Skip to content

Architecture Notes

Design distributed systems that stay correct, resilient, and observable — from first deploy to full scale.

Explore the patterns

Overview

Architecture Notes is a practical field guide built from sixteen years of designing and running enterprise platforms for global banks. Each post explains one pattern the way an architect would: the problem it solves, how it works, a diagram, working code, the trade-offs, and the mistakes that show up in production.

The first series, the Microservices Patterns Field Guide, covers 53 patterns across 6 categories — from sagas and transactional outboxes to circuit breakers, service meshes and distributed tracing.

Written by Karthik Sukumaran, AI & GenAI Solutions Architect.

Start with pattern #1: Saga

The 6 categories

Distributed transactions Keeping data correct when one business operation spans several services and databases. 9 patterns Reliable messaging Publishing events exactly when data changes, delivering them, and processing them safely. 9 patterns Data management Owning, shaping, and querying data across service boundaries. 9 patterns Resilience Surviving slow, failing, and overloaded dependencies without taking the whole system down. 10 patterns Integration How clients and services find, call, and coordinate with one another. 12 patterns Observability Seeing what a distributed system is actually doing — and proving who did what. 4 patterns

Distributed transactions

Keeping data correct when one business operation spans several services and databases.

Displaying 1-4 (9)

pattern 01 2026-10-17·8 min read Saga Chain of local transactions with compensations on failure. pattern 02 2026-10-17·7 min read Two-Phase Commit (2PC) Coordinator runs prepare, then commit or abort across all participants. pattern 03 2026-10-17·6 min read Three-Phase Commit (3PC) Adds a pre-commit phase to reduce 2PC blocking. pattern 04 2026-10-17·7 min read Try-Confirm/Cancel (TCC) Reserve first, then confirm or cancel. pattern 05 2026-10-17·7 min read Compensating Transaction Undoes a completed step with an opposite action. pattern 06 2026-10-17·7 min read Eventual Consistency Data syncs over time instead of instantly. pattern 07 2026-10-17·6 min read Reservation Temporarily holds a resource with an expiry. pattern 08 2026-10-17·7 min read Process Manager Stateful router driving a multi-step workflow. pattern 09 2026-10-17·6 min read Routing Slip Message carries its own list of steps.

Reliable messaging

Publishing events exactly when data changes, delivering them, and processing them safely.

Displaying 1-4 (9)

pattern 10 2026-10-17·6 min read Transactional Outbox Saves data and event in the same DB transaction. pattern 11 2026-10-17·5 min read Transactional Inbox Stores incoming messages before processing to dedupe. pattern 12 2026-10-17·6 min read Change Data Capture (CDC) Turns DB log changes into events. pattern 13 2026-10-17·7 min read Polling Publisher Periodically reads the outbox and publishes. pattern 14 2026-10-17·6 min read Idempotent Consumer Processing a message twice has the same effect as once. pattern 15 2026-10-17·6 min read Dead Letter Queue (DLQ) Parks messages that keep failing. pattern 16 2026-10-17·5 min read Guaranteed Delivery Persists messages until acknowledged. pattern 17 2026-10-17·5 min read Competing Consumers Multiple workers share one queue. pattern 18 2026-10-17·6 min read Claim Check Sends a reference instead of a large payload.

Data management

Owning, shaping, and querying data across service boundaries.

Displaying 1-4 (9)

pattern 19 2026-10-17·7 min read Database per Service Each service owns a private DB. pattern 20 2026-10-17·6 min read Shared Database Services share one DB (often an anti-pattern). pattern 21 2026-10-17·7 min read Event Sourcing Stores state as a sequence of events. pattern 22 2026-10-17·6 min read CQRS Separates write and read models. pattern 23 2026-10-17·6 min read API Composition Merges data from multiple services for a query. pattern 24 2026-10-17·6 min read Materialized View Precomputed read-optimized view. pattern 25 2026-10-17·6 min read Domain Events Publishes meaningful business events. pattern 26 2026-10-17·6 min read Aggregate (DDD) Cluster of objects forming one consistency boundary. pattern 27 2026-10-17·6 min read Snapshot Saves state periodically to avoid full event replay.

Resilience

Surviving slow, failing, and overloaded dependencies without taking the whole system down.

Displaying 1-4 (10)

pattern 28 2026-10-17·6 min read Retry with Backoff Retries with growing delays plus jitter. pattern 29 2026-10-17·5 min read Circuit Breaker Stops calling a failing service and fails fast. pattern 30 2026-10-17·6 min read Bulkhead Isolates resources per dependency. pattern 31 2026-10-17·6 min read Timeout Caps how long to wait on a call. pattern 32 2026-10-17·6 min read Fallback Returns a default or cached response on failure. pattern 33 2026-10-17·6 min read Rate Limiting / Throttling Caps request volume. pattern 34 2026-10-17·6 min read Load Shedding Drops low-priority work under load. pattern 35 2026-10-17·6 min read Health Check Reports liveness and readiness. pattern 36 2026-10-17·6 min read Leader Election Picks one instance to do a task. pattern 37 2026-10-17·7 min read Distributed Lock Allows one process at a time on a resource.

Integration

How clients and services find, call, and coordinate with one another.

Displaying 1-4 (12)

pattern 38 2026-10-17·5 min read API Gateway Single entry point for clients. pattern 39 2026-10-17·6 min read Backend for Frontend (BFF) Separate gateway per client type. pattern 40 2026-10-17·5 min read Choreography Services coordinate by reacting to events. pattern 41 2026-10-17·7 min read Orchestration A central coordinator directs services. pattern 42 2026-10-17·5 min read Publish-Subscribe Producers publish to topics, many subscribers receive. pattern 43 2026-10-17·6 min read Async Request-Reply Request on one queue, reply on another. pattern 44 2026-10-17·6 min read Correlation ID One ID tracks a flow across services. pattern 45 2026-10-17·6 min read Anti-Corruption Layer Translates a legacy or external model. pattern 46 2026-10-17·7 min read Strangler Fig Gradually replaces a monolith. pattern 47 2026-10-17·6 min read Sidecar Helper container next to a service. pattern 48 2026-10-17·6 min read Service Mesh Manages service-to-service traffic. pattern 49 2026-10-17·6 min read Service Discovery Services find each other dynamically.

Observability

Seeing what a distributed system is actually doing — and proving who did what.

Displaying 1-4 (4)

pattern 50 2026-10-17·6 min read Distributed Tracing Follows a request across services. pattern 51 2026-10-17·6 min read Log Aggregation Centralizes logs. pattern 52 2026-10-17·6 min read Application Metrics Tracks latency, errors, throughput. pattern 53 2026-10-17·7 min read Audit Logging Records who did what and when.