OTTL profile context exposes profile.sample_type but provides no way to read its data

[henrikrexed]

Component(s)

pkg/ottl

What happened?

Description

Version: v0.137.0 with service.profilesSupport feature gate

The OTTL profile context exposes profile.sample_type but the object is completely opaque - no form of access works.
All attempts fail:

# Returns Empty()
set(resource.attributes["type"], profile.sample_type)

# Error: "path section 'typ' was not used by the context"
set(resource.attributes["type"], profile.sample_type.typ)

# Error: "path section 'type' was not used by the context"  
set(resource.attributes["type"], profile.sample_type.type)

# Error: "keys indexing 'sample_type' were not used by the context"
set(resource.attributes["type"], profile.sample_type["typ"])

Steps to Reproduce

Deploy the profiler collector.

Architecture

Two-tier OpenTelemetry Collector setup:

• DaemonSet collectors - receive profiles from eBPF profiler, forward to gateway
• Gateway collector - receives profiles from DaemonSets, attempts to process with OTTL
  Minimal ConfigurationGateway Collector (otelcol-gateway.yaml):

receivers:
  otlp:
    protocols:
      grpc:
        endpoint: 0.0.0.0:4317

processors:
  transform/test:
    error_mode: ignore
    profile_statements:
      - context: profile
        statements:
          # All of these fail to access sample_type data
          - set(resource.attributes["test1"], profile.sample_type)
          - set(resource.attributes["test2"], profile.sample_type.typ)
          - set(resource.attributes["test3"], profile.sample_type["typ"])

exporters:
  debug:
    verbosity: detailed

service:
  pipelines:
    profiles:
      receivers: [otlp]
      processors: [transform/test]
      exporters: [debug]

Expected Result

sample_type is a struct with fields Typ, Unit, and AggregationTemporality.
OTTL should expose these fields so users can differentiate CPU vs memory profiles.

For now Cannot use signaltometrics or transform processor to create separate metrics for CPU and memory profiles. The connector is unusable for profiling use cases that need type-based filtering.

Please expose sample_type.typ and sample_type.unit as readable fields in the OTTL profile context.

Collector version

v0.137.0

Environment information

Environment

Kubernetes v1.33.4

OpenTelemetry Collector configuration

receivers:

      otlp:
        protocols:
          grpc: {}
          http: {}

    processors:
      batch:
        send_batch_max_size: 1000
        timeout: 30s
        send_batch_size : 800

      memory_limiter:
        check_interval: 1s
        limit_percentage: 70
        spike_limit_percentage: 30


      k8sattributes/profile:
        auth_type: "serviceAccount"
        passthrough: false
        wait_for_metadata: true
        wait_for_metadata_timeout: 60s
        extract:
          metadata:
            - k8s.pod.name
            - k8s.pod.uid
            - k8s.deployment.name
            - k8s.namespace.name
            - k8s.statefulset.name
            - k8s.daemonset.name
            - k8s.cronjob.name
            - k8s.node.name
            - k8s.cluster.uid
          labels:
            - tag_name: app.label.component
              key: app.kubernetes.io/component
              from: pod
          otel_annotations: true
        pod_association:
          - sources:
              - from: resource_attribute
                name: container.id
              - from: resource_attribute
                name: k8s.node.name





      transform/fix_container_id:
        error_mode: ignore
        profile_statements:
          - context: resource
            statements:
              # Add containerd:// prefix if container.id exists and doesn't have it
              - set(attributes["container.id"], Concat(["containerd://", attributes["container.id"]], "")) where attributes["container.id"] != nil and not IsMatch(attributes["container.id"], "^(containerd|docker|cri-o)://")
              - set(attributes["k8s.node.name"], attributes["host.name"]) where attributes["host.name"] != nil
          - context: profile
            statements:
              # Extract the sample type (cpu, alloc_space, etc.)
              # Try accessing the TypeStrindex method
              - set(resource.attributes["profile.type_strindex"], profile.sample_type["typ"]) where profile.sample_type["typ"] != nil

              # Or try with camelCase
#              - set(resource.attributes["profile.type_strindex"], Int(profile.sample_type.TypeStrindex)) where profile.sample_type.TypeStrindex != nil




      
    connectors:
      signaltometrics:

        profiles:
          - name: profile.cpu.count
            description: "Number of CPU profile records per container"
            sum:
              value: "1"  # Just count each profile
            attributes:
              - key: container.short_id
              - key: k8s.node.name
            conditions:
              - 'resource.attributes["profile.type_strindex"] == 1'
          - name: profile.memory.inuse.count
            description: "Number of CPU profile records per container"
            sum:
              value: "1"  # Just count each profile
            attributes:
              - key: container.short_id
              - key: k8s.node.name
            conditions:
              - 'resource.attributes["profile.type_strindex"] == 3'
          - name: profile.memory.alloc.count
            description: "Number of CPU profile records per container"
            sum:
              value: "1"  # Just count each profile
            attributes:
              - key: container.short_id
              - key: k8s.node.name
            conditions:
              - 'resource.attributes["profile.type_strindex"] == 5'
    exporters:


      otlphttp:
        endpoint: ${DT_ENDPOINT}/api/v2/otlp
        headers:
          Authorization: "Api-Token ${DT_API_TOKEN}"
      debug:
        verbosity: detailed

    service:
      pipelines:

       
        profiles:
          receivers: [ otlp ]
          processors: [ memory_limiter, transform/fix_container_id, k8sattributes/profile ]
          exporters: [ signaltometrics,debug ]
        metrics/profile:
          receivers: [ signaltometrics ]
          processors: [ memory_limiter,filter,cumulativetodelta,batch ]
          exporters: [otlphttp,debug]
      telemetry:
        logs:
          level: info
          initial_fields:
            service: otelcol

Log output

Error: invalid configuration: processors::transform/fix_container_id: unable to parse OTTL statement "set(resource.attributes[\"profile.type_strindex\"], profile.sample_type[\"typ\"]) where profile.sample_type[\"typ\"] != nil": error while parsing arguments for call to "set": invalid argument at position 1: the keys indexing "sample_type" were not used by the context - this likely means you are trying to index a path that does not support indexing

Additional context

No response

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}

[github-actions]

Pinging code owners:

• pkg/ottl: @TylerHelmuth @evan-bradley @edmocosta

See Adding Labels via Comments if you do not have permissions to add labels yourself.

[edmocosta]

Thanks for opening this issue @henrikrexed!

OTTL cannot access complex types that are not described here, without exposing new paths for that.
The profile’s path profile.sample_type currently exposes the pprofile.ValueType type, which isn't support out of the box by OTTL, and needs those extra context's paths for accessing this object type fields.

In terms of implementation, I think we can probably came up with a few extra paths, abstracting the profile format/table lookups:

• profile.sample_type.type: string value from the table.
• profile.sample_type.unit: string value from the table.
• profile.sample_type.aggregation_temporality: int64 (OTTL type), we also need to expose the enums for this path values, defined on the profiles proto.

Would that work for you? @florianl WDYT?

[florianl]

Sounds like another subissue/subtask of #40489.

profile.sample_type.aggregation_temporality no longer exists. Other than that, the approach sounds good to me.

[henrikrexed]

I have checked several fields fro the profile schema... :
-> has.profile_id: Str(yes)
-> has.duration: Str(yes)
-> has.sample_type: Str(yes)
-> has.period_type: Str(yes)
-> has.period: Str(yes)
-> has.attribute_indices: Str(yes)
-> has.original_payload_format: Str(yes)
From those fields I can only get value on profile_id and period. all the other ones are returning nil or empty.
I don't know if it is a bug but for now it seems difficult to use transform, filter or signaltometrics on profile

[edmocosta]

I have checked several fields fro the profile schema... : -> has.profile_id: Str(yes) -> has.duration: Str(yes) -> has.sample_type: Str(yes) -> has.period_type: Str(yes) -> has.period: Str(yes) -> has.attribute_indices: Str(yes) -> has.original_payload_format: Str(yes) From those fields I can only get value on profile_id and period. all the other ones are returning nil or empty. I don't know if it is a bug but for now it seems difficult to use transform, filter or signaltometrics on profile

If the data is present on the profile payload, those paths shouldn't return nil or empty values. If they're doing so, it's probably a bug. Do you have any example of those path issues? I've tested a few paths and they seems to be working, except for the period_type, which needs extra paths as well, similar to the sample_type.
Please note that a few data types are not supported by the attributes, which includes the time.Time type. If you're testing it by copying that value to an attribute, it probably won't do anything, and you need to convert those values for int timestamps or strings before setting it. Thanks!

[edmocosta]

For now, let's add support for the missingsample_type and period_type fields (.type, .unit), as they're essential to work with profiles on the transform and filter processors. I'll try to put a PR together soon. Thanks again for opening the issue.