Knowledge Check - Data and Storage
Test your understanding of the key concepts from Module 5. 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 reads customer profiles and product catalogues far more often than it writes them. The primary of the three-instance DBaaS for PostgreSQL cluster is near its CPU limit at peak, and the auto-scaled portal tier opens more connections every month. Which change should the architect recommend?
PostgreSQL standbys are warm standbys that do not accept read connections, and horizontal scaling adds availability, not performance, so extra instances only add cost. A cache takes repeated reads off the primary, and the pooler lets many client connections share fewer database connections below the RAM-derived ceiling. MariaDB scaling behaves the same way, and dropping to one instance gives up automatic failover without adding read capacity.
FinCorp's risk office requires that a database failover never loses a confirmed payment, and that the payment database keeps accepting writes while one node is in maintenance. How should the architect configure the DBaaS for PostgreSQL cluster?
Strictly Synchronous mode confirms a commit only after a synchronous standby has it, so a failover loses no confirmed data; it needs at least three instances and two operating nodes to accept writes, which keeps writes flowing with one node down. Asynchronous mode can lose non-replicated transactions at failover regardless of the number of standbys, and PITR can only restore what the uploaded WAL contains. Two instances fall below the mode's minimum.
FinCorp's audit archive is a contract-owned bucket in eu-central-4 with Object Lock in Compliance mode. The regulator now asks for a second, equally immutable copy in another region. Which design meets the requirement?
A bucket with Object Lock cannot be a replication source, and cross-regional replication is not supported for contract-owned buckets, so the second copy must come from the writer. Snapshots are bound to the location where they were created and copy a volume, not the archive. Only the Standard storage class exists, so lifecycle rules delete objects but cannot move them to another class or region.
The portal servers fc-app-01 and fc-app-02 in Frankfurt need the same templates and assets, and the Berlin recovery VDC must be able to serve the same files after a failover. Which storage design fits?
NFS gives the Frankfurt pair one shared copy with POSIX permissions and no data transfer charge, while a cluster serves only the private LAN it is attached to, so cross-region access belongs to Object Storage. A Block Storage volume is a server's own device, and snapshots stay in their creation location. Per-server volumes with a sync job create exactly the drift that shared file storage avoids.
FinCorp publishes payment events to Event Streams for Apache Kafka. Each account's events must be processed in order, the fraud service must scale out to six consumers at peak, and the cluster has the standard three brokers. How should the topic be designed?
Kafka orders messages within a partition, so keying by account ID keeps each account's events in order while different accounts are processed in parallel, and six partitions allow six consumers in one group. Six is a multiple of the three brokers, which avoids an uneven distribution, and a replication factor of 3 is the recommended setting. Random keys break per-account ordering, one partition caps parallelism at one consumer, and adding partitions later moves keys and breaks per-key ordering.
FinCorp's onboarding case files in a MongoDB cluster must be retained for ten years, and the cluster may be replaced by a new one during that period. Which protection design should the architect choose?
MongoDB snapshots are retained for seven days in every edition, and destroying a cluster deletes all of its snapshots, so long retention needs dumps that live outside the cluster, and Object Lock in Compliance mode makes them impossible to delete early. Point-in-time recovery is an Enterprise feature that reaches back up to 24 hours, not years, and Business does not offer it. Neither the snapshots nor PITR survive the replacement of the cluster.