Expose Pipeline and Namespace-Level Metrics in Tekton Pipelines

[lluisrubio]

Feature Request: Expose Pipeline and Namespace-Level Metrics in Tekton Pipelines

Problem statement

Tekton Pipelines currently exposes only controller-level metrics through the controller-service (port 9090). These metrics:

• Are global, not tied to any specific PipelineRun, TaskRun, Pipeline, or namespace.
• Do not include labels such as pipeline, task, pipelineRun, or namespace.
• Cannot be used to build per-tenant or per-pipeline dashboards, alerts, or SLA indicators.

As a result, users operating multi‑tenant clusters or large-scale CI/CD platforms cannot:

• Measure performance per pipeline.
• Monitor reliability per namespace.
• Produce accurate per-tenant observability.
• Track execution durations, success/failure rates, or trends per pipeline.

---

Why this feature is needed

1. Tekton only exposes controller metrics today

The official documentation shows that only controller-level Prometheus metrics are available and they are not pipeline-aware nor namespace-aware.

2. Modern production systems require per-pipeline and per-namespace metrics

Examples:

• Error rates per pipeline
• Duration per pipeline or task
• Tenant‑level SLAs (by namespace)
• Cost allocation per CI pipeline
• Alerts tied to specific pipelines

3. Large multi‑tenant clusters cannot rely on global aggregation

Without pipeline/ns labels, metrics are impossible to group meaningfully.

4. Tekton Results is not a metrics system

While Tekton Results allows storing and querying PipelineRuns, it does not expose Prometheus metrics, meaning a custom pipeline metrics layer is still needed.

5. The CNCF ecosystem expects detailed metrics

Prometheus, Grafana, Argo, and Kubernetes controllers typically expose granular metrics with labels.

---

Proposed feature

Expose metrics for the following entities:

• PipelineRuns
• Pipelines
• TaskRuns
• Tasks
• Namespaces

With labels such as:

namespace="team-a"
pipeline="build-app"
pipelineRun="build-app-9x2kz"
task="compile"
status="success|failed|running"

Proposed metrics include:

Pipeline-level

tekton_pipeline_run_duration_seconds{namespace="team-a", pipeline="build"} 12.4
tekton_pipeline_run_count{namespace="team-a", pipeline="deploy", status="success"} 42
tekton_pipeline_run_count{namespace="team-a", pipeline="deploy", status="failed"} 3

Task-level

tekton_task_run_duration_seconds{namespace="team-a", pipeline="build", task="compile"} 4.8

Namespace-level

tekton_pipelinerun_total{namespace="team-a"} 1034
tekton_taskrun_total{namespace="team-a"} 5821

Implementation ideas:

• Add new metrics directly in the Tekton controller.
• Introduce a lightweight Tekton Metrics sidecar.
• Add pluggable metric collectors.
• Use Tekton Results as a backend for metrics aggregation.

---

Summary

Adding pipeline-level and namespace-level metrics to Tekton Pipelines would:

• Dramatically improve observability.
• Enable multi‑tenant and enterprise use cases.
• Reduce operational overhead for large-scale users.
• Bring Tekton in line with other CNCF project observability models.

[AlanGreene]

Tekton Pipelines does expose a number of metrics with labels including pipeline, pipelinerun, task, taskrun, namespace, status, reason, etc., as documented in https://tekton.dev/docs/pipelines/metrics/

You can also find some relevant design decisions and related resources in https://github.com/tektoncd/community/blob/main/teps/0073-simplify-metrics.md