[gpum] Derive gpu.sm_active from GPM SM cycle counters

[martavicentenavarro]

What does this PR do?

Adds a source for gpu.sm_active computed from the GPM counters SM_CYCLES_ELAPSED (248) and SM_CYCLES_ACTIVE (249) on physical GPUs.

NVML returns these counters as cumulative values at Sample2 (not as the difference between the two samples), so the GPM collector keeps the previous reading and reports 100 * Δactive / Δelapsed.

• Low priority by default (fallback only), High with the new gpu.prefer_sm_cycles_sm_active option.
• Not computed when gpu.legacy_sm_active is enabled (the legacy value takes precedence) nor on MIG devices, where the counters haven't been validated.
• RemoveDuplicateSamples now resolves equal-priority ties by collector name instead of map iteration order, so sm_active can't alternate between the ebpf and gpm sources at Low. No other metric is affected: there were no equal-priority ties across collectors before this change.
• The counter IDs are defined locally because go-nvml v0.13.1-0 stops at ID 210.

Motivation

AXT-33

Describe how you validated your changes

Unit tests

Manual validation on an H100 80GB HBM3 (driver 595.91.07, NVML 13.595.91.07) with agent check gpu:

• Querying NVML directly confirmed the counters are cumulative and that the value is returned for Sample2.
• Idle: sm_active = 0.
• gpu-burner at 60% (gpu-burner --run_time 300 auto --target_sm 60):
  • default config: sm_active = 61–62 from the sampling source;
  • gpu.prefer_sm_cycles_sm_active: true: sm_active = 59.3–59.6 from the new source.
• With 33 of 132 SMs busy: the new source reports 25%, while gr_engine_active is ~100%.
• 300 windows under full saturation: Δactive never exceeded Δelapsed.

Additional Notes

• The derived value is the same as GPM_METRIC_SM_UTIL (SM activity averaged over all SMs), not the percentage of time any SM was active, so with the option enabled gpu.sm_active matches gpu.sm_utilization.
• With the default config the new source adds two GpmMetricsGet calls per GPM device per run, and its sample is deduplicated, which increases duplicate_metrics telemetry by one per device.
• Not validated on MIG (disabled there) nor with system-probe/ebpf on hardware.

[github-actions]

AI review by Codex (OpenAI) - workflow run

The patch is correct. No actionable issues were found in the counter calculations, source priority handling, configuration wiring, or tests. Validation was limited to static review; tests were not run in the read-only environment.

[datadog-prod-us1-6]

🎯 Code Coverage (details)
• Patch Coverage: 97.30%
• Overall Coverage: 58.62% (+0.01%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 189db58 | Docs | Give us feedback!}

[dd-octo-sts]

Files inventory check summary

File checks results against ancestor 66e6c51f:

Results for datadog-agent_7.86.0~devel.git.306.189db58.pipeline.143779268-1_amd64.deb:

No change detected

Results for datadog-iot-agent_7.86.0~devel.git.306.189db58.pipeline.143779268-1_amd64.deb:

No change detected

[dd-octo-sts]

Static quality checks

✅ Please find below the results from static quality gates
Comparison made with ancestor 66e6c51
📊 Static Quality Gates Dashboard
🔗 SQG Job

Successful checks

Info

	Quality gate	Change	Size (prev → curr → max)
✅	agent_deb_amd64	+8.41 KiB (0.00% increase, -0.16% of buffer)	755.256 → 755.264 → 760.520
✅	agent_deb_amd64_fips	+12.41 KiB (0.00% increase, -0.23% of buffer)	688.278 → 688.290 → 693.570
✅	agent_rpm_amd64	+8.41 KiB (0.00% increase, -0.16% of buffer)	755.240 → 755.248 → 760.500
✅	agent_rpm_amd64_fips	+12.41 KiB (0.00% increase, -0.23% of buffer)	688.261 → 688.274 → 693.550
✅	agent_rpm_arm64	+8.4 KiB (0.00% increase, -0.15% of buffer)	725.784 → 725.793 → 731.100
✅	agent_rpm_arm64_fips	+8.06 KiB (0.00% increase, -0.15% of buffer)	664.371 → 664.379 → 669.740
✅	agent_suse_amd64	+8.41 KiB (0.00% increase, -0.16% of buffer)	755.240 → 755.248 → 760.500
✅	agent_suse_amd64_fips	+12.41 KiB (0.00% increase, -0.23% of buffer)	688.261 → 688.274 → 693.550
✅	agent_suse_arm64	+8.4 KiB (0.00% increase, -0.15% of buffer)	725.784 → 725.793 → 731.100
✅	agent_suse_arm64_fips	+8.06 KiB (0.00% increase, -0.15% of buffer)	664.371 → 664.379 → 669.740
✅	docker_agent_amd64	+8.41 KiB (0.00% increase, -0.16% of buffer)	811.423 → 811.431 → 816.720
✅	docker_agent_arm64	+8.41 KiB (0.00% increase, -0.15% of buffer)	806.590 → 806.598 → 811.970
✅	docker_agent_jmx_amd64	+8.42 KiB (0.00% increase, -0.16% of buffer)	1002.292 → 1002.301 → 1007.590
✅	docker_agent_jmx_arm64	+8.4 KiB (0.00% increase, -0.15% of buffer)	986.240 → 986.248 → 991.620
✅	docker_host_profiler_arm64	-64.22 KiB (0.02% reduction, +8.77% of buffer)	319.865 → 319.802 → 320.580

18 successful checks with minimal change (< 2 KiB)

	Quality gate	Current Size
✅	agent_heroku_amd64	319.145 MiB
✅	agent_msi	653.833 MiB
✅	docker_cws_instrumentation_amd64	7.443 MiB
✅	docker_cws_instrumentation_arm64	6.877 MiB
✅	docker_dogstatsd_amd64	39.543 MiB
✅	docker_dogstatsd_arm64	37.629 MiB
✅	docker_host_profiler_amd64	308.722 MiB
✅	dogstatsd_deb_amd64	30.290 MiB
✅	dogstatsd_deb_arm64	28.278 MiB
✅	dogstatsd_rpm_amd64	30.290 MiB
✅	dogstatsd_suse_amd64	30.290 MiB
✅	iot_agent_deb_amd64	47.371 MiB
✅	iot_agent_deb_arm64	43.838 MiB
✅	iot_agent_deb_armhf	44.653 MiB
✅	iot_agent_rpm_amd64	47.372 MiB
✅	iot_agent_suse_amd64	47.371 MiB
✅	docker_cluster_agent_amd64	211.197 MiB
✅	docker_cluster_agent_arm64	224.162 MiB

[cit-pr-commenter-54b7da]

Regression Detector

Regression Detector Results

Metrics dashboard
Target profiles
Job ID: bb0ad39f-d238-46b1-820a-a7790fcc7085

Baseline: 66e6c51
Comparison: 189db58
Diff

Optimization Goals: ✅ No significant changes detected

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	dsd_uds_10mb_3k_timestamped_contexts_memory	memory utilization	+1.62	[+1.40, +1.84]	1	Logs
➖	quality_gate_security_no_fs_load	memory utilization	+0.26	[+0.19, +0.34]	1	Logs bounds checks dashboard
➖	quality_gate_idle	memory utilization	+0.26	[+0.21, +0.30]	1	Logs bounds checks dashboard
➖	dsd_uds_10mb_3k_timestamped_contexts_cpu	% cpu utilization	-0.05	[-0.29, +0.19]	1	Logs
➖	quality_gate_idle_all_features	memory utilization	-0.07	[-0.15, +0.00]	1	Logs bounds checks dashboard
➖	quality_gate_security_idle	memory utilization	-0.18	[-0.21, -0.14]	1	Logs bounds checks dashboard
➖	python_openmetrics	% cpu utilization	-0.22	[-0.90, +0.46]	1	Logs bounds checks dashboard
➖	quality_gate_metrics_logs	memory utilization	-0.38	[-0.60, -0.15]	1	Logs bounds checks dashboard
➖	quality_gate_logs	% cpu utilization	-0.49	[-1.34, +0.36]	1	Logs bounds checks dashboard
➖	quality_gate_security_mean_fs_load	memory utilization	-0.58	[-0.62, -0.55]	1	Logs bounds checks dashboard
➖	quality_gate_private_action_runner	memory utilization	-0.63	[-0.75, -0.51]	1	Logs bounds checks dashboard
➖	dsd_uds_client_drop_detector_cpu	% cpu utilization	-0.99	[-1.48, -0.49]	1	Logs

Bounds Checks: ✅ Passed

perf	experiment	bounds_check_name	replicates_passed	observed_value	links
✅	python_openmetrics	checks_execution_time	10/10	80.05 ≤ 100	bounds checks dashboard
✅	python_openmetrics	cpu_usage	10/10	1340.66 ≤ 1500	bounds checks dashboard
✅	python_openmetrics	memory_usage	10/10	4.31GiB ≤ 4.75GiB	bounds checks dashboard
✅	quality_gate_idle	intake_connections	10/10	4 ≤ 5	bounds checks dashboard
✅	quality_gate_idle	memory_usage	10/10	179.54MiB ≤ 181MiB	bounds checks dashboard
✅	quality_gate_idle	total_bytes_received	10/10	774.59KiB ≤ 819.20KiB	bounds checks dashboard
✅	quality_gate_idle_all_features	intake_connections	10/10	2 ≤ 5	bounds checks dashboard
✅	quality_gate_idle_all_features	memory_usage	10/10	471.30MiB ≤ 542MiB	bounds checks dashboard
✅	quality_gate_idle_all_features	total_bytes_received	10/10	1.14MiB ≤ 1.25MiB	bounds checks dashboard
✅	quality_gate_logs	intake_connections	10/10	18 ≤ 40	bounds checks dashboard
✅	quality_gate_logs	memory_usage	10/10	212.05MiB ≤ 228MiB	bounds checks dashboard
✅	quality_gate_logs	missed_bytes	10/10	0B = 0B	bounds checks dashboard
✅	quality_gate_logs	total_bytes_received	10/10	263.88MiB ≤ 292MiB	bounds checks dashboard
✅	quality_gate_metrics_logs	cpu_usage	10/10	371.56 ≤ 2000	bounds checks dashboard
✅	quality_gate_metrics_logs	intake_connections	10/10	20 ≤ 40	bounds checks dashboard
✅	quality_gate_metrics_logs	memory_usage	10/10	412.82MiB ≤ 455MiB	bounds checks dashboard
✅	quality_gate_metrics_logs	missed_bytes	10/10	0B = 0B	bounds checks dashboard
✅	quality_gate_metrics_logs	total_bytes_received	10/10	0.95GiB ≤ 1.04GiB	bounds checks dashboard
✅	quality_gate_private_action_runner	memory_usage	10/10	73.62MiB ≤ 77MiB	bounds checks dashboard
✅	quality_gate_security_idle	cpu_usage	10/10	30.57 ≤ 100	bounds checks dashboard
✅	quality_gate_security_idle	memory_usage	10/10	326.95MiB ≤ 357MiB	bounds checks dashboard
✅	quality_gate_security_mean_fs_load	cpu_usage	10/10	63.11 ≤ 200	bounds checks dashboard
✅	quality_gate_security_mean_fs_load	memory_usage	10/10	306.43MiB ≤ 337MiB	bounds checks dashboard
✅	quality_gate_security_no_fs_load	cpu_usage	10/10	22.62 ≤ 100	bounds checks dashboard
✅	quality_gate_security_no_fs_load	memory_usage	10/10	338.43MiB ≤ 348MiB	bounds checks dashboard

Explanation

Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%

Performance changes are noted in the perf column of each table:

• ✅ = significantly better comparison variant performance
• ❌ = significantly worse comparison variant performance
• ➖ = no significant change in performance

A regression test is an A/B test of target performance in a repeatable rig, where "performance" is measured as "comparison variant minus baseline variant" for an optimization goal (e.g., ingress throughput). Due to intrinsic variability in measuring that goal, we can only estimate its mean value for each experiment; we report uncertainty in that value as a 90.00% confidence interval denoted "Δ mean % CI".

For each experiment, we decide whether a change in performance is a "regression" -- a change worth investigating further -- if all of the following criteria are true:

1. Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.

2. Its 90.00% confidence interval "Δ mean % CI" does not contain zero, indicating that if our statistical model is accurate, there is at least a 90.00% chance there is a difference in performance between baseline and comparison variants.

3. Its configuration does not mark it "erratic".

CI Pass/Fail Decision

✅ Passed. All Quality Gates passed.

• quality_gate_private_action_runner, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_security_idle, bounds check cpu_usage: 10/10 replicas passed. Gate passed.
• quality_gate_security_idle, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check total_bytes_received: 10/10 replicas passed. Gate passed.
• quality_gate_idle_all_features, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_idle_all_features, bounds check total_bytes_received: 10/10 replicas passed. Gate passed.
• quality_gate_idle_all_features, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_security_mean_fs_load, bounds check cpu_usage: 10/10 replicas passed. Gate passed.
• quality_gate_security_mean_fs_load, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check missed_bytes: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check total_bytes_received: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check cpu_usage: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check missed_bytes: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check total_bytes_received: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_security_no_fs_load, bounds check cpu_usage: 10/10 replicas passed. Gate passed.
• quality_gate_security_no_fs_load, bounds check memory_usage: 10/10 replicas passed. Gate passed.