Task · SKT-0087.05

Emit NVIDIA driver kernel events (Xid, SXid, NVRM) in GPU host logs

Description

The GPU fault workflow starts from an Xid burst and pivots from the DCGM Xid counter to the driver’s kernel log lines for the same GPU and time window. The host construct already emits journald streams (signals/host.md) but no NVIDIA driver content, and NVSwitch-bearing systems also log through the fabric manager service (SXid). Without these lines the GPU fault, failed job and support-evidence journeys stop at a metric.

Acceptance Criteria

Definition of Done

Implementation Notes

loop47: admitted R05 pinned NVIDIA driver-log contract research; implementation waits for fixture/DCGM landing

loop47 dependency park, implementation0: DCGM consumer SKT-0087.04 (GPU exporter profiles) source-blocked before code. Resume after its source-correct green landing; I05 dual host/k8s log adapters serialize before I06 lifecycle/KSM/capacity/operator adapters. Generic fixture Done does not release these consumer joins or authorize invented vendor events.

loop48 in-loop dependency park, no admitted implementation/attempt consumed: narrowed DCGM prerequisite is Parked after admitted attempt and independent owner/seam diagnosis. Resume only after DCGM lands with sourced physical constraint/clock semantics and applicable proof; then retain host/node-log ownership and public Xid/SXid/NVRM projection checks. No driver-log source support inferred from the HGX fixture alone.

References

View the source file on GitHub