Skip to main content

ADR-003: Caching Strategy

Status​

Accepted

Date​

2026-01-05

Context​

A caching layer is required to:

  • Reduce database load for frequently accessed data
  • Store session and rate limit data
  • Provide pub/sub for real-time notifications
  • Implement distributed locks for workflow coordination

Options Considered​

  1. Redis - In-memory, data structures, pub/sub
  2. Memcached - Simple key-value, high performance
  3. Application-level caching - In-process, no network

Decision​

Redis 7 was selected for caching and coordination.

Rationale​

Data Structures​

  • Hashes for session objects
  • Sorted sets for rate limiting windows
  • Lists for queues
  • Streams for event processing

Features​

  • Pub/sub for real-time finding notifications
  • Lua scripting for atomic operations (rate limiting)
  • TTL support with millisecond precision
  • Cluster mode for horizontal scaling

Persistence​

  • Optional RDB/AOF persistence for recovery
  • Not required for pure caching, but useful for sessions

Why Not Memcached​

  • No pub/sub support
  • Limited data structures
  • No persistence option
  • No Lua scripting

Cache Strategy​

TTL by Data Type​

Data TypeTTLReason
Compliance frameworks24hRarely changes
Finding by ID1hMay be updated
Deduplication keys7dLong-term lookup
User sessions8hSecurity policy
Rate limit counters1minPer-window
AI responses30minCost optimization

Cache Invalidation​

  1. Write-through: Update cache on write
  2. Event-driven: Invalidate on compliance framework updates
  3. TTL-based: Automatic expiration for most data

Consequences​

Positive​

  • Reduced database queries by 80%+
  • Sub-millisecond cache reads
  • Built-in rate limiting support
  • Real-time notifications via pub/sub

Negative​

  • Additional infrastructure to manage
  • Cache invalidation complexity
  • Memory costs (~$0.017/GB/hour on ElastiCache)

Mitigations​

  • Use ElastiCache Serverless for auto-scaling
  • Implement cache-aside pattern with fallback
  • Monitor cache hit rates
  • ADR-002: Database Selection
  • ADR-005: Rate Limiting Implementation