From 967d9a9f4ea12815bc28f909c5fd0110f4955845 Mon Sep 17 00:00:00 2001 From: Alex Telpis Date: Wed, 15 Oct 2025 12:07:58 -0500 Subject: [PATCH] Whitelist AMD power and throttle related metrics --- cmd/do-agent/aggregation.go | 2 ++ cmd/do-agent/whitelist.go | 2 ++ 2 files changed, 4 insertions(+) diff --git a/cmd/do-agent/aggregation.go b/cmd/do-agent/aggregation.go index b21f8b7..5c3c6e8 100644 --- a/cmd/do-agent/aggregation.go +++ b/cmd/do-agent/aggregation.go @@ -187,8 +187,10 @@ var gpuAggregationSpec = map[string][]string{ "amd_gpu_violation_ppt_residency_accumulated": amdAggregatedLabels, "amd_gpu_violation_hbm_thermal_residency_accumulated": amdAggregatedLabels, "amd_gpu_violation_vr_thermal_tracking_accumulated": amdAggregatedLabels, + "amd_gpu_violation_current_accumulated_counter": amdAggregatedLabels, "amd_gpu_junction_temperature": amdAggregatedLabels, "amd_gpu_power_usage": amdAggregatedLabels, + "amd_gpu_package_power": amdAggregatedLabels, "amd_gpu_memory_temperature": amdAggregatedLabels, "amd_gpu_gfx_activity": amdAggregatedLabels, "amd_gpu_prof_sm_active": amdAggregatedLabels, diff --git a/cmd/do-agent/whitelist.go b/cmd/do-agent/whitelist.go index 9cbad10..d81b931 100644 --- a/cmd/do-agent/whitelist.go +++ b/cmd/do-agent/whitelist.go @@ -238,11 +238,13 @@ var gpuWhitelist = map[string]bool{ "amd_gpu_clock": true, "amd_gpu_violation_proc_hot_residency_accumulated": true, "amd_gpu_violation_soc_thermal_residency_accumulated": true, + "amd_gpu_violation_current_accumulated_counter": true, "amd_gpu_violation_ppt_residency_accumulated": true, "amd_gpu_violation_hbm_thermal_residency_accumulated": true, "amd_gpu_violation_vr_thermal_tracking_accumulated": true, "amd_gpu_junction_temperature": true, "amd_gpu_power_usage": true, + "amd_gpu_package_power": true, "amd_gpu_memory_temperature": true, "amd_gpu_gfx_activity": true, "amd_gpu_prof_sm_active": true,