[CONTP-499] Parsing GPU tags on kubeapiserver collector

[gabedos]

What does this PR do?

• Allow GPU tag parsing on cluster agent's kube-apiserver collector.
• Migrate GPU name conversion to shared location

Motivation

• Support gpu tagger on the cluster agent and cluster check runners
• Allow flexibility for other components to use the same GPU naming conventions

Describe how to test/QA your changes

Deploy basic agent configuration with cluster tagger

datadog:
  kubelet:
    tlsVerify: false
  clusterName: <INSERT_CLUSTER_NAME>
  clusterTagger:
    collectKubernetesTags: true
clusterAgent:
  image:
    tag: <INSERT_TAG>
    pullPolicy: IfNotPresent
  enabled: true
  replicas: 1

Deploy a dummy GPU workload k apply -f deployment.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: dummy-nginx-app
spec:
  replicas: 1
  selector:
    matchLabels:
      app: dummy-nginx-app
  template:
    metadata:
      labels:
        app: dummy-nginx-app
    spec:
      containers:
        - name: dummy-nginx-app
          image: nginx
          resources:
            requests:
              memory: "64Mi"
              cpu: "250m"
              nvidia.com/mig-something: "0"
              amd.com/gpu: "0"
              gpu.intel.com/xe: "0"
            limits:
              memory: "128Mi"
              cpu: "500m"
              nvidia.com/mig-something: "0"
              amd.com/gpu: "0"
              gpu.intel.com/xe: "0"

Check for the GPU tags on the cluster agent

k exec -it datadog-cluster-agent-xxxxx -- agent tagger-list
=== Entity kubernetes_pod_uid://46bc5ffe-a5b9-45f3-98b9-8fd64aac1e36 ===
== Source workloadmeta-kubernetes_pod =
=Tags: [gpu_vendor:amd gpu_vendor:intel gpu_vendor:nvidia kube_cluster_name:gabedos-test-cluster kube_deployment:dummy-nginx-app kube_namespace:default kube_ownerref_kind:replicaset kube_ownerref_name:dummy-nginx-app-6d47fbbbc5 kube_qos:Burstable kube_replica_set:dummy-nginx-app-6d47fbbbc5 pod_name:dummy-nginx-app-6d47fbbbc5-ggjb7 pod_phase:running]
===

Possible Drawbacks / Trade-offs

Additional Notes

Considered creating 1 ParsePods method shared across kubeapiserver and kubelet collectors however the overhead work to convert the types or implement interfaces seem like more work than supporting the two separate parsers.

[cit-pr-commenter]

Go Package Import Differences

Baseline: 5e0c347
Comparison: 2c5d869

binary	os	arch	change
agent	linux	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
agent	linux	arm64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
agent	windows	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
agent	darwin	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
agent	darwin	arm64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
cluster-agent	linux	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
cluster-agent	linux	arm64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
dogstatsd	linux	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
dogstatsd	linux	arm64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
process-agent	linux	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
process-agent	linux	arm64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
process-agent	windows	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
process-agent	darwin	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
process-agent	darwin	arm64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
security-agent	linux	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
security-agent	linux	arm64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu
security-agent	windows	amd64	+1, -0 +github.com/DataDog/datadog-agent/pkg/util/gpu

[cit-pr-commenter]

Regression Detector

Regression Detector Results

Metrics dashboard
Target profiles
Run ID: d2dfc9c6-f0f4-43c7-a756-f444387214eb

Baseline: 5e0c347
Comparison: 2c5d869
Diff

Optimization Goals: ✅ Improvement(s) detected

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
✅	basic_py_check	% cpu utilization	-8.65	[-12.29, -5.00]	1	Logs

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	otel_to_otel_logs	ingress throughput	+1.88	[+1.17, +2.59]	1	Logs
➖	quality_gate_idle_all_features	memory utilization	+1.17	[+1.06, +1.29]	1	Logs bounds checks dashboard
➖	file_to_blackhole_500ms_latency	egress throughput	+0.41	[-0.37, +1.18]	1	Logs
➖	file_to_blackhole_0ms_latency	egress throughput	+0.16	[-0.69, +1.01]	1	Logs
➖	quality_gate_idle	memory utilization	+0.13	[+0.09, +0.18]	1	Logs bounds checks dashboard
➖	file_to_blackhole_1000ms_latency_linear_load	egress throughput	+0.11	[-0.35, +0.58]	1	Logs
➖	file_to_blackhole_300ms_latency	egress throughput	+0.09	[-0.54, +0.73]	1	Logs
➖	file_to_blackhole_1000ms_latency	egress throughput	+0.05	[-0.74, +0.84]	1	Logs
➖	file_to_blackhole_100ms_latency	egress throughput	+0.03	[-0.66, +0.72]	1	Logs
➖	uds_dogstatsd_to_api	ingress throughput	+0.01	[-0.07, +0.09]	1	Logs
➖	tcp_dd_logs_filter_exclude	ingress throughput	+0.00	[-0.01, +0.01]	1	Logs
➖	tcp_syslog_to_blackhole	ingress throughput	-0.01	[-0.07, +0.05]	1	Logs
➖	file_tree	memory utilization	-0.10	[-0.24, +0.05]	1	Logs
➖	uds_dogstatsd_to_api_cpu	% cpu utilization	-0.13	[-0.86, +0.60]	1	Logs
➖	pycheck_lots_of_tags	% cpu utilization	-0.31	[-3.76, +3.14]	1	Logs
➖	quality_gate_logs	% cpu utilization	-0.57	[-3.54, +2.40]	1	Logs
✅	basic_py_check	% cpu utilization	-8.65	[-12.29, -5.00]	1	Logs

Bounds Checks: ❌ Failed

perf	experiment	bounds_check_name	replicates_passed	links
❌	file_to_blackhole_500ms_latency	lost_bytes	9/10
✅	file_to_blackhole_0ms_latency	lost_bytes	10/10
✅	file_to_blackhole_0ms_latency	memory_usage	10/10
✅	file_to_blackhole_1000ms_latency	memory_usage	10/10
✅	file_to_blackhole_1000ms_latency_linear_load	memory_usage	10/10
✅	file_to_blackhole_100ms_latency	lost_bytes	10/10
✅	file_to_blackhole_100ms_latency	memory_usage	10/10
✅	file_to_blackhole_300ms_latency	lost_bytes	10/10
✅	file_to_blackhole_300ms_latency	memory_usage	10/10
✅	file_to_blackhole_500ms_latency	memory_usage	10/10
✅	quality_gate_idle	memory_usage	10/10	bounds checks dashboard
✅	quality_gate_idle_all_features	memory_usage	10/10	bounds checks dashboard
✅	quality_gate_logs	lost_bytes	10/10
✅	quality_gate_logs	memory_usage	10/10

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_idle_all_features, 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_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check lost_bytes: 10/10 replicas passed. Gate passed.

[agent-platform-auto-pr]

Test changes on VM

Use this command from test-infra-definitions to manually test this PR changes on a VM:

inv aws.create-vm --pipeline-id=50334515 --os-family=ubuntu

Note: This applies to commit 2c5d869

[clamoriniere]

can we wrap this logic in one function to be reused in comp/core/workloadmeta/collectors/util/kubernetes_resource_parsers/pod.go ?
I feel like the code is very similar

[gabedos]

I've tried implementing a function such as the following:

func GetGPUResourcesFromResourceList[T corev1.ResourceList | kubelet.ResourceList](gpuSet *map[string]struct{}, resourceList T) {
	for resourceName := range resourceList {
		gpuName, found := ExtractSimpleGPUName(ResourceGPU(resourceName))
		if found {
			(*gpuSet)[gpuName] = struct{}{}
		}
	}
}

However, I get an error stating "cannot range over resourceList: no core type". It seems that Go's type system cannot infer the underlying map structure directly from T corev1.ResourceList | kubelet.ResourceList.
I've tried directly setting it to [T ~map[corev1.ResourceName]resource.Quantity | ~map[kubelet.ResourceName]resource.Quantity] and get the same issue. It's interesting that corev1.ResourceName and kubelet.ResourceName are both simply strings, so these types are essentially the same. I tried type casting the input map[string]resource.Quantity(container.Resources.Requests) but it doesn't allow me either.

I can't think of a lightweight way for wrapping this logic into one function that can be reused. Do you have any ideas? It would be very appreciated. Thanks!

[gabedos]

/merge

[dd-devflow]

Devflow running: /merge

View all feedbacks in Devflow UI.

---

2024-12-05 17:58:34 UTC ℹ️ MergeQueue: pull request added to the queue

The median merge time in main is 23m.