Lee ChanyangBackend Engineer

Language

← Back to design approaches

Process Evidence · 05

Select infrastructure options that fit quality attributes

I evaluated infrastructure configurations against execution time, failure cost, retryability, observability, and cost constraints.

I separated synchronous APIs, workers, object storage, Redis, CloudWatch, and deployment boundaries according to the nature of each domain flow.

How I worked

Actual Working Approach

  1. Input

    I examined execution time, file size, failure-propagation scope, zero-downtime constraints, cost conditions, deployment-tool-generated artifacts, and human-managed source resources together.

  2. Questions to ask

    I decided whether the API server should handle the work directly, whether it should be separated into a queue/worker or external-storage adapter, and which resources should be owned by code.

  3. Output

    I documented where asynchronous processing runs, object-storage responsibilities, ASG · CodeDeploy · IaC ownership boundaries, and observability criteria in logs and operational documents.

Source evidence

Selecting infrastructure evidence by execution characteristics, deployment safety, and observability

For the infrastructure card, I retained evidence showing where long-running or high-impact failure flows were placed and which states support recovery during deployment, rather than a list of technologies.

Operational documentation
I checked production-service documents for CodeDeploy, deployment orchestration, DR, SLOs, Outbox operating procedures, CloudWatch · ALB health checks, and long-term region-pinning criteria.
Deployment decisions
I grouped reflections on CI/CD separation, the production CodeDeploy transition, the OIDC migration path, and brownfield IaC organization as evidence of deployment safety.
Asynchronous execution
Material on CSV ingestion, Firebase · Cloud Functions, queues and workers, and external-storage adapters serves as supporting examples of work separated from the request-response boundary.
Metric limitations
I do not invent p95, uptime, RPS, TPS, or cost figures that cannot be verified from accessible logs; I use only operational criteria supported by documents and code.
Detailed work example

Choosing infrastructure boundaries to match domain-flow execution characteristics

I separated synchronous APIs, asynchronous workers, file processing, deployment resources, and observability criteria based on the execution time implied by requirements. Requests needing fast responses were not grouped in the same execution boundary as long-running or retryable work.

From quality attributes to infrastructure choicesMermaid

Rendering the diagram.

Where This Led to Implementation

  • Multi-Gateway Subscription Billing System

    I separated synchronous payment APIs, subscription-renewal workers, Redis · BullMQ, PostgreSQL, gateway callbacks, and CodeDeploy deployment checks into distinct execution responsibilities.

    View the payment runtime architecture
  • Points Accounting & Rewards System

    I separated the execution responsibilities of APIs, workers, Redis · BullMQ, PostgreSQL, CodeDeploy, and CloudWatch, and defined boundaries for codifying the running Blue/Green resources.

    View infrastructure architecture
  • On-Chain Statistics Data Pipeline

    I separated user APIs, administrator approval flows, and statistics batches into independent execution units, using Redis locks and cursor checkpoints to define batch rerun criteria.

    View the batch execution structure
  • Content SaaS Operations Backend

    I distinguished external-infrastructure boundaries such as Firebase, MongoDB, and S3 in location-based AR commerce from the limitations of publicly shareable deployment evidence.

    View early infrastructure boundaries
  • Legacy Domain Modernization

    The life-planning content modernization has traces of Docker/ECR scripts and migration-transition evidence, but deployment logs and server topology are marked as limitations because evidence is insufficient.

    View transition and deployment limitations
  • Evidence-driven Engineering Portfolio

    I used Next.js static routes, sitemap/robots, and lint/typecheck/build/check as pre-publication verification gates.

    View portfolio deployment verification
Input
I considered execution time, file size, failure-propagation scope, zero-downtime constraints, cost conditions, deployment-tool-generated artifacts, and human-managed source resources together.
Judgment
I kept requests requiring immediate responses on the API server, moved retryable follow-up work to workers and queues, and separated file and attachment storage behind external-storage adapters.
Boundaries
In the running Blue/Green environment, I distinguished deployment-tool-generated artifacts from original resources that should be owned by code.
Output
I documented where asynchronous processing runs, object-storage responsibilities, ASG · CodeDeploy · IaC ownership boundaries, and observability criteria in logs and operational documents.
Verification criteria

Distinguishing operational verification criteria from measurable metrics

Infrastructure and operational materials contain criteria for assessing operability, such as CloudWatch, ALB health checks, operational documents, and deployment boundaries. I do not invent unavailable production p95 or error-rate figures; I use only verification criteria that can be disclosed.

Operational criteria
I documented health-check, logging, and reprocessing criteria through ALB · ASG · CodeDeploy · CloudWatch, Outbox operational documents, and SLO documentation.
Measurable
Monthly points settlement is an approximate figure observed in operation at 1 point = KRW 1, including small community rewards and operational payouts based on offline package payments. Search-window sizes, processing-batch sizes, and renewal-failure counts are stated only when supported by documents or code.
Not measurable
I do not present p95, error rates, or uptime from AWS operational logs that are currently inaccessible as performance results.