Deployment, Provisioning, and Operations
Provisioning vs configuration vs orchestration
Provisioning creates resources (a VM, a bucket, a load balancer). Configuration shapes them (packages, OS settings, application config). Orchestration sequences many resources into a running system. Questions often blur these; read for whether the task is creating or shaping.
Release strategies
Recreate/big-bang (simple, downtime). Rolling (replace instances in batches, partial exposure to a bad version). Blue-green (two identical environments, switch traffic, instant rollback, roughly double the cost). Canary (send a small slice of traffic to the new version, then progressively widen). Feature flags decouple deploy from release. Environment parity - staging resembling production - is what makes these strategies trustworthy.
Operating the estate
Monitoring collects metrics, logs and traces; observability is the ability to answer new questions from that telemetry without redeploying. Know SLI (measured indicator), SLO (internal target) and SLA (contractual promise with credits), plus error budgets, which let teams trade reliability against release speed. Patching cadence, runbooks, immutable infrastructure (replace, don't patch in place) and chaos engineering belong here too.
Recovery vocabulary
RTO = maximum tolerable downtime. RPO = maximum tolerable data loss, measured in time. Backup protects against deletion and corruption; replication propagates them, so replication is not backup. Snapshots are typically incremental and regional; cross-region copies are a separate configuration.
Rule of thumb: RPO sets backup frequency; RTO sets recovery method. Daily backups imply up to 24 hours of data loss no matter how fast you can rebuild.
Trap: swapped RTO/RPO in scenario questions, and choosing blue-green (cost) when a canary (risk control) is what the scenario describes.