Knowledge Check - Operations, Resilience, and Performance
Test your understanding of the key concepts from Module 7. Select the best answer for each question, then submit to see your results. You need to score at least 60% to pass.
FinCorp's portal runs in Frankfurt with a pilot light in the Berlin VDC: network, reserved addresses, ALB, images and stopped portal servers. The portal database is a DBaaS for PostgreSQL cluster in Frankfurt with its backup location outside Frankfurt. The external monitor can rewrite the portal record through the Cloud DNS API. Frankfurt is lost. What failover sequence should the runbook prescribe?
Steering must not run ahead of data: a record that points at a site without its database is an outage with a new address, and DNS steers new connections within one TTL whether or not the backend is ready. The cluster's replication runs between instances inside one location, so there is no standby in Berlin to promote; recovery is a new cluster created from the off-site backup. ALB health checks only remove failed targets inside a site; they cannot make a missing database appear.
After building fc-pg-portal, an engineer turned on Enable logs and Enable metrics in its Observability settings. Weeks later, the predefined PostgreSQL dashboards in Grafana are still empty, and the cluster shows no error. What is the most likely cause, and the fix?
Product-initiated forwarding has two layers: the contract-level Central Logging and Central Monitoring switch per region, and the product's own toggles. The cluster accepts its toggles while the region is inactive, which is why there is no error, but no data flows until both layers are on. Managed database instances are not agent hosts, the predefined dashboards cover PostgreSQL, MariaDB and In-Memory DB, and an expired Grafana login would block access rather than show empty dashboards.
A self-managed reporting database on a 2-core VM uses an 80 GB Performance volume and saturates its storage I/O during month-end. Finance wants the cheapest change that fixes the bottleneck, and the database will stay under 100 GB of data for years. What should the architect recommend?
Balanced and Performance scale with volume size, are recommended at 100 GB or more, and reach their per-volume ceilings only with a minimum server size, so an 80 GB volume on 2 cores is below both floors. Sizing for I/O rather than for data, and comparing the classes on price per IOPS, fixes the bottleneck at the lowest cost. Essential is static at a low rate, NIC Multi-Queue addresses network rather than storage I/O, and a snapshot is a rollback point, not a read replica.
FinCorp's core-banking systems run on vSphere with VMware-certified vendor appliances. The risk office allows each service a cutover window of under one hour and demands a tested way back for every step. The estate has several hundred VMs in tightly coupled groups. Which migration design fits?
VCDA keeps the VMware operating model, replicates asynchronously in the background, and reduces each cutover to the final synchronisation and boot, while reverse failover gives every wave a way back. Waves follow dependency clusters so that no group is split across the hybrid link. Image conversion makes the downtime the whole disk copy and changes the guest, a single big-bang wave has no contained rollback, and moving databases ahead of their applications splits every cluster across the link.
The two Windows servers of FinCorp's payment switch boot with legacy BIOS on-premises. They must move to Dedicated Core servers on a CPU family that requires UEFI, and the team wants the guest adapted without rebuilding the operating systems. Which path should the architect choose?
A full machine backup and recovery is the documented route across boot firmware and CPU generations: the recovery adapts the boot partition for the BIOS-to-UEFI change, and Universal Restore injects the VirtIO storage drivers for Windows. The image metadata flag only declares that an image is UEFI-compatible; it does not convert the guest. Storage re-attachment and snapshots require source and target to support the same boot firmware, and VCDA keeps workloads on VMware rather than moving them to VDC compute.