[k8sattributes] Revisit regex based metadata extraction

[ChrsMark]

As it was mentioned at #23067 (comment) recently, we should ensure that metadata extraction is consistent for different types of workloads.

Right now we use some regexp patterns for extracting owners' names:

opentelemetry-collector-contrib/processor/k8sattributesprocessor/internal/kube/client.go

Lines 135 to 145 in 07228c0

	// Extract replicaset name from the pod name. Pod name is created using
	// format: [deployment-name]-[Random-String-For-ReplicaSet]
	var rRegex = regexp.MustCompile(`^(.*)-[0-9a-zA-Z]+$`)
	// Extract CronJob name from the Job name. Job name is created using
	// format: [cronjob-name]-[time-hash-int]
	var cronJobRegex = regexp.MustCompile(`^(.*)-\d+$`)
	// Extract Deployment name from the ReplicaSet name. Deployment name is created using
	// format: [deployment-name]-[hash]
	var deploymentHashSuffixPattern = regexp.MustCompile(`^[a-z0-9]{10}$`)

#42534 added support for extracting the deployment name from the replicaset's name behind a setting. But for other workloads that's not configurable ie:

opentelemetry-collector-contrib/processor/k8sattributesprocessor/internal/kube/client.go

Line 892 in 9cbe6fb

parts := c.cronJobRegex.FindStringSubmatch(ref.Name)

We should revisit this and aim for consistency.

As it was also mentioned in the other issue this logic is not guaranteed: kubernetes/kubernetes#121687 (comment). Hence we should also evaluate if we actually want this pattern.

FYI @dashpole @JeffLuoo

[github-actions]

Pinging code owners for processor/k8sattributes: @dmitryax @fatsheep9146 @TylerHelmuth @ChrsMark. See Adding Labels via Comments if you do not have permissions to add labels yourself. For example, comment '/label priority:p2 -needs-triage' to set the priority and remove the needs-triage label.

[basti1302]

As it was also #23067 (comment) in the other issue this logic is not guaranteed: kubernetes/kubernetes#121687 (comment). Hence we should also evaluate if we actually want this pattern.

I think right now having this pattern is absolutely crucial for some scenarios, see #44407 (comment)

[dipeti]

Hello - this works as expected provided the replicaset's name matches the following regex:

// Extract Deployment name from the ReplicaSet name. Deployment name is created using
// format: [deployment-name]-[hash]
var deploymentHashSuffixPattern = regexp.MustCompile(`^[a-z0-9]{10}$`)

The problem I am facing is ReplicaSets do not always get created with a 10 character hash, sometimes they come with fewer characters.
See the issue covering this in detail. kubernetes/kubernetes#121687
When a ReplicaSet's generated hash is fewer than 10 characters, the k8s.deployment.name metadata is no longer extracted.

[dipeti]

The pod-template-hash podLabel is an automatically generated label used by Kubernetes controllers like Deployments and ReplicaSets to uniquely identify pods belonging to a specific revision of a workload. Wouldn't relying on the value of this label rather than on regex be more robust?

apiVersion: v1
kind: Pod
metadata:
  labels:
    app.kubernetes.io/component: opentelemetry-collector
    app.kubernetes.io/instance: otel-system.otel-deployment
    app.kubernetes.io/managed-by: opentelemetry-operator
    app.kubernetes.io/name: otel-deployment-collector
    app.kubernetes.io/part-of: opentelemetry
    app.kubernetes.io/version: 0.141.0
    pod-template-hash: 67d576cd9c
  name: otel-deployment-collector-67d576cd9c-qvnfj
  namespace: otel-system
  ownerReferences:
  - apiVersion: apps/v1
    blockOwnerDeletion: true
    controller: true
    kind: ReplicaSet
    name: otel-deployment-collector-67d576cd9c
    uid: df58c534-adb0-44fc-bbb1-7a5edece925c

[dipeti]

FYI, ended up with this workaround until the k8sattributesprocessor's logic is amended.

processors:
  k8sattributes:
    extract:
      deployment_name_from_replicaset: true
      labels:
        - from: pod
          key: pod-template-hash
  transform/infer_deployment_name_from_pod_template_hash:
    error_mode: ignore
    metric_statements:
      - set(resource.attributes["k8s.deployment.name"], resource.attributes["k8s.pod.name"]) where resource.attributes["k8s.deployment.name"] == nil and resource.attributes["k8s.pod.labels.pod-template-hash"] != nil
      - replace_pattern(resource.attributes["k8s.deployment.name"], Format("-%s-[a-z0-9]+$", [resource.attributes["k8s.pod.labels.pod-template-hash"]]) , "") where resource.attributes["k8s.pod.labels.pod-template-hash"] != nil
      - delete_key(resource.attributes, "k8s.pod.labels.pod-template-hash")
service:
  pipelines:
    metrics:
      processors:
        - k8sattributes
        - transform/infer_deployment_name_from_pod_template_hash
# Rest is omitted for brevity...

[JeffLuoo]

I took a look of how the name of replicaset is constructed - https://github.com/kubernetes/kubernetes/blob/bf0e7b694f3a7355c200241825b6688ad391fae8/pkg/controller/deployment/sync.go#L198C2-L198C21

It's using the podTemplateSpecHash to construct the name. Get back to maintainer of the processor - @ChrsMark WDYT of using the pod template hash to trim the name and get the deployment name?