Requirements, VM or container build, network, LB/TLS, monitoring, go-live, handover to the business
Practise these as flashcards. One or two lines each; the ones you miss come back first.
Start freeIntake (owner, criticality, users, deadline), requirements and NFRs (stack, sizing, ports and dependencies, uptime, RTO/RPO), design and architecture review, approvals (security, firewall, LB, cert, CAB), provision with IaC, configure (baseline, hardening, agents, runtime, secrets), expose (DNS, LB, TLS, WAF, firewall), observe (monitoring, logs, alerts with runbooks), protect (backup with test restore, patching, vuln scan, DR), validate (smoke, load, failover, UAT), go-live with a cutover runbook and rollback trigger, hypercare, then handover with runbook, KT, CMDB and business sign-off.
Owner and criticality, user base and locations, tech stack and OS, sizing, database, every inbound port and outbound dependency, authentication method, NFRs (availability, RTO/RPO, peak load, latency), environments, support team and deploy method, and the go-live date with any compliance needs.
How the system must behave rather than what it does: availability, performance and latency, scalability, RTO/RPO, security and compliance, retention, supportability. They drive most infra decisions: zones, sizing, DR, monitoring.
So clients and resolvers pick up the new record within minutes, and so a rollback (flipping the record back) takes effect quickly. Lower it a day or more ahead, since the old TTL must expire first; raise it again after.
What the app does and its dependencies, how to start/stop and check health, where logs are, common alerts and what to do, failover and recovery steps, escalation contacts, and links to dashboards and the architecture diagram. Written so on-call can act without the builder.
Start with the app: vendor app needing full OS → VM. Standard web app or API → App Service. Containerized with a few services and no K8s skills → Container Apps. Many microservices and a team that runs Kubernetes → AKS. Prefer the most managed option that meets the requirements, and state who owns patching and scaling in each.
A table of every required connection: source, destination, port, protocol, purpose. It becomes the firewall and NSG requests, the security review evidence and the troubleshooting map. Building it early prevents go-live surprises.
Hit it directly with the real hostname: curl -v --resolve app.example.com:443:<VIP> https://app.example.com/health and openssl s_client -servername. Check backend health on the LB, test each backend directly with the Host header, and run a short load test. Then DNS is the only change at cutover.
Probe path and host: does /health return 200-399 without auth or redirect? Is the app listening on the backend port on a non-loopback address? Do NSGs on the backend subnet allow the App Gateway subnet, and does the App Gateway subnet allow GatewayManager 65200-65535? For HTTPS backends, does the cert match the host name and is the root trusted? az network application-gateway show-backend-health usually tells you which.
Use vendor guidance or current metrics, then a load test: measure what one instance handles at a safe CPU level, divide peak load by it, add N+1, and spread across zones. Size subnets for growth. Revisit after 30 days of real usage.
A defined period after go-live, usually one to four weeks, when the project team watches closely, fixes issues fast, tunes alerts and capacity, and answers support. It ends with a formal transition to BAU support.
The app can be supported and recovered by someone other than the builder: monitoring and alerts routed with runbooks, central logs, backups with a test restore, patching assigned, vuln scan clean, certs inventoried, secrets in a vault, CMDB updated, DR documented, and the support team trained.