AutoGineer
Generated banner for Amazon Elasticache
Generated logo for Amazon Elasticache

Amazon Elasticache

AWSOfficialPublished by AWS

  • Skill
  • Tool integration
  • Cloud and DevOps
  • Apache-2.0
  • Medium risk
  • Has scripts
Download ZIP

From the author

Activate when developers have latent caching needs: slow API responses, database read bottlenecks, DynamoDB throttling or cost, RDS/Aurora scaling pressure, Bedrock latency or cost, or adding a cache; activate when working with Redis, Valkey, Memcached, or any in-memory data store, cache-aside patterns, session stores, rate limiting, leaderboards, counters, streams, queues, pub/sub, distributed locks, feature flags, shopping carts, or other caching strategies. Activate for GenAI and ML retrieval: vector similarity search for low-latency retrieval, semantic caching, RAG, LLM response caching, embedding stores, AI agent memory, recommendation, personalization. Activate for ElastiCache lifecycle: provisioning (serverless or node-based), engine selection, CloudFormation/CDK/Terraform IaC, VPC connectivity, TLS, RBAC, IAM auth, Global Datastore, monitoring, troubleshooting, cost optimization, and migration from self-managed Redis. Do not trigger for browser caches, CDN/CloudFront, HTTP Cache-Control, CPU caches.
— the description by AWS

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The Amazon ElastiCache skill provides a structured, modular approach to managing in-memory data stores on AWS. It guides developers through the entire lifecycle of ElastiCache and Valkey, from initial requirements gathering and provisioning to complex data modeling, AI-driven vector search, and operational monitoring. The skill enforces best practices for connection safety, security, and infrastructure-as-code, ensuring reliable performance for caching, session management, and GenAI workloads.

What it does

  • Assesses workload requirements to recommend optimal ElastiCache configurations.
  • Automates provisioning and connectivity setup for serverless and node-based clusters.
  • Provides implementation patterns for common use cases like cache-aside, session stores, and rate limiting.
  • Supports advanced AI workloads including semantic caching, RAG, and vector similarity search.
  • Delivers operational diagnostics for performance tuning, cost optimization, and troubleshooting.
  • Manages migration paths from self-managed Redis or between different deployment models.

When to use it

  • When you need to reduce database load or improve API response times with an in-memory cache.
  • When implementing AI agent memory, RAG retrieval, or semantic caching for LLM applications.
  • When migrating existing Redis workloads to a managed AWS environment.
  • When troubleshooting performance issues like high latency, replication lag, or memory pressure.
  • When setting up secure, production-ready caching infrastructure using IaC.

Advantages

  • Standardizes deployment and operational patterns across development teams.
  • Ensures compliance with security best practices like TLS and IAM authentication.
  • Reduces configuration errors through automated input validation and tagging.
  • Provides clear guidance on engine selection and version-specific feature availability.
  • Simplifies complex tasks like cross-region migration and vector search setup.

Usage examples

“I need to speed up my RDS database. Should I use ElastiCache?”
The agent will initiate the requirements sub-skill to analyze your workload, stack, and latency needs, then provide a recommendation on whether ElastiCache is the right fit.
“How do I set up a Valkey cluster for vector search?”
The agent will route to the genai sub-skill, verify you are using a node-based Valkey 8.2+ cluster, and provide the necessary configuration and code to enable vector similarity search.
“My cache hit rate is low and latency is spiking. Can you help me troubleshoot?”
The agent will activate the monitoring sub-skill to analyze your metrics, identify potential hot keys or memory imbalance, and suggest specific remediation steps.
“Help me migrate my self-managed Redis to ElastiCache.”
The agent will guide you through the migration sub-skill, performing preflight checks and validating your source configuration for a zero-downtime cutover.

Install

  1. Download the skill as a ZIP:

    https://mp-staging.autogineer.com/agent-skills/aws/amazon-elasticache/download

  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip amazon-elasticache.zip -d .agents/skills

Documentation by AWS

ElastiCache

A modular ElastiCache toolkit organized as a registry of sub-skills. Each sub-skill handles one domain of ElastiCache work. The router below matches user intent to the right sub-skill, then loads only the references needed for that sub-skill.

How this skill works

  1. Match the user's request against the semantic categories in the registry below. Match on meaning, not exact wording ("help me figure out which data structures to use" matches data-modeling even without the word "pattern").

  2. Disambiguation: If the user's intent matches multiple sub-skills, apply these rules in order:

    • If .elasticache/requirements.json exists with infrastructure.endpoint set, prefer monitoring or data-modeling (the user has an existing cache).

    • If no cache exists (no requirements.json or no endpoint), prefer requirements.

    • If still ambiguous, ask one clarifying question: "Are you looking to set up something new, or troubleshoot something existing?"

  3. Check the Guardrails section before recommending an engine or deployment model.

  4. Read references/{sub-skill-id}/instructions.md for the matched sub-skill. If the file is not found at a relative path, check your prompt or environment for the skill directory absolute path and retry with {skill-directory}/references/{sub-skill-id}/instructions.md.

  5. If the request spans multiple sub-skills, execute them in pipeline order.

  6. If a sub-skill requires upstream context (engine, deployment model, endpoint) not yet in session memory, route to the upstream sub-skill first.

  7. If no sub-skill matches, activate requirements first.

  8. If a script or CLI call fails, show the error to the user and suggest a specific fix before retrying.

Sub-skill registry

Each entry has: an ID (directory name under references/), a domain description, semantic categories for matching, and upstream/downstream dependencies.

IDNameDomainSemantic CategoriesUpstreamDownstream
requirementsSolution FitGathers workload, stack, scale, latency, persistence, and budget through workspace scan + structured interview. Decides whether ElastiCache is the right service and hands off with a routing recommendation.I need a cache, speed up my app, reduce database load, lower Bedrock cost, should I use ElastiCache, what's best for my workload, evaluating cache options, ElastiCache vs X, Valkey vs X, vague new workload—setup, data-modeling, genai, monitoring, migration
setupCreate and ConnectProvisioning, connectivity, security, authentication, IaC, deployment choice. Gets the user to a working cache with least friction. Covers engine selection, serverless vs node-based, VPC, TLS, RBAC/IAM, jump-host/SSM tunnels, CLI/SDK/CFN/CDK/Terraform starters.create a cache, set up ElastiCache, provision, Valkey cluster, connect Lambda/ECS/EKS/EC2, VPC, security groups, TLS, RBAC, IAM auth, jump host, SSM tunnel, CloudFormation, CDK, Terraform, engine selection, serverless vs node-based, backup, snapshot, restore, exportrequirements (optional)data-modeling, genai, monitoring
data-modelingApplication PatternsPicks data structures, key schema, TTL strategy, invalidation approach, and client code for non-AI patterns: cache-aside, session store, rate limiting, leaderboards, counters, pub/sub, streams, shopping carts, job queues, activity feeds.session store, rate limiting, leaderboard, cache-aside, query caching, counters, streams, pub/sub, shopping cart, job queue, activity feed, key schema, TTL, invalidation, data structuressetup (cache must exist)monitoring
genaiAI and Vector WorkloadsClassifies request into Mode 1 (plain cache), Mode 2 (semantic response cache), or Mode 3 (full vector search). Selects Valkey and forces node-based Valkey 8.2 or above (recommend 9.0) when server-side vector similarity is needed. Covers semantic caching, agent memory, RAG retrieval, recommendation, personalization, conversation/session persistence for AI agents, and framework wiring (Strands, mem0, LangChain).semantic cache, RAG, agent memory, conversational memory, vector search, embeddings, recommendation, personalization, Bedrock latency, Bedrock cost, LLM caching, Strands, mem0, LangChain, conversation history, AI session store, embedding provider, framework integrationsetup (cache must exist)monitoring
monitoringOperate and ObserveDiagnoses performance, cost, and reliability using metrics first, then recommends the smallest change. Covers dashboards, alarms, log delivery, cost reporting, event routing, troubleshooting high CPU / memory / replication lag / connection spikes / low hit rate / hot keys / big keys / slot imbalance / latency spike root cause.cache is slow, cost too high, hit rate low, high CPU, memory pressure, replication lag, connection spikes, dashboards, alarms, CloudWatch, cost comparison, troubleshoot, hot key, uneven shard load, one node pinned, big key, memory bloat, which key is biggest, keyspace distribution, prefix analysis, cost attribution by tenant, memory imbalance, one shard full, slot memory skew, latency spike, slow command incident, root cause for latency bump—setup, migration
migrationEngine and Platform MigrationSelects the migration path and sequences preflight, validation, cutover, and rollback. Covers self-managed Redis → ElastiCache, Redis OSS → Valkey, node-based ↔ serverless, version upgrades. Hard validate-before-migrate gate.migrate, Redis OSS to Valkey, self-managed to ElastiCache, node-based to serverless, serverless to node-based, engine upgrade, version upgrade, zero-downtime cutover, rollback—setup, monitoring

Pipeline order

Sub-skills run independently, but common multi-step journeys follow these pipelines:

  • requirements → setup → (data-modeling | genai) → monitoring

  • migration → setup → monitoring

  • monitoring → setup | migration (if metrics indicate)

State handoff: requirements.json

.elasticache/requirements.json is the single source of truth for cross-sub-skill state. Each sub-skill reads it at start and writes its section after completing work. Read before writing; merge, do not overwrite.

SectionOwnerKey fields
top-levelrequirementsengine, deployment_model, region, runtime, patterns, use_case, vpc_id, subnet_ids, security_group_ids
infrastructuresetupcache_name, resource_id, engine_version, topology, endpoint, port, auth_model, tls, client_library, execution_path, access_mode, tunnel_instance_id, embedding_provider, embedding_model, embedding_dim, embedding_module
genaigenaimode, mode_2_path, framework
migrationmigrationsource_type, source_host, migration_path, cutover_status

Ownership note: deployment_model is set by requirements during initial interview. migration may update it after an engine or deployment model switch (e.g., node-based to serverless).

requirements.json should include "schema_version": 1 and "last_updated": "<ISO timestamp>" at the top level. Every sub-skill that writes to requirements.json must update last_updated. If last_updated is older than 7 days, warn the user that cached state may be stale.

requirements.json tracks one active cache. If the user works with multiple caches in the same project, confirm which cache is active before reading or writing state.

When a sub-skill needs upstream context (engine, endpoint, auth model), check requirements.json first. If the field is null or the file does not exist, route to the upstream sub-skill.

Global rules (apply to every sub-skill)

  1. Execution path. Use AWS CLI, SDK (boto3), CloudFormation, or CDK as the primary path for control-plane work. Use valkey-py as the primary path for data-plane work.

  2. Response depth. Summary (2-3 sentences) for "should I" or "which" questions. Standard (recommendation + config + code + next steps) by default. Expert (full decision matrix with alternatives, cost, security caveats) for "why" or "compare all" questions. Escalate on user request; never downgrade unprompted.

  3. Session memory. Track region, VPC, engine, deployment model, auth model, compute runtime, and language. Carry forward across sub-skills. Do not re-ask. If the user overrides a value, update it everywhere. Inferred values (from workspace scan or IaC) must be re-confirmed before high-risk decisions (engine, deployment model, security posture); low-risk inferences (language, framework, region) can be used as defaults silently.

  4. Source priority. Always answer from skill-local files first (sub-skill references, then scripts/). Do not fetch external documentation or web search unless the local files cannot answer the query. When local files are insufficient, fall back to official AWS docs: https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/ for features and https://aws.amazon.com/elasticache/pricing/ for pricing. Never invent price points or version constraints. If the user references a Valkey or Redis version, feature, or pricing tier not covered in local files, fall back to https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/ before answering. Do not extrapolate from local content that may be outdated.

  5. Freshness disclaimer. When outputting pricing, version constraints, or feature availability, include a one-line disclaimer: "For current pricing see https://aws.amazon.com/elasticache/pricing/. For current feature availability see https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/."

  6. Connection safety. Never create a Valkey/Redis client at module level (top of file, import time). Initialize connections inside a function or on first use. Module-level connections crash applications that import the module before the cache is reachable.

  7. Code environment. Use python3 (not python) in all commands and shebangs. Do not generate pip install or pytest commands in application code. Prefer the agent's file-writing tool over shell redirection for writing files.

  8. Input validation (write operations only). Before running any create, modify, or delete CLI command or SDK call that uses user-provided values (cache name, region, subnet IDs, security group IDs, VPC ID, KMS key, snapshot name), validate them with scripts/input_validator.py. Run python3 scripts/input_validator.py cache_name=<val> region=<val> ... or import validate_all as a library. Skip validation for read-only operations (describe, list) where bad input is harmless.

  9. Resource tagging (always apply on resource creation). When creating any cache (create-serverless-cache, create-replication-group, create-cache-cluster), ALWAYS include these tags:

    --tags Key=managed_by,Value=aws-skills Key=skill,Value=elasticache Key=skill_version,Value=1.0.0 Key=created_by,Value=elasticache-skill Key=generation_model,Value={your-model-id}

    Example: --tags Key=managed_by,Value=aws-skills Key=skill,Value=elasticache Key=skill_version,Value=1.0.0 Key=created_by,Value=elasticache-skill Key=generation_model,Value=claude-sonnet-4-20250514

    Set generation_model to your own model identifier. Include these tags even if the user does not mention tagging. If the user provides additional tags, append these to their tags rather than replacing them.

Reference loading

Load additional references only when the current turn's answer requires them.

On-demand pointers (not preloaded; fetch when the trigger applies):

  • references/shared-ux/production-readiness.md — when the user asks if their cache is ready for production, or after setup completes and the user wants to go to production

  • references/shared-ux/action-safety.md — before any destructive action (risk levels, never-auto-execute list)

  • references/shared-ux/error-remediation.md — when the user hits a specific ElastiCache error code (MOVED, CROSSSLOT, CLUSTERDOWN, MULTI/EXEC+IAM, etc.)

  • references/shared-foundation/boundary-doc.md — when the user asks what this skill covers

  • references/shared-foundation/attribution.md — when generating CLI commands, SDK code, or IaC templates

  • references/shared-foundation/architecture-diagrams.md — when the user asks for architecture diagrams or visual reference

  • references/shared-runtime/lambda.md — when connecting from Lambda (cold start gotchas, IAM auth code, lazy init)

  • references/shared-runtime/ecs.md — when connecting from ECS (SIGTERM shutdown, connection pool drain, task definition)

  • references/shared-runtime/eks.md — when connecting from EKS (IRSA, service mesh bypass, SecurityGroupPolicy CRD)

  • references/shared-runtime/api-gateway.md — when integrating with API Gateway (no direct path, caching layers comparison)

  • references/shared-runtime/rds-acceleration.md — when caching RDS/Aurora queries (thundering herd, stampede protection, invalidation)

  • references/shared-runtime/secret-injection.md — when the user asks about credential management per compute platform

  • references/shared-security/encryption-defaults.md — when adding encryption to an existing unencrypted cluster (TLS two-step migration, at-rest immutability)

  • references/shared-security/config-guardrails.md — when the user wants continuous compliance monitoring (AWS Config rules, custom Lambda rules)

  • references/shared-security/vpc-patterns.md — when debugging port/security-group issues (port 6380 serverless reader, anti-patterns)

Folder convention: references/ contains 10 folders. 6 match the sub-skills (requirements, setup, data-modeling, genai, monitoring, migration) and are routing destinations. The 4 shared-* folders (shared-foundation, shared-ux, shared-security, shared-runtime) are cross-cutting material loaded on demand, not routing destinations.

Guardrails

PriorityRule
CRITICALVector search MUST use node-based Valkey 8.2 or above. Serverless does NOT support vector search. Never suggest serverless for vector search. Apply this regardless of which sub-skill activates.
CRITICALDo not invent price points or version constraints. Use scripts/price_calculator.py and current AWS docs when precision matters.
HIGHDo not recommend Memcached when the user needs persistence, replication, RBAC or IAM auth, sorted sets, streams, pub/sub, or vector search.
HIGHDo not assume local laptop access works directly. ElastiCache is VPC-centric; explain VPC, tunnel, or jump-host access when needed.
STANDARDDo not trigger on every generic Redis mention. Trigger when the user is clearly asking about AWS, managed caching, migration, connectivity, pricing, operations, or AWS service integration.
STANDARDFor ambiguous "cache" requests inside AWS contexts, activate this skill and start with requirements.

Product truths

  • ElastiCache Serverless deploys in under a minute and removes infrastructure management.

  • Valkey serverless pricing is 33% lower than other supported engines; node-based Valkey pricing is 20% lower.

  • Serverless caches have in-transit encryption always enabled (cannot be disabled).

  • IAM auth is available for all ElastiCache Valkey versions (7.2 is the baseline Valkey version on ElastiCache) and Redis OSS 7.0+.

  • Valkey version ladder: 7.2 (baseline), 8.0 (20% more data per node (capacity improvement), per-slot metrics), 8.1 (Bloom filters, COMMANDLOG, SET IFEQ, 20% less memory via new hash table (efficiency improvement)), 8.2 (vector search), 9.0 (recommended default for new clusters). Recommend Valkey 9.0 for new clusters unless a specific feature dictates otherwise.

  • Vector search is available for Valkey 8.2 or above on node-based clusters (recommend 9.0).

  • Global Datastore is available for node-based clusters only. It does not support IPv6 or Local Zones. Global Datastore supports AUTH and RBAC. Cross-region failover must be promoted manually (no autofailover across regions). At-rest encryption must be enabled on all clusters in the Global Datastore, but each cluster can use a separate KMS key per region.

  • Online migration from self-managed Redis to ElastiCache requires: (source) AUTH must not be enabled, protected-mode set to no, replication and administrative commands must not be renamed (e.g., sync, psync, info, config, command, cluster); (target) encryption in-transit disabled, Multi-AZ enabled, engine version Redis OSS 5.0.6+ or Valkey 7.2+, not part of a Global Datastore, data tiering disabled. Shard counts must match between source and target. All source Redis instances must use the same port. Online migration is not supported for serverless caches (node-based targets only). See references/migration/topology-validation.md for the full checklist.

Files

references/setup/serverless-vs-node.md

Serverless vs Node-Based

Hard Routing Rules (non-negotiable)

These constraints override all other guidance. Never recommend serverless for these workloads.

  1. Vector search MUST use node-based Valkey 8.2 or above (recommend 9.0). Serverless does not support vector search. Any workload requiring vector similarity (semantic cache, RAG, conversational memory with vector retrieval, recommendations via embeddings, semantic search) MUST be deployed on a node-based Valkey 8.2 or above cluster (recommend 9.0). Do NOT suggest serverless as a fallback or future option for vector search.

  2. Global Datastore MUST use node-based clusters. Multi-Region replication via Global Datastore is not available on serverless. Any workload requiring cross-Region reads or disaster recovery via Global Datastore MUST use a node-based deployment.

ElastiCache Serverless supports Valkey 7.2+, Redis OSS 7.1+, and Memcached 1.6.22+. Regional availability may vary; check the ElastiCache pricing page or the AWS Region Table before selecting a deployment model.

Quick Decision

Default to serverless unless one of the hard routing rules or node-based requirements below applies.

FactorServerlessNode-Based
Setup time1-5 minutes5-15 minutes
Capacity planningAutomaticManual (choose node type, count)
ScalingInstant, automaticManual or auto-scaling policies
Maintenance windowsNone (zero downtime)Required for patching
Pricing modelPay per GB stored + ECPUs consumedPay per node-hour (reserved or on-demand)
Minimum cost~$6/month (Valkey)Depends on node type
In-transit encryption (TLS)Always on (cannot disable)Optional (recommended)
AuthenticationRBAC/IAM only (no AUTH tokens)RBAC/IAM and AUTH tokens (legacy)
Best forVariable/unpredictable traffic, new projects, getting startedPredictable high-throughput, cost optimization at scale

When serverless wins

  • New projects or prototypes

  • Variable or spiky traffic patterns

  • Teams that don't want to manage infrastructure

  • When time-to-value matters (under 1 minute to create)

  • Workloads under ~100 GB or with unpredictable growth

When node-based is required or preferred

  • Vector search -- requires Valkey 8.2 or above on node-based clusters (recommend 9.0). Available in all AWS Regions at no additional cost. Not supported on data-tiering instances (r6gd) or serverless.

  • Data Tiering -- the r6gd instance family (SSD-backed data tiering) is only available on node-based clusters.

  • Global Datastore -- multi-Region replication requires node-based clusters. Global Datastore has specific engine version requirements and is available only in regions that support multi-Region replication. Check the ElastiCache documentation for supported regions and engine versions.

  • Reserved instances -- reserved nodes save approximately 30-55% depending on term length and payment option (run python3 scripts/price_calculator.py --mode node --node-type <type> --nodes <N> --show-ri-options for current estimates)

  • Fine-grained parameter tuning -- custom parameter groups for advanced optimization

  • Predictable, steady-state high throughput with strict sub-millisecond latency at 50k+ connections

  • Large datasets where reserved instances save significant cost

  • Compliance requirements needing specific instance families

Serverless Constraints and Limits

These constraints are inherent to the serverless deployment model and cannot be changed. Verify each one against the workload requirements before choosing serverless.

Encryption and TLS

  • TLS is always on. In-transit encryption cannot be disabled on serverless caches. All clients must connect with TLS enabled (ssl=True in valkey-py, tls: {} in ioredis, --tls in valkey-cli). If the application cannot use TLS, use node-based instead.

  • At-rest encryption is always on. Cannot be disabled.

Authentication

  • RBAC or IAM only. AUTH token (password-only) authentication is not supported on serverless. Applications using legacy AUTH tokens must switch to RBAC users or IAM-based authentication before migrating to serverless.

Connections

  • 65,000 max concurrent connections. Beyond this threshold, additional connections may or may not succeed depending on current load. If the workload consistently approaches this limit, use node-based with multiple nodes, or implement connection pooling.

  • Connection pooling is strongly recommended regardless of deployment model.

Storage and Throughput Limits

  • Maximum storage: 5,000 GiB. Configurable via CacheUsageLimits.DataStorage.Maximum. Maximum per-slot storage is 32 GiB.

  • Maximum ECPU/s: 15,000,000 ECPU/s. This is the hard platform maximum. CacheUsageLimits.ECPUPerSecond.Maximum lets you set a lower ceiling for cost control, but cannot exceed 15,000,000. Per-slot throughput ranges from 30,000 to 90,000 ECPU/s.

  • For workloads requiring storage beyond 5 TB or predictable high throughput, evaluate node-based for cost efficiency.

Feature Restrictions

  • No vector search. Vector search requires Valkey 8.2 or above on node-based clusters (recommend 9.0).

  • No Global Datastore. Multi-region replication requires node-based clusters.

  • Cluster-mode-enabled only. All serverless caches operate in cluster mode. Clients must support cluster protocol (MOVED/ASK redirects). Non-cluster clients will fail to connect properly.

  • No keyspace notifications. Keyspace events are not supported on serverless caches. Workloads relying on keyspace notifications must use node-based.

  • Eviction policy is volatile-lru (not configurable). Only keys with a TTL set are candidates for eviction. Keys without TTL are never evicted. If storage reaches the maximum and no keys can be evicted, writes receive OOM errors. Always set a TTL on serverless.

  • No custom parameter groups. Parameter tuning is not available on serverless.

  • Manual snapshots supported. Use CreateServerlessCacheSnapshot for on-demand snapshots. Restore via create-serverless-cache --snapshot-arns-to-restore. Automatic daily snapshots are not enabled by default; set SnapshotRetentionLimit > 0 to enable them.

  • SELECT command not supported. Multiple logical databases (SELECT 0, SELECT 1, etc.) are not available. All data lives in a single logical namespace. Use key prefixes for logical separation.

  • Lua scripting restrictions. Scripts that access keys across multiple slots may fail. Keep scripts single-key or use hash tags to colocate keys.

Operations

  • No maintenance windows. Patching is applied transparently with zero downtime. No operator action required.

  • Provisioning time: 1-5 minutes. While the console may say "under a minute," actual creation takes 1-5 minutes in practice.

Networking

  • Port 6379 and 6380 on the same hostname. Serverless uses port 6379 for writes and port 6380 for reads, both on the same endpoint hostname. Security groups must allow inbound TCP on both ports. Many clients connect to both ports automatically. Since TLS is mandatory, ensure no intermediate proxies strip TLS.

  • VPC required. Serverless caches are always deployed inside a VPC.

Latency

  • Single-digit millisecond latency. Serverless does not guarantee sub-millisecond latency. For workloads requiring consistent sub-millisecond response times at high concurrency (50,000+ connections), use node-based.

Serverless Performance Tuning

ECPU consumption model

Serverless bills in ElastiCache Processing Units (ECPUs). How commands are charged:

  • Fixed commands (GET, SET, HGET, INCR, EXISTS, PING, DEL, UNLINK, and other O(1) operations): each request consumes at least 1 ECPU, with 1 ECPU per KB of data transferred. A GET returning 3.2 KB of data consumes 3.2 ECPUs (see ElastiCache pricing).

  • Non-fixed commands (EVAL, SORT, MGET, HGETALL, and other variable-cost operations): consume ECPUs based on the higher of vCPU time or data transferred. For example, an HMGET that takes 3x the vCPU time of a GET and transfers 3.2 KB consumes 3.2 ECPUs (data wins). If it transfers only 2 KB, it consumes 3 ECPUs (vCPU wins). See the ElastiCache Serverless pricing blog.

  • Pipelined commands are charged individually (each command in the pipeline costs its own ECPUs), but pipelining still reduces round-trip latency and network overhead.

  • Non-key commands (ACL, SELECT, ECHO, TIME, etc.): these consume ECPUs as fixed commands. INFO and PUBLISH are exceptions — they are not metered.

  • Pub/sub commands (SUBSCRIBE, PUBLISH, etc.): SUBSCRIBE and UNSUBSCRIBE are not metered. SPUBLISH, SSUBSCRIBE, and SUNSUBSCRIBE are metered.

  • Metadata commands (AUTH, MULTI, EXEC, CONFIG): these are not metered.

  • Replication/cluster commands (REPLCONF, PSYNC, CLUSTER): these are internal to the serverless infrastructure and are not metered.

Use scripts/command_classifier.py with INFO commandstats output for ECPU estimation of an existing workload. Note: the classifier uses heuristic approximations; actual ECPU consumption may differ. For precise billing, use CloudWatch ECPU metrics on a running serverless cache.

Key CloudWatch metrics for serverless
MetricWhat it tells you
ElastiCacheProcessingUnits (Sum)Total ECPUs consumed per period. Primary cost driver.
ThrottledCmds (Sum)Commands rejected due to ECPU limit. Any value > 0 needs action.
SuccessfulReadRequestLatency (p99)Read latency. Baseline is single-digit ms; spikes suggest hot keys or large payloads.
SuccessfulWriteRequestLatency (p99)Write latency. Same baseline expectations as reads.
BytesUsedForCacheStorage consumed. Drives the storage component of the bill.
CurrConnectionsActive connections against the 65K limit.
When to increase ECPU limits

If ThrottledCmds > 0 sustained across multiple minutes, raise the ceiling:

aws elasticache modify-serverless-cache \
  --serverless-cache-name <name> \
  --cache-usage-limits '{"ECPUPerSecond": {"Maximum": <higher-value>}}' \
  --region <region>

Before increasing, check whether expensive commands (SORT, large LRANGE, unoptimized Lua scripts) are inflating ECPU consumption. Optimizing command patterns is cheaper than raising limits.

Client-side optimization

  • Keep payloads small. Every KB transferred adds ECPUs. Compress values over 1 KB (gzip or lz4) when the workload is read-heavy.

  • Use pipelining to batch commands and reduce round trips. Each command still costs ECPUs, but latency drops significantly.

  • Prefer MGET/MSET over loops of GET/SET for bulk operations.

  • Use hash tags {tag} to colocate related keys on the same slot when using multi-key commands.

Serverless Connection Management

Proper connection pooling is critical given the 65K per-cache connection limit (documented above). For per-library pool configuration and code examples, see ../monitoring/client-tuning-and-diagnostics.md (Connection Pooling section).

Key serverless-specific guidance:

  • Monitor CurrConnections in CloudWatch. Alert at 50K to leave headroom for spikes.

  • Lambda and short-lived compute: reuse connections across invocations (module-level client). Each new TLS handshake adds ~5 ms latency and a connection slot.

Serverless Backup Behavior

  • Automatic daily snapshots: serverless caches can take one automated snapshot per day, but this must be explicitly enabled by setting SnapshotRetentionLimit > 0. The default is 0 (disabled).

  • On-demand snapshots: use CreateServerlessCacheSnapshot for on-demand logical snapshots of serverless caches.

  • Retention: SnapshotRetentionLimit controls how many days automatic snapshots are kept (0 disables, max 35 days). Set via --snapshot-retention-limit on create-serverless-cache or modify-serverless-cache.

  • Cross-region snapshot copy: not available natively for serverless. As a workaround, use the export-serverless-cache-snapshot API to export a snapshot to S3, then copy the S3 object to the target region. For long-term retention beyond 35 days, consider node-based clusters which support copy-snapshot --target-bucket for S3 export.

Cost comparison tip

For workloads over ~50 GB with steady traffic, run scripts/price_calculator.py --mode serverless --data-gb <N> --ops-per-sec <N> and --mode node --node-type <type> --nodes <N> --show-ri-options to compare. Reserved node-based instances can be cheaper at scale, but serverless eliminates operational overhead.

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Needs Python, Needs Node.js

Security

What our scanner noticed· Medium risk

  • References to credential files
  • Instructions that use the network
  • Scripts that your agent may run

This skill includes scripts that your agent may run. Read them before you use it.

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Published by
AWS
Publisher
AWS (Official)
Source
aws/agent-toolkit-for-aws · skills/specialized-skills/database-skills/amazon-elasticache · commit 0a218f8
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE)
  • NOTICE (added by AutoGineer from NOTICE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by AWS.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download amazon-elasticache.zip