3-month-devops_plan.md

• Deployment Pipeline (re-instate)

  • CI (shippable or semaphore) triggers pipeline on successful master build
    • Probably 'Go' - has misgivings but does the job for now
  • builds and pushes docker container artifact
  • deploy to staging
    • smoke test
    • acceptance test
  • Deploy to prod (auto-deploy can be optional e.g. supporter)
    • smoke test
    • Rollback on failure

• Common flow for all services - adding a new one should be trivial

• Security updates (e.g. base images) can have own pipelines that are dependencies of deployment ones

  • if the update passes acceptance tests, it can be shipped
  • tracability of which base image was used for a version

• New Go Server on staging-build - Core OS running Go in docker. Same host can be used as shippable host.

  • Each job runs in a container, so dependencies can be encapsulated. Old Go hosts required every dependency available.

• Dynamic Routing

• Registrator to roll service locations to nginx

• Support internal only services (e.g. rous)

• GliderGun style deployments

  • new version in parallel to current version - private route?
  • (only enter into routing pool after smoke test?)
  • clean up old version
  • Instance tags for service locations

• Build resilience into AWS architecture

  • Auto-scale groups around every host group - even with static size for resilience
  • Launch configuration/User data bootstraps a node enough to join pool automatically

• Staging and Production uniformity

  • All changes rolled through staging
  • Terraform?

• Standard Health checks across services

  • Consul health checks

• Continue general monitoring/alerting

• Continue general log shipping/aggregation