Component Library
Generic architecture roles. Vendor names are examples of the pattern, not requirements.
Clients & entry
Edge & traffic
- DNSResolves a hostname to IP addresses; can steer users to regions (geo/latency routing).
- CDN / edge cacheGeographically distributed caches that serve static and cacheable content near users.
- Firewall / WAFFilters malicious or malformed traffic before it reaches the application.
- Load balancerDistributes requests across healthy instances and removes failed ones via health checks.
- Reverse proxyTerminates TLS and forwards requests to upstreams; often also compresses and caches.
- API gateway / rate limiterSingle front door for APIs: authentication, rate limiting, routing and request shaping.
- Route 53Managed DNS with latency, geo and failover routing policies.
- CloudFrontGlobal CDN caching static and dynamic content at edge locations.
- AWS WAF / ShieldWeb application firewall and DDoS protection in front of CloudFront, ALB or API Gateway.
- Elastic Load Balancing (ALB/NLB)Layer 7 (ALB) or layer 4 (NLB) load balancing across targets with health checks.
- API GatewayManaged REST/HTTP/WebSocket APIs with throttling, auth and usage plans.
- Cloud DNSManaged authoritative DNS with routing policies.
- Cloud CDNEdge caching on Google's global network, fronted by the load balancer.
- Cloud ArmorWAF and DDoS protection at the Google edge.
- Cloud Load BalancingGlobal anycast HTTP(S) or regional TCP/UDP load balancing.
- Apigee / API GatewayAPI management with quotas, keys and analytics.
- Azure DNS / Traffic ManagerDNS hosting and DNS-based global traffic routing.
- Front Door / Azure CDNGlobal edge entry with caching, routing and acceleration.
- Azure WAF / DDoS ProtectionWeb application firewall on Front Door or Application Gateway.
- Application Gateway / Load BalancerLayer 7 (App Gateway) or layer 4 load balancing.
- API ManagementAPI gateway with policies, quotas, keys and developer portal.
Compute
- Application / API serviceStateless business logic that handles requests and calls data stores and other services.
- MicroserviceAn independently deployable service owning one business capability and its data.
- Serverless functionEvent-triggered compute that scales per invocation with no servers to manage.
- Stateless worker / consumerBackground process that pulls messages from a queue and processes them.
- Scheduled jobPeriodic batch trigger (cron) that emits work on a schedule.
- EC2 (Auto Scaling)Virtual machines, usually in an Auto Scaling group behind a load balancer.
- ECS / FargateManaged container orchestration; Fargate removes server management.
- EKS (Kubernetes)Managed Kubernetes control plane for containerised services.
- LambdaEvent-driven functions billed per request and duration; cold starts add latency.
- EventBridge SchedulerCron and one-off schedules that invoke targets.
- Compute Engine (MIG)Virtual machines in managed instance groups with autoscaling.
- Cloud RunServerless containers that scale with requests, including to zero.
- GKE (Kubernetes)Managed Kubernetes, Standard or Autopilot.
- Cloud FunctionsEvent-driven functions (now Cloud Run functions).
- Cloud SchedulerManaged cron jobs that call HTTP, Pub/Sub or functions.
- DataflowManaged Apache Beam stream and batch processing.
- Virtual Machine Scale SetsAutoscaled groups of identical VMs.
- App ServiceManaged web app hosting with autoscale.
- Container AppsServerless containers with scale-to-zero, built on Kubernetes.
- AKS (Kubernetes)Managed Kubernetes service.
- Azure FunctionsEvent-driven serverless functions.
- Logic Apps (timer)Scheduled and event-driven workflows.
Messaging & async
- Message queueDurable buffer between producers and consumers; absorbs bursts and decouples failures.
- Event stream / topicPartitioned, replayable append-only log consumed by independent consumer groups.
- Dead-letter queueHolds messages that repeatedly fail processing so they stop blocking the main queue.
- SQSFully managed message queue (standard or FIFO).
- SNSPub/sub fan-out to queues, functions, HTTP and mobile push.
- Kinesis Data StreamsSharded, ordered event stream with replay.
- MSK (Managed Kafka)Managed Apache Kafka clusters.
- EventBridgeServerless event bus routing events by rules.
- Cloud TasksManaged task queue with rate control and retries.
- Pub/SubGlobal pub/sub messaging with at-least-once delivery.
- Service BusEnterprise message broker with queues, topics and sessions.
- Storage QueuesSimple, cheap queue on Azure Storage.
- Event HubsPartitioned event ingestion stream, Kafka-compatible.
- Event GridEvent routing service with push delivery.
Data & storage
- Cache (in-memory)In-memory key-value store that answers hot reads and shields the database.
- Relational databaseACID transactions, joins and flexible queries over structured data.
- Key-value / wide-column storePartitioned store accessed by key; scales horizontally for massive volume.
- Document storeStores JSON-like documents with flexible schema and secondary indexes.
- Graph storeOptimised for traversing relationships (friends-of-friends, fraud rings).
- Object / blob storageDurable, cheap storage for large immutable objects (images, video, backups).
- Search indexInverted index for full-text and faceted search.
- ElastiCache (Redis/Valkey)Managed in-memory cache and data store.
- RDS / AuroraManaged relational databases (PostgreSQL, MySQL); Aurora adds fast replicas.
- DynamoDBServerless key-value / document store partitioned by key; capacity in RCUs/WCUs.
- DocumentDBMongoDB-compatible document database.
- NeptuneManaged graph database (Gremlin, openCypher, SPARQL).
- S3Durable object storage for files, media, backups and data lakes.
- OpenSearch ServiceManaged search and log analytics; also supports vector search.
- Memorystore (Redis/Valkey)Managed in-memory cache.
- Cloud SQL / AlloyDBManaged PostgreSQL/MySQL; AlloyDB adds high-performance Postgres.
- SpannerGlobally distributed, strongly consistent relational database.
- BigtableWide-column store for very high throughput, low-latency workloads.
- FirestoreServerless document database with real-time sync.
- Cloud StorageDurable object storage with storage classes.
- BigQueryServerless analytics warehouse; scans are not for request-path latency.
- Azure Cache for RedisManaged Redis cache.
- Azure SQL / PostgreSQL FlexibleManaged relational databases.
- Cosmos DBGlobally distributed multi-model database with tunable consistency; capacity in RU/s.
- Table StorageLow-cost key-value table store.
- Blob StorageObject storage with hot, cool and archive tiers.
- Azure AI SearchSearch index with keyword, semantic and vector retrieval.
AI / ML
- LLM inferenceServes a language model; latency = time to first token + output tokens / decode speed.
- Embedding modelTurns text or images into vectors for similarity search.
- Vector databaseApproximate nearest-neighbour search over embeddings (e.g. HNSW, IVF).
- RerankerCross-encoder that re-scores retrieved candidates for precision.
- Retrieval serviceEmbeds the query, searches indexes, filters by permission, returns passages.
- Prompt / context builderAssembles system prompt, history and retrieved context within the token budget.
- Agent / orchestratorRuns the plan–act–observe loop: calls the LLM, invokes tools, keeps state.
- Tool / MCP serverExposes actions and data to agents through a standard protocol (e.g. MCP).
- Guardrails / content filterChecks inputs and outputs for policy, PII and prompt injection.
- Feature storeServes the same features online (low latency) and offline (training) to avoid skew.
- Model serving (online)Hosts a trained (non-LLM) model behind a prediction API.
- Evaluation / monitoringLogs traces, scores outputs and tracks quality, cost and drift.
- BedrockManaged access to foundation models (Claude, Llama, Titan…) via API.
- SageMaker EndpointHosted real-time inference for your own models.
- Bedrock Knowledge BasesManaged RAG: ingest, embed and retrieve documents.
- Bedrock AgentsManaged agent orchestration with tool (action group) calls.
- Bedrock GuardrailsContent filters, PII redaction and topic policies for model I/O.
- Vertex AI — GeminiManaged Gemini and partner models via Vertex AI.
- Vertex AI EndpointOnline prediction endpoints for custom models.
- Vertex AI EmbeddingsText and multimodal embedding models.
- Vertex AI Vector SearchManaged approximate nearest-neighbour index.
- Vertex AI Agent BuilderBuild agents and search apps grounded in your data.
- Azure OpenAIGPT and other OpenAI models hosted in Azure regions.
- Azure ML Online EndpointManaged real-time inference for custom models.
- Azure OpenAI EmbeddingsEmbedding models served by Azure OpenAI.
- AI Foundry Agent ServiceManaged agents with tools and memory.
- AI Content SafetyDetects harmful content and prompt attacks.
Platform
- Auth serviceAuthenticates callers and issues/validates tokens; enforces authorisation.
- Secrets / configCentral store for secrets and runtime configuration.
- Metrics / logs / tracesCollects telemetry so operators can see and debug system behaviour.
- CognitoUser sign-up, sign-in and token issuance.
- Secrets ManagerStores and rotates secrets and credentials.
- CloudWatch / X-RayMetrics, logs, alarms and distributed tracing.
- Identity PlatformCustomer identity and authentication.
- Secret ManagerStores and versions secrets.
- Cloud Monitoring / LoggingMetrics, logs, traces and alerting.
- Entra ID (B2C)Identity and access management for users and apps.
- Key VaultSecrets, keys and certificate storage.
- Azure Monitor / App InsightsMetrics, logs, traces and alerts.