|$ curl https://forge-ai.dev/api/markdown?path=docs/architecture/devops
$cat docs/devops-lifecycle.md
updated Today·20-28 min read·published

DevOps Lifecycle

ArchitectureSDLCDevOpsIntermediate🎯Free Tools
Introduction

DevOps is the cultural and technical movement to deliver software changes safely, quickly, and sustainably by removing walls between development and operations. It is not a job title alone, a tool chain alone, or "developers on call without support." It is shared ownership of production outcomes plus automation that makes frequent change low-risk.

Architecture and DevOps co-determine each other: deployable units, configuration strategy, observability, and rollback design are architectural decisions that enable (or block) continuous delivery.

info

Hands-on delivery docs: CI/CD and Ops. This page focuses on lifecycle culture and architectural implications.
CALMS
PillarMeaningSmell if missing
CultureBlame-aware learning; shared goalsTicket tossing; hero culture
AutomationRepeatable build/test/deploySnowflake servers; manual release
LeanSmall batches; limit WIP; reduce wasteHuge releases; long queues
MeasurementDORA + SLOs + business metricsOpinion-only debates
SharingDocs, on-call, postmortemsTribal knowledge; bus factor 1

Use CALMS as a diagnostic: if you bought a CI product but culture still punishes failure, you have tooling without DevOps.

Continuous Everything

Continuous Integration

Every change integrates to mainline frequently with automated build and tests. Branch lifetimes stay short. Architecture must allow independent testing of modules.

Continuous Delivery

Every green mainline build is releasable. Release is a business decision. Requires DoD, env parity, and migration discipline.

Continuous Deployment

Every green build deploys to production automatically. Needs strong automated verification, progressive delivery, and fast rollback.

Continuous Learning

Postmortems, experiment results, and customer metrics feed the backlog — closing the SDLC loop.

DORA metricWhat it indicates
Deployment frequencyBatch size / throughput
Lead time for changesFlow from commit to production
Change fail rateQuality of verification & design
Time to restore serviceResilience & incident readiness
Shift-Left

Shift-left means moving quality, security, and operability earlier in the lifecycle — where fixes are cheaper:

  • Security: threat modeling in design; SAST/dependency scans in CI; secrets hygiene.
  • Testing: unit/contract tests with code; perf budgets on critical paths.
  • Operability: design for logs/metrics/traces; define SLOs with features.
  • Accessibility and privacy: treat as NFRs with acceptance checks.

best practice

Shift-left is not 'dump ops work on developers alone.' Provide platforms, paved roads, and specialist partnership.
DevOps Lifecycle Loop

A common loop: Plan → Code → Build → Test → Release → Deploy → Operate → Monitor → (feedback) Plan. Architecture decisions appear in every stage:

StageArchitecture concern
PlanEnablers & NFRs on backlog
CodeModule boundaries; ports/adapters
Build/TestFitness functions; contract tests
Release/DeployDeployable topology; flags; migrations
Operate/MonitorSLOs; failure domains; runbooks
delivery-sketch.yml
YAML
1stages: [build, test, harden, deploy]
2build: [compile, unit]
3test: [integration, contract]
4harden: [sast, deps, smoke]
5deploy:
6 strategy: canary
7 rollback: automatic_on_slo_burn
Platform Engineering

Platform engineering provides internal developer platforms (IDPs): paved roads for CI/CD, observability, environments, and golden paths. Teams self-serve without filing tickets for every cluster detail.

  • Why: reduce cognitive load; standardize security baselines; improve DORA metrics.
  • How: treat the platform as a product with users (app teams), SLAs, and docs.
  • Architecture link: platforms encode architectural standards (service templates, module layouts) as defaults — not only wiki pages.
📝

note

A platform that forces one architecture on every domain will be bypassed. Offer golden paths with escape hatches and documented tradeoffs.
When DevOps Practices Are Required
RequirementDevOps response
Multiple deploys per weekCI/CD + small batches + flags
High availability SLOsProgressive delivery + fast restore
Compliance with frequent changeAutomated evidence generation
Many teams sharing infraPlatform engineering
Chronic release fearAutomation + blameless learning culture
Common Mistakes
  • Renaming ops to DevOps without changing incentives or automation.
  • CI that is flaky — teams bypass it and lose the safety net.
  • Continuous deployment into an unobservable system.
  • Ignoring database migration and schema compatibility.
  • Architecture that requires downtime for every release.
How to Adopt DevOps Incrementally
  • Make mainline green and trusted — fix flaky tests first.
  • Automate one painful release step per week until releases are boring.
  • Add progressive delivery before increasing deploy frequency.
  • Define SLOs for critical user journeys; alert on burn rate.
  • Run blameless postmortems with action items in the backlog.
  • Invest in a thin platform: CI templates, observability defaults, env provisioning.

Architecture enablement

Strangle monoliths into deployable modules; expand contract tests; eliminate shared-DB coupling before splitting teams.

Security enablement

Shift-left scans + secrets management + least privilege cloud roles as paved road.

Anti-patternReplace with
Monthly release train onlySmall batches + flags; trains optional for marketing
Separate QA after code completeQA engineers embedded; automation in CI
Ops as ticket sinkPlatform self-service + shared on-call
No rollback storyBackward-compatible deploys + tested rollback
DevSecOps Notes

Integrate security into the same lifecycle: policy-as-code, image scanning, secret detection, and least-privilege runtime identities. Security gates should be automated and fast enough not to recreate a late Waterfall security phase.

Community

Get help on Slack, Discord or VIP

Stuck on a guide? Join the community and ask.