The primary gateway for this data is the OCI Monitoring API, which categorizes hardware health into specific "Namespaces."

1. The Infrastructure Health API

This is the most critical API for monitoring the physical state of your rented servers. It uses the oci_compute_infrastructure_health namespace.

• Metric Names: * instance_status: Returns 0 for healthy and 1 for hardware-level failure.

  • maintenance_status: Indicates if the physical host is scheduled for a firmware or hardware update.

  • health_error_count: Specifically for Bare Metal, this reports the number of detected physical defects on the motherboard or power supply.

• Use Case: Automating "failover" scripts. If the API returns a 1 for maintenance status, your script can automatically spin up a new instance and move the traffic.

2. The GPU & HPC Telemetry API

For AI and high-performance computing, OCI exposes specialized hardware metrics via the gpu_infrastructure_health and rdma_infrastructure_health namespaces.

• Hardware Sensors:

  • Thermal: gpu_temperature (returns the maximum Celsius reading).

  • Power: gpu_power_draw (returns wattage consumption).

  • Memory Integrity: gpu_ecc_double_bit_errors (crucial for detecting failing VRAM in AI clusters).

• Interconnect Health: The RDMA API tracks rdma_link_speed_fault and rdma_pcie_ber_check_fault, ensuring your multi-node AI training cluster isn't being slowed down by a bad cable or a loose PCIe connection.

3. The Block Volume Performance API

To monitor the health of your "rented" storage, the oci_blockstore namespace provides hardware-level I/O metrics.

• Key Metrics: VolumeThroughput and VolumeIops.

• Latency Monitoring: Use the VolumeLatency metric via the API to detect if the physical SSD array is experiencing "micro-spikes" that could affect your database performance.

API Endpoint Summary (2026)

4. How to call these APIs

In 2026, most developers use the OCI SDK (Python, Java, or Go) or the OCI CLI to pull this data. A typical CLI call to check your hardware health looks like this:

# Example: Checking for hardware-level accessibility issues
oci monitoring metric-data summarize-metrics-data \
    --namespace oci_compute_instance_health \
    --query-text "instance_accessibility_status[1m].max()" \
    --compartment-id <your_ocid>

5. OpenTelemetry & Integration (2026 Shift)

The biggest change in 2026 is the support for OpenTelemetry (OTLP).

• You can now configure the Oracle Cloud Agent on your server to stream these hardware metrics directly to an OTel Collector.

• This allows you to view your GPU temperatures and physical server health directly in third-party tools like Grafana, Datadog, or New Relic without writing custom API polling scripts.

Pro-Tip: The "Metric-Data Post" API

If you have custom hardware sensors (e.g., in a Cloud@Customer on-premises deployment), you can use the PostMetricData API to push your own hardware telemetry into OCI Monitoring. This lets you see your local data-center temperature alongside your OCI cloud metrics in a single pane of glass.