Passive Health Checks with Oracle Cloud Agent
Oracle Cloud Agent passive checks examine supported GPU-node hardware and software state without generating active workload traffic.
What Passive Checks Examine
The passive-check catalog covers the following areas. Specific checks run only where they apply to the shape and are enabled in the deployed release.
| Area | Examples of evidence examined |
|---|---|
| GPU inventory and software readiness | Expected GPU count, driver state and supported versions, GPU operating mode and clocks, required peer-memory support, and fabric services. |
| GPU memory and error state | NVIDIA XID events, correctable and uncorrectable memory errors, row-remap state, retired or bad memory pages, and SRAM error counters. |
| GPU interconnects and PCIe | NVLink or AMD xGMI state where applicable, reported link errors, expected PCIe speed and width, and missing devices in the PCIe topology. |
| Network and remote direct memory access (RDMA) | Expected network adapter count, link state and speed, authentication, RDMA configuration, link flaps, discarded packets, and physical-link error counters. |
These are observations of the instance and its devices. They don't generate the sustained GPU or network load used by active tests.
Test Catalog and Enablement Matrix
To learn about the health checks available for your GPU shape, review the test catalog and test enablement matrix. These documents are included with the container image and are the definitive reference for the checks provided by that specific release. See Find Tests Supported by Your Configuration.
How Passive Checks Improve Fault and Repair Decisions
A raw device event isn't always enough to decide whether a node needs a reboot, a configuration change, or hardware repair. Passive checks add context such as the expected device count, the affected component, and the relevant error condition. The resulting fault and recommendation give the operator and repair integration more specific evidence for the next action.
For example, a GPU-count check can identify a node that exposes fewer GPUs than its shape requires. The recommendation can be to reboot and check again, with hardware repair required if the GPU remains unavailable. This avoids treating every instance-level problem as an immediate hardware replacement.
Follow the reported recommendation and the applicable OCI repair guidance. An unavailable or failed test isn't a healthy result, and a successful reboot doesn't prove that the underlying fault has cleared.