docs(gpu-operator): reflect recording-rule dependency for gpu-quotas

Signed-off-by: Arsolitt <arsolitt@gmail.com>
This commit is contained in:
Arsolitt 2026-04-19 11:08:47 +03:00
parent a3241bf51b
commit 2cc60f170c
No known key found for this signature in database
GPG key ID: 4D8302CE6A9247C4

View file

@ -81,7 +81,7 @@ What each dashboard needs on top of the upstream DCGM Exporter
| `gpu-performance` | Per-node, per-GPU deep dive | `DCGM_FI_DEV_POWER_VIOLATION`, `DCGM_FI_DEV_THERMAL_VIOLATION` |
| `gpu-efficiency` | Per-workload util vs tensor active | `DCGM_FI_DEV_POWER_VIOLATION`, `DCGM_FI_DEV_THERMAL_VIOLATION` (via `gpu:*_throttle_fraction:rate5m` recording rules) |
| `gpu-fleet` | Cluster-wide admin inventory | `DCGM_FI_DEV_POWER_MGMT_LIMIT` (for the TDP vs draw panel) |
| `gpu-quotas` | Kube-quota vs live usage | nothing (kube-state-metrics + default counters) |
| `gpu-quotas` | Kube-quota vs live usage | `kube_pod_container_resource_requests`, `kube_pod_status_phase`, `DCGM_FI_DEV_GPU_UTIL` (via `namespace:gpu_count:allocated` / `cluster:gpu_count:*` recording rules) |
| `gpu-tenants` | Per-namespace tenant view | nothing (works on default counters) |
`DCGM_FI_PROF_PIPE_TENSOR_ACTIVE` and `DCGM_FI_PROF_GR_ENGINE_ACTIVE`