An artifact repository is all about keeping built deliverables in a centralized, versioned store so deployment can pull exactly what was built. It holds artifacts like binaries, libraries, and container images, and it tracks versions so you can reproduce a deployment by selecting a specific artifact revision. This immutability and traceability enable safe promotion across environments (dev, QA, staging, production) and ensure the same artifact is used in every run. Deployment tools fetch the needed artifact from the repository, guaranteeing consistency and a predictable release process. The other options don’t match this role: hosting user data is typically handled by data stores or databases, running applications happens on compute environments, and runtime metrics come from monitoring systems rather than an artifact store.

In route synchronization, the two basic movements describe how updates are coordinated between devices: planned synchronization and immediate synchronization. Planned synchronization schedules updates at a predefined time or interval, allowing coordinated, batched changes that improve stability and make verification easier across multiple routers. Immediate synchronization pushes updates as soon as they occur, keeping routing information current and reducing convergence time, though it can increase short-term update churn. Together these cover the main approaches: one emphasizes control and predictability, the other emphasizes speed and freshness. The other options describe directions, scope, or order rather than how updates are timed and propagated.

A VPC is a virtual private cloud—a virtual network in a cloud environment that isolates resources and enables segmentation. It gives you control over networking constructs like IP address ranges, subnets, route tables, gateways, and security controls (security groups and network ACLs). This isolation lets you segment resources for different environments or workloads and precisely control who and what can communicate with what. It’s not simply a public cloud service for hosting websites, not a local data center appliance, and not a data backup vault—the defining idea is the isolated, segmented virtual network.

When releasing changes through canary deployments, the goal is to catch problems quickly and limit impact by having guardrails that respond automatically. The safest practice is to define clear thresholds on key health and performance metrics and have the system roll back automatically if those metrics breach the limits. This creates a fast, consistent safety net: if error rates rise, latency spikes, or availability drops beyond what you’ve tolerated, traffic is redirected back to the known good version without waiting for a human to intervene. It helps protect users and keeps deployments from drifting into unstable states. This approach relies on observable signals tied to reliable service expectations, such as error rate, latency (P95 or P99), and saturation, often aligned with service level objectives. Automatic rollback enforces the policy consistently, reducing recovery time and minimizing blast radius, which is the whole point of canary testing. Rolling back only after a manual audit introduces delays that defeat the purpose of rapid, incremental releases. Never rolling back once deployed leaves you exposed to failures that can worsen over time. Ignoring metrics during rollout eliminates the very guardrail that makes canaries safe.