Oracle NoSQL Database Cloud Service is a fully managed, serverless database. The service abstracts the underlying hardware infrastructure, including compute instances and their CPU utilization, from the user. Users manage performance by provisioning throughput (Read and Write Capacity Units) and storage. Autoscaling is triggered by the consumption of these provisioned units. Metrics like rejected (throttled) requests and operation latency are direct indicators of whether the provisioned capacity is sufficient for the workload. CPU utilization of the host instances is an internal Oracle-managed metric and is not exposed to the user or used as a basis for customer-configured autoscaling.

B. Number of rejected read requests due to insufficient RCUs.

This is a primary metric (ReadThrottleCount) indicating the provisioned read capacity is exceeded, making it highly relevant for scaling up.

C. Number of rejected write requests due to insufficient WCUs.

This is a primary metric (WriteThrottleCount) indicating the provisioned write capacity is exceeded, making it highly relevant for scaling up.

D. Latency of read and write operations.

High latency is a key performance indicator that the system is approaching its capacity limits, making it a very relevant metric for scaling decisions.

1. Oracle Cloud Infrastructure Documentation, About Oracle NoSQL Database Cloud Service: "Oracle NoSQL Database Cloud Service is a fully managed database service that handles the administration of the underlying infrastructure, including hardware and software provisioning, and scaling." This statement confirms the abstraction of hardware metrics like CPU from the user.

Source: Oracle Cloud Infrastructure Documentation, "Overview of NoSQL Database Cloud Service," section "About Oracle NoSQL Database Cloud Service."

2. Oracle Cloud Infrastructure Documentation, NoSQL Database Metrics: This document lists the available metrics for the NoSQL Database service. The list includes ReadThrottleCount, WriteThrottleCount, GetRequestLatency, and PutRequestLatency. It does not include any metric related to the CPU utilization of the underlying service infrastructure, confirming that this is not a user-facing metric for monitoring or autoscaling.

Source: Oracle Cloud Infrastructure Documentation, "Monitoring," section "NoSQL Database Metrics."

3. Oracle Cloud Infrastructure Documentation, Estimating and Monitoring Capacity: "If you exceed the provisioned throughput for a table for an extended period of time, you will experience a ThrottlingException for the excess requests." This directly links rejected requests (throttling) to exceeding provisioned capacity, establishing the relevance of options B and C.

Source: Oracle Cloud Infrastructure Documentation, "Using Oracle NoSQL Database Cloud Service," section "Estimating and Monitoring Capacity."