Event-Driven Patterns for Financial Services on AWS
Domain-specific architectural guidance for building event-driven financial-services systems on AWS: real-time market-data pipelines, alert/subscription matching, multi-channel notification delivery, and the idempotency, observability, and infrastructure patterns that hold them together.
This skill is a router. Load the reference file that matches the aspect you are designing; each file is a self-contained deep-dive with official AWS documentation sources at the bottom.
When to Load Reference Files
| You are working on… |
Load |
| Event-driven pattern selection (notification / carried-state / sourcing / CQRS), event design, schema versioning, ordering |
references/eda-foundations.md |
| MSK topic & partition design, Flink vs Lambda vs ECS stream-processing choice, subscription-matching engine, scale tiers, latency targets |
references/stream-processing.md |
| Multi-channel push/email/SMS delivery, delivery guarantees, retry & DLQ strategy, preference/compliance-aware routing |
references/notification-delivery.md |
| Idempotency keys, atomic conditional writes, effectively-once processing, dedup vs offset commit |
references/idempotency.md |
| Metrics to instrument, alarm thresholds, distributed tracing across Lambda/ECS/DynamoDB/Kafka |
references/observability.md |
| CDK stack decomposition, cross-stack references, event-driven infrastructure layout |
references/cdk-patterns.md |
| Test strategy (unit/integration/contract/E2E/perf) for event pipelines |
references/testing.md |
Core Principles
- Events are immutable facts describing what happened — carry enough context to be useful without back-channel lookups, version every schema, and partition by the entity whose ordering matters (instrument symbol, user ID, order ID).
- Design for at-least-once transport, achieve effectively-once through idempotency. No AWS-native stream transport in this design guarantees exactly-once end-to-end; correctness comes from idempotent, atomically-guarded consumers. Duplicate notification = annoying; duplicate trade = catastrophic.
- Match compute to the processing shape, not by default: stateless per-record → Lambda; stateful/windowed → Managed Service for Apache Flink; persistent-session/long-lived → ECS/Fargate.
- Dead letters are first-class — every consumer handles poison pills and surfaces DLQ depth as an alarm.
- Let CDK synthesize cross-stack wiring by passing construct references between stacks; reserve manual imports for pre-existing/cross-app resources.
1---2name: aws-event-driven-patterns3description: Domain architecture guidance for event-driven financial-services systems on AWS — real-time market-data pipelines, stream processing, alert/subscription matching, multi-channel notification delivery, idempotency, and event-driven CDK stack design. Triggers on phrases like: event-driven architecture for FSI, market data stream processing, MSK topic and partition design, Flink vs Lambda vs ECS for streaming, subscription matching engine, push notification delivery, effectively-once processing, idempotency key, event-driven observability. For generic Lambda/EventBridge how-to unrelated to FSI event pipelines, use the aws-lambda skill instead; for MSK/Kafka service mechanics use the aws-messaging-and-streaming skill instead.4license: MIT-05---67# Event-Driven Patterns for Financial Services on AWS89Domain-specific architectural guidance for building event-driven financial-services systems on AWS: real-time market-data pipelines, alert/subscription matching, multi-channel notification delivery, and the idempotency, observability, and infrastructure patterns that hold them together.1011This skill is a router. Load the reference file that matches the aspect you are designing; each file is a self-contained deep-dive with official AWS documentation sources at the bottom.1213## When to Load Reference Files1415| You are working on… | Load |16|---|---|17| Event-driven pattern selection (notification / carried-state / sourcing / CQRS), event design, schema versioning, ordering | [references/eda-foundations.md](references/eda-foundations.md) |18| MSK topic & partition design, Flink vs Lambda vs ECS stream-processing choice, subscription-matching engine, scale tiers, latency targets | [references/stream-processing.md](references/stream-processing.md) |19| Multi-channel push/email/SMS delivery, delivery guarantees, retry & DLQ strategy, preference/compliance-aware routing | [references/notification-delivery.md](references/notification-delivery.md) |20| Idempotency keys, atomic conditional writes, effectively-once processing, dedup vs offset commit | [references/idempotency.md](references/idempotency.md) |21| Metrics to instrument, alarm thresholds, distributed tracing across Lambda/ECS/DynamoDB/Kafka | [references/observability.md](references/observability.md) |22| CDK stack decomposition, cross-stack references, event-driven infrastructure layout | [references/cdk-patterns.md](references/cdk-patterns.md) |23| Test strategy (unit/integration/contract/E2E/perf) for event pipelines | [references/testing.md](references/testing.md) |2425## Core Principles2627- **Events are immutable facts** describing what happened — carry enough context to be useful without back-channel lookups, version every schema, and partition by the entity whose ordering matters (instrument symbol, user ID, order ID).28- **Design for at-least-once transport, achieve effectively-once through idempotency.** No AWS-native stream transport in this design guarantees exactly-once end-to-end; correctness comes from idempotent, atomically-guarded consumers. Duplicate notification = annoying; duplicate trade = catastrophic.29- **Match compute to the processing shape**, not by default: stateless per-record → Lambda; stateful/windowed → Managed Service for Apache Flink; persistent-session/long-lived → ECS/Fargate.30- **Dead letters are first-class** — every consumer handles poison pills and surfaces DLQ depth as an alarm.31- **Let CDK synthesize cross-stack wiring** by passing construct references between stacks; reserve manual imports for pre-existing/cross-app resources.
aws-samples