Skip to main content

Redis

Redis provides lowest latency for high-throughput scenarios where some durability trade-offs are acceptable.

Why Redis​

StrengthBenefit
Sub-millisecond latencyFastest event writes
Horizontal scalingRedis Cluster for partitioning
Memory-firstNo disk I/O on hot path
Pub/Sub nativeCan double as event bus

Trade-offs​

ConcernConsideration
DurabilityConfigurable (AOF, RDB, none)
Memory costEvents live in RAM until evicted
ComplexityCluster mode adds operational overhead

Configuration​

[storage]
backend = "redis"

[storage.redis]
url = "redis://localhost:6379"
# For cluster mode
# urls = ["redis://node1:6379", "redis://node2:6379"]

Environment Variables​

export REDIS_URL="redis://localhost:6379"
export STORAGE_BACKEND="redis"

Data Model​

Redis stores events in sorted sets keyed by aggregate:

events:{domain}:{edition}:{root}
  └── ZADD with sequence as score
      └── Member: serialized event

positions:{handler}:{domain}:{edition}:{root}
  └── String: sequence number

snapshots:{domain}:{edition}:{root}
  └── Hash: { sequence, state }

Durability Options​

AOF (Append-Only File)​

# redis.conf
appendonly yes
appendfsync everysec  # fsync every second (default)
# appendfsync always  # fsync every write (slower, safest)

RDB Snapshots​

# redis.conf
save 900 1    # Save if 1 key changed in 900 seconds
save 300 10   # Save if 10 keys changed in 300 seconds
save 60 10000 # Save if 10000 keys changed in 60 seconds

Production Recommendation​

Use both AOF and RDB for durability with fast recovery:

appendonly yes
appendfsync everysec
save 900 1

Helm Deployment​

# values.yaml
storage:
  backend: redis

redis:
  enabled: true
  host: redis.cache.svc.cluster.local
  port: 6379
  credentials:
    secretName: redis-credentials
    passwordKey: password

Testing​

# Run Redis tests (requires testcontainers)
cargo test --test storage_redis --features redis

# Requires podman socket
systemctl --user start podman.socket

When to Use Redis​

  • Gaming/real-time — Latency-critical applications
  • Ephemeral data — Acceptable to lose on crash
  • Caching layer — In front of durable storage
  • Development — Fast iteration without disk I/O

Next Steps​