Infrastructure

Managed Redis

Assistance-operated Redis for cache, sessions, queues, rate limits, and real-time data paths


Managed Redis is for teams using Redis as a production dependency: cache, session store, rate limiter, lightweight queue, pub/sub channel, or real-time data structure. Assistance operates the Redis runtime so your engineers can focus on application behavior and performance.

Best-fit use cases#

Use caseWhy Redis fits
Application cacheReduce database pressure and improve read latency for hot data
Session storageCentralize session state for horizontally scaled applications
Rate limitingTrack counters and windows with low-latency operations
Queues and background jobsUse lists, streams, or sorted sets for lightweight work coordination
Real-time featuresPub/sub, leaderboards, counters, and ephemeral state

What Assistance operates#

AreaIncluded managed service responsibility
ProvisioningRedis topology selection, memory sizing, network placement, secure defaults, and connection details
ReliabilitySentinel or Cluster design, replication health, failover procedure, persistence choices, backup options, and runbooks
PerformanceLatency, memory, command, slow log, eviction, and connection monitoring
MaintenancePatch planning, version lifecycle guidance, maintenance windows, and controlled restarts
SecurityTLS options, ACLs, password/secret rotation support, network isolation, and audit-friendly access practices
ObservabilityDashboards and alerts for memory pressure, latency, replication lag, persistence errors, and connection saturation
SupportSeverity-based support for platform incidents and escalation for covered production services

Ownership boundary#

ResponsibilityAssistance ownsCustomer owns
Runtime operationsConfigure, monitor, patch, scale, back up where applicable, and operate RedisClient usage, connection pooling, key design, and command patterns
DurabilityImplement agreed persistence and backup approachDecide which data is cache-only, recoverable, or business-critical
EvictionConfigure selected policy and monitor pressureDefine TTL strategy and acceptable behavior when memory fills
AvailabilityOperate Sentinel/Cluster/failover inside agreed topologyApplication retry behavior and idempotency for Redis-dependent flows
AccessACL model and rotation procedureUser approval, service credentials in applications, internal reviews

Deployment options#

OptionWhen to use it
Assistance physical serversDevelopment, staging, CI, internal tools, and predictable cache/session workloads
Customer cloud accountProduction workloads needing low latency to application services or cloud network controls
Cache-only deploymentPerformance optimization where data can be rebuilt from a source of truth
Durable Redis deploymentQueues, sessions, or stateful workflows that need persistence and recovery expectations

Topology choices#

TopologyUse when
Single nodeDevelopment, test, or non-critical cache-only workloads
SentinelYou need HA failover for a single primary with replicas and manageable memory size
Redis ClusterYou need horizontal scale, higher write throughput, or data larger than a single node comfortably supports
Read replicasYou need read scaling or isolation for reporting/analytics-style consumers

Reliability and support model#

TopicManaged Redis approach
AvailabilityDefined by topology and support tier; not all cache-only environments require production SLA language
PersistenceRDB, AOF, hybrid, or no persistence selected based on data semantics
RecoveryRestore expectations documented where persisted Redis is part of production recovery
PerformanceMemory pressure, latency, and eviction are monitored because Redis failure is often capacity-driven
ResponseP1 response targets scoped in the support plan; 24/7 critical response available for covered production services

Onboarding#

1. Workload classification#

We classify Redis usage: cache, session, rate limit, queue, pub/sub, streams, or mixed. This determines durability, topology, and operational risk.

2. Configuration design#

Assistance defines memory sizing, persistence, eviction policy, topology, network access, ACLs, monitoring, and support model.

3. Integration and migration#

We provision Redis, provide connection details, help migrate keys where needed, and validate application connection pooling and timeout behavior.

4. Operate#

We monitor latency, memory, command rates, replication, persistence, and failover signals. Capacity changes are planned before memory pressure becomes an outage.

Not included by default#

  • Designing every application cache key and invalidation path
  • Rewriting queue consumers or job processing logic
  • Guaranteeing durability for data declared cache-only
  • Unlimited memory, node count, or retention outside the plan
  • Debugging application-level race conditions unless scoped separately
  • Managed PostgreSQL — Source-of-truth relational database commonly paired with Redis
  • Managed MySQL — Relational backend for cache and session-heavy applications
  • Managed Prometheus — Metrics and alerting for cache hit ratio, latency, and saturation
  • SRE as a Service — Incident response, SLOs, and runbooks around Redis-dependent systems

Getting started#

Frequently asked questions#

Should we use Redis Sentinel or Redis Cluster? Sentinel is usually simpler for HA around one primary. Cluster is appropriate when data size or throughput needs horizontal partitioning. We choose based on workload and growth, not default complexity.

Can Redis be used for queues? Yes, but queue semantics must be explicit. Redis Streams can be a good fit for lightweight queues, while Kafka is usually better for durable replayable event streams.

Is Redis data always persisted? No. Cache-only workloads may disable persistence for performance. Sessions, queues, or stateful workflows usually need persistence and recovery expectations.

Who owns cache invalidation? Your application team owns invalidation logic and TTL choices. Assistance monitors and operates the Redis platform and can advise on safe patterns.

What happens when memory fills? The agreed eviction policy applies. We monitor memory pressure and recommend capacity or TTL changes before Redis reaches unsafe saturation.