]?B?$" }
          total_items: { type: integer, minimum: 1 }
          update_frequency: { type: string, enum: [real_time, seconds, minutes, hours, days] }
          access_pattern: { type: string, enum: [uniform, hot_cold, temporal] }
      requirements:
        type: object
        properties:
          hit_rate_target: { type: number, minimum: 0, maximum: 100 }
          max_latency_ms: { type: integer, minimum: 1 }
          consistency: { type: string, enum: [strict, eventual] }
          budget_monthly: { type: string }

validation_rules:
  - name: "hit_rate_feasibility"
    rule: "hit_rate_target <= 99.9"
    error: "100% hit rate is not achievable in practice"
  - name: "memory_estimate"
    rule: "size_per_item * total_items <= available_memory"
    warning: "May require cache eviction or sharding"

Output Schema

output:
  type: object
  properties:
    architecture:
      type: object
      properties:
        layers: { type: array }
        technology: { type: string }
        topology: { type: string }
    configuration:
      type: object
      properties:
        memory_allocation: { type: string }
        eviction_policy: { type: string }
        ttl_strategy: { type: object }
        connection_pool: { type: object }
    invalidation:
      type: object
      properties:
        strategy: { type: string }
        triggers: { type: array }
        implementation: { type: string }
    metrics:
      type: object
      properties:
        expected_hit_rate: { type: number }
        memory_usage: { type: string }
        latency_p99: { type: string }

Core Patterns

Cache Layers

L1: Application Memory
├── Technology: Caffeine, Guava
├── Latency: ~0.1ms
├── Size: MB range
├── TTL: Seconds
└── Use: Hot data, thread-local

L2: Distributed Cache
├── Technology: Redis, Memcached
├── Latency: 1-5ms
├── Size: GB-TB range
├── TTL: Minutes to hours
└── Use: Shared state, sessions

L3: CDN Edge
├── Technology: CloudFront, Fastly
├── Latency: 5-50ms (network)
├── Size: Unlimited
├── TTL: Hours to days
└── Use: Static assets, API responses

L4: Database Cache
├── Technology: Query cache, buffer pool
├── Latency: ~1ms
├── Size: GB range
├── TTL: Until invalidated
└── Use: Query results

Cache Patterns

Cache-Aside (Lazy Loading):
├── Read: Check cache → Miss → DB → Store → Return
├── Write: Update DB → Invalidate cache
├── Pros: Simple, resilient to cache failure
├── Cons: Cache miss penalty, stale on DB update
└── Use: Read-heavy, tolerance for staleness

Write-Through:
├── Write: Update cache + DB atomically
├── Read: Always from cache
├── Pros: Cache always fresh
├── Cons: Write latency, complexity
└── Use: Read-after-write needed

Write-Behind:
├── Write: Update cache → Async DB write
├── Pros: Low write latency
├── Cons: Data loss risk, complexity
└── Use: Write-heavy, acceptable loss

Read-Through:
├── Read: Cache handles DB fetch on miss
├── Pros: Simplified application
├── Cons: Cache dependency
└── Use: Predictable access patterns

Invalidation Strategies

TTL-Based:
├── Simple time expiry
├── Formula: TTL = max_acceptable_staleness
├── Jitter: TTL * (1 + random(-0.1, 0.1))
└── Prevents: Thundering herd

Event-Based:
├── Invalidate on data change
├── Implementation: CDC, Pub/Sub
├── Latency: Near real-time
└── Complexity: Event system required

Version-Based:
├── Key: user:{id}:v{version}
├── Bump version on change
├── Old versions expire naturally
└── Benefit: No explicit invalidation

Tag-Based:
├── Associate keys with tags
├── Invalidate by tag
├── Example: Tag "product:123" on all related
└── Use: Related data groups

Retry Logic

Cache Operation Retry

retry_config:
  cache_read:
    max_attempts: 2
    timeout_ms: 50
    on_failure: proceed_without_cache

  cache_write:
    max_attempts: 3
    timeout_ms: 100
    on_failure: log_and_continue

  redis_connection:
    max_attempts: 5
    initial_delay_ms: 100
    max_delay_ms: 5000
    multiplier: 2.0

  circuit_breaker:
    failure_threshold: 5
    reset_timeout_seconds: 30
    half_open_requests: 1

Logging & Observability

Log Format

log_schema:
  level: { type: string }
  timestamp: { type: string, format: ISO8601 }
  skill: { type: string, value: "caching-strategies" }
  event:
    type: string
    enum:
      - cache_hit
      - cache_miss
      - cache_set
      - cache_invalidate
      - cache_evict
      - ttl_expired
      - circuit_open
  context:
    type: object
    properties:
      key: { type: string }
      ttl_seconds: { type: integer }
      latency_ms: { type: number }
      size_bytes: { type: integer }

example:
  level: INFO
  event: cache_hit
  context:
    key: "user:123:profile"
    latency_ms: 0.5
    size_bytes: 1024

Metrics

metrics:
  - name: cache_requests_total
    type: counter
    labels: [operation, result]  # hit, miss, error

  - name: cache_latency_seconds
    type: histogram
    labels: [operation]
    buckets: [0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05]

  - name: cache_memory_bytes
    type: gauge
    labels: [cache_name]

  - name: cache_evictions_total
    type: counter
    labels: [policy]

  - name: cache_hit_ratio
    type: gauge
    labels: [cache_name]

Troubleshooting

Common Issues

IssueCauseResolution
Low hit rateTTL too shortIncrease TTL, analyze patterns
High latencyCache miss + DBWarm cache, optimize DB
Memory pressureToo much dataIncrease memory, evict
Stale dataTTL mismatchEvent-based invalidation
Thundering herdMass expiryJittered TTL, singleflight
Hot keyPopularity skewReplicate, local cache

Debug Checklist

□ Hit rate measured (>90% target)?
□ Memory usage within limits?
□ Eviction rate acceptable?
□ Latency p99 within SLA?
□ Invalidation working?
□ Cluster health OK?
□ Connection pool sized right?

Unit Test Templates

Cache Configuration Tests

# test_caching_strategies.py

def test_valid_caching_context():
    params = {
        "caching_context": {
            "use_case": "session",
            "data_profile": {
                "size_per_item": "1KB",
                "total_items": 1000000,
                "update_frequency": "minutes",
                "access_pattern": "hot_cold"
            },
            "requirements": {
                "hit_rate_target": 99,
                "max_latency_ms": 5,
                "consistency": "eventual"
            }
        }
    }
    result = validate_parameters(params)
    assert result.valid == True

def test_memory_estimation():
    result = estimate_memory(
        size_per_item="1KB",
        total_items=1000000,
        overhead_factor=1.5  # Redis overhead
    )
    assert result.total == "1.5GB"
    assert result.recommended_allocation == "2GB"  # 25% buffer

def test_ttl_with_jitter():
    base_ttl = 3600
    jittered = apply_jitter(base_ttl, factor=0.1)
    assert 3240 <= jittered <= 3960  # ±10%

def test_hit_rate_infeasibility():
    params = {
        "caching_context": {
            "requirements": {
                "hit_rate_target": 100  # Impossible
            }
        }
    }
    result = validate_parameters(params)
    assert result.valid == False
    assert "not achievable" in result.errors[0]

Invalidation Tests

def test_ttl_invalidation():
    cache = MockCache()
    cache.set("key", "value", ttl=60)

    assert cache.get("key") == "value"
    advance_time(61)
    assert cache.get("key") is None

def test_event_based_invalidation():
    cache = MockCache()
    cache.set("user:123", {"name": "Alice"})

    # Simulate update event
    event = {"type": "user_updated", "id": 123}
    process_invalidation_event(cache, event)

    assert cache.get("user:123") is None

def test_tag_based_invalidation():
    cache = MockCache()
    cache.set("product:123:details", data, tags=["product:123"])
    cache.set("product:123:reviews", reviews, tags=["product:123"])

    cache.invalidate_by_tag("product:123")

    assert cache.get("product:123:details") is None
    assert cache.get("product:123:reviews") is None

def test_thundering_herd_prevention():
    cache = MockCache()

    # Simulate concurrent requests for same key
    requests = [fetch_with_singleflight(cache, "hot_key") for _ in range(100)]
    results = await asyncio.gather(*requests)

    # Only one DB call should have been made
    assert cache.db_calls == 1
    assert all(r == results[0] for r in results)

Version History

VersionDateChanges
2.0.02025-01Production-grade rewrite with invalidation patterns
1.0.02024-12Initial release

Keep looking

Skills are one crate of 610,220. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.