• Irreversible Decisions
    • Constraints First
    • Coupling and Cohesion in Practice
    • Trade-off Sliders
    • The Boring Baseline
    • Monolith First, Split on Evidence
    • Service Boundaries That Survive Reorgs
    • Synchronous vs Asynchronous Integration
    • API Contracts and Versioning
    • Idempotency and Retries
    • Choosing a Datastore Without Regret
    • Schema Migrations Without Downtime
    • Consistency Models You Actually Need
    • Caching: The Four Questions
    • Event Sourcing and CDC: When It Pays
    • Trunk-Based Development and Branch Reality
    • Pipeline Design: Fast Feedback, Slow Gates
    • Progressive Delivery
    • Build Reproducibility and Artifact Promotion
    • Rollback Is a Feature
    • Kubernetes: What You Sign Up For
    • Infrastructure as Code That Doesn't Drift
    • Environments, Config, and Secrets
    • Multi-Tenancy and Cost Boundaries
    • The Internal Platform as a Product
    • SLOs, Error Budgets, and Saying No
    • Observability: What to Actually Wire
    • Capacity, Load Shedding, and Backpressure
    • Failure Modes: Timeouts, Breakers, Bulkheads
    • Incident Response and Blameless Postmortems
    • Threat Modeling in One Hour
    • Identity, Authentication, Authorisation
    • Secrets and Key Management
    • Supply Chain: Dependencies, SBOM, Signing
    • Least Privilege and Auditability
    • Architecture Decision Records That Get Read
    • Diagrams That Age Well
    • Runbooks and On-Call Docs
    • Design Reviews and RFCs
    • Building a Team Architecture Memory
    • GitHub
  • to navigate
  • to select
  • to close
    • Home
    • Data and State
    On this page
    database

    Data and State

    The layer that is hardest to change and easiest to get wrong: storage choice, migrations, consistency, and caching.

    storage

    Choosing a Datastore Without Regret

    Start from Postgres and make the alternative prove itself. Most 'we need NoSQL' claims do not survive contact with numbers.

    swap_horiz

    Schema Migrations Without Downtime

    Every schema change is a distributed system problem, because old and new code run at the same time.

    balance

    Consistency Models You Actually Need

    Almost nothing needs global strong consistency. Find the few things that do and be strict there.

    bolt

    Caching: The Four Questions

    A cache is a second copy of the truth. Answer four questions before you add one, or you have added a bug with a hit rate.

    hub

    Event Sourcing and CDC: When It Pays

    Two very different ideas that get confused. One is a storage model, the other is an integration mechanism.

    © 2026 Architecture Field Notes. Built with Lotus Docs