Metrics Migration: OpenCensus to OpenTelemetry

Breaking Changes - Action Required

Related PR: #9043

Executive Summary

Tekton Pipelines has migrated from OpenCensus (deprecated) to OpenTelemetry (CNCF standard). This is a BREAKING CHANGE requiring updates to dashboards, alerts, and configuration.

What Changed

Infrastructure Metrics (HIGH IMPACT)

• Workqueue: tekton_pipelines_controller_workqueue_* → kn_workqueue_*
• K8s Client: tekton_pipelines_controller_client_* → Standard HTTP/K8s metrics
• Go Runtime: tekton_pipelines_controller_go_* → go_*

Core Tekton Metrics (LOW IMPACT)

• Names preserved: tekton_pipelines_controller_pipelinerun_* and tekton_pipelines_controller_taskrun_*
• Labels unchanged by default
• Optional reason label available (disabled by default, see Section 2.2)

Configuration (MEDIUM IMPACT)

• metrics.backend-destination → metrics-protocol
• New OTLP export options (gRPC, HTTP/protobuf)

Removed Metrics (LOW IMPACT)

• reconcile_count and reconcile_latency → Use kn_workqueue_* instead

Action Required

1. Update ConfigMap: metrics.backend-destination: prometheus → metrics-protocol: prometheus
2. Update Dashboards: Replace infrastructure metric names (see tables below)
3. Update Alerts: Use new metric names
4. Test First: Validate in staging before production

1. Infrastructure Metrics - Breaking Changes

1.1 Workqueue Metrics

1.2 Kubernetes Client Metrics

1.3 Go Runtime Metrics

All Go runtime metrics renamed: tekton_pipelines_controller_go_* → go_*

Examples:

• tekton_pipelines_controller_go_goroutines → go_goroutines
• tekton_pipelines_controller_go_memstats_alloc_bytes → go_memstats_alloc_bytes
• tekton_pipelines_controller_go_threads → go_threads

2. Core Tekton Metrics - No Breaking Changes

Core metrics retain their names and labels. By default, there are no changes to PipelineRun and TaskRun metrics.

2.1 Backward Compatibility

These metrics maintain full backward compatibility with the OpenCensus implementation when using default configuration.

2.2 Optional Feature: reason Label (Disabled by Default)

An opt-in feature allows adding a reason label to duration metrics for more granular failure analysis.

To enable (not recommended for high-volume clusters):

data:
  metrics.count.enable-reason: "true"

When enabled, duration metrics gain the reason label:

Reason values: Succeeded, Failed, Completed, Cancelled, PipelineRunCancelled, TimedOut, StoppedRunFinally, CancelledRunFinally

⚠️ Cardinality Impact: Enabling this increases time series by 3-5x. Use sum by(le) aggregation in queries to mitigate.

Note: Total counters (*_total) never include the reason label, even when enabled.

2.3 Preserved Behavior

3. Removed Metrics

4. Configuration Changes

4.1 Before (OpenCensus) - REMOVED

apiVersion: v1
kind: ConfigMap
metadata:
  name: config-observability
  namespace: tekton-pipelines
data:
  metrics.backend-destination: prometheus  # DEPRECATED - no longer supported

4.2 After (OpenTelemetry) - Current

apiVersion: v1
kind: ConfigMap
metadata:
  name: config-observability
  namespace: tekton-pipelines
data:
  # ===== Metrics Configuration =====
  metrics-protocol: prometheus                    # Required: prometheus, grpc, http/protobuf, none
  metrics-endpoint: ""                            # Optional: OTLP endpoint (for grpc/http)
  metrics-export-interval: "30s"                  # Optional: Export frequency

  # ===== Tracing Configuration (NEW) =====
  tracing-protocol: none                          # Optional: grpc, http/protobuf, none, stdout
  tracing-endpoint: ""                            # Optional: OTLP tracing endpoint
  tracing-sampling-rate: "1.0"                    # Optional: 0.0-1.0 (1.0 = 100%)

  # ===== Runtime Configuration =====
  runtime-profiling: disabled                     # Optional: enabled, disabled
  runtime-export-interval: "15s"                  # Optional: Runtime metrics export interval

  # ===== Tekton-Specific Settings =====
  metrics.taskrun.level: "task"                   # task, namespace, cluster
  metrics.taskrun.duration-type: "histogram"      # histogram, gauge
  metrics.pipelinerun.level: "pipeline"           # pipeline, namespace, cluster
  metrics.pipelinerun.duration-type: "histogram"  # histogram, gauge
  metrics.count.enable-reason: "false"            # Add reason label to duration metrics (see Section 2.2)

Key Change: metrics.backend-destination → metrics-protocol

4.3 Supported Protocols

4.4 Configuration Examples

Example 1: Prometheus (Default)

data:
  metrics-protocol: prometheus

Example 2: OTLP gRPC to OpenTelemetry Collector

data:
  metrics-protocol: grpc
  metrics-endpoint: "otel-collector.observability.svc.cluster.local:4317"
  metrics-export-interval: "30s"

Example 3: OTLP HTTP with Tracing

data:
  metrics-protocol: http/protobuf
  metrics-endpoint: "http://otel-collector.observability.svc.cluster.local:4318/v1/metrics"

  tracing-protocol: grpc
  tracing-endpoint: "otel-collector.observability.svc.cluster.local:4317"
  tracing-sampling-rate: "0.1"  # 10% sampling

5. Migration Steps

5.1 Pre-Migration Checklist

• Inventory all dashboards using Tekton metrics
• Inventory all alerts using Tekton metrics
• Back up config-observability ConfigMap
• Plan maintenance window

5.2 Step-by-Step Process

Step 1: Update Configuration

kubectl edit configmap config-observability -n tekton-pipelines
# Change: metrics.backend-destination: prometheus
# To:     metrics-protocol: prometheus

Step 2: Upgrade Tekton Pipelines

# Apply the new version containing the OTel migration
kubectl apply -f https://storage.googleapis.com/tekton-releases/pipeline/latest/release.yaml

# Wait for rollout
kubectl rollout status deployment/tekton-pipelines-controller -n tekton-pipelines

Step 3: Verify Metrics Endpoint

# Port-forward to controller
kubectl port-forward -n tekton-pipelines deployment/tekton-pipelines-controller 9090:9090

# Check metrics (in another terminal)
curl http://localhost:9090/metrics | grep -E "(kn_workqueue|tekton_pipelines_controller)"

Step 4: Update Dashboards

Note: Core Tekton metrics (PipelineRun/TaskRun) queries remain unchanged with default configuration.

Step 5: Update Alerts

Before:

- alert: HighWorkqueueDepth
  expr: tekton_pipelines_controller_workqueue_depth > 100

After:

- alert: HighWorkqueueDepth
  expr: kn_workqueue_depth > 100

Step 6: Verify and Monitor

• Confirm metrics appear in Prometheus
• Check dashboards display correctly
• Verify alerts work
• Monitor for errors in controller logs

6. Query Examples

Example 1: PipelineRun Success Rate

Before:

sum(rate(tekton_pipelines_controller_pipelinerun_total{status="success"}[5m]))
/
sum(rate(tekton_pipelines_controller_pipelinerun_total[5m]))

After: ✅ No change needed (reason label not added to *_total)

sum(rate(tekton_pipelines_controller_pipelinerun_total{status="success"}[5m]))
/
sum(rate(tekton_pipelines_controller_pipelinerun_total[5m]))

Example 2: P95 PipelineRun Duration

Before:

histogram_quantile(0.95,
  rate(tekton_pipelines_controller_pipelinerun_duration_seconds_bucket[5m])
)

After: ✅ No change needed (with default configuration)

histogram_quantile(0.95,
  rate(tekton_pipelines_controller_pipelinerun_duration_seconds_bucket[5m])
)

Note: If you enable metrics.count.enable-reason: "true", you’ll need to aggregate by le:

histogram_quantile(0.95,
  sum by(le) (
    rate(tekton_pipelines_controller_pipelinerun_duration_seconds_bucket[5m])
  )
)

# OR keep reason for granular failure analysis:
histogram_quantile(0.95,
  sum by(le, reason) (
    rate(tekton_pipelines_controller_pipelinerun_duration_seconds_bucket[5m])
  )
)

Example 3: Controller Memory

Before:

tekton_pipelines_controller_go_memstats_alloc_bytes

After: ✅ Simple prefix change

go_memstats_alloc_bytes

Example 4: Workqueue Processing Rate

Before:

rate(tekton_pipelines_controller_workqueue_adds_total{name="pipelinerun"}[5m])

After: ✅ Prefix change only

rate(kn_workqueue_adds_total{name="pipelinerun"}[5m])

7. Troubleshooting

7.1 Verifying Metrics Export

# Check metrics endpoint
kubectl port-forward -n tekton-pipelines deployment/tekton-pipelines-controller 9090:9090

# Verify new metrics exist
curl http://localhost:9090/metrics | grep kn_workqueue
curl http://localhost:9090/metrics | grep tekton_pipelines_controller

7.2 Prometheus Not Scraping

Check Prometheus targets:

1. Navigate to Prometheus UI → Status → Targets
2. Look for tekton-pipelines-controller
3. Verify State is UP

Verify ServiceMonitor:

kubectl get servicemonitor -n tekton-pipelines -o yaml

7.3 High Cardinality Warnings

Problem: Prometheus warns about high series cardinality after enabling metrics.count.enable-reason

Root Cause: The reason label increases time series by 3-5x

Solution 1: Disable the feature (recommended):

data:
  metrics.count.enable-reason: "false"

Solution 2: Aggregate away the reason label in queries:

# Instead of:
histogram_quantile(0.95, rate(tekton_pipelines_controller_pipelinerun_duration_seconds_bucket[5m]))

# Use:
histogram_quantile(0.95, sum by(le) (rate(tekton_pipelines_controller_pipelinerun_duration_seconds_bucket[5m])))

Note: This issue only occurs if you explicitly enabled metrics.count.enable-reason: "true".

7.4 Configuration Not Taking Effect

# Restart controller to pick up ConfigMap changes
kubectl rollout restart deployment/tekton-pipelines-controller -n tekton-pipelines
kubectl rollout status deployment/tekton-pipelines-controller -n tekton-pipelines

# Check logs
kubectl logs -n tekton-pipelines deployment/tekton-pipelines-controller | grep -i "observability"

7.5 OTLP Export Failures

Check controller logs:

kubectl logs -n tekton-pipelines deployment/tekton-pipelines-controller | grep -i "otel\|export\|metric"

Test endpoint connectivity:

kubectl exec -n tekton-pipelines deployment/tekton-pipelines-controller -- \
  nc -zv otel-collector.observability.svc.cluster.local 4317

7.6 Enable Debug Logging

kubectl edit configmap config-logging -n tekton-pipelines
# Add: loglevel.controller: "debug"

kubectl rollout restart deployment/tekton-pipelines-controller -n tekton-pipelines

8. Frequently Asked Questions

Q: Do I need to upgrade immediately?

A: This migration is included starting from a future Tekton Pipelines release. Plan to migrate when upgrading to that version. Test in staging first.

Q: Will old metrics continue to work during transition?

A: No. This is a hard cutover. Only OpenTelemetry metrics will be available after upgrade.

Q: What if I don’t update my configuration?

A: If using Prometheus, the default behavior is preserved, but dashboards/alerts will break due to metric name changes.

Q: Can I use both OpenCensus and OpenTelemetry?

A: No. The controller only emits OpenTelemetry metrics after upgrade.

Q: How do I test without affecting production?

A: Deploy in a test/staging environment first. Verify all metrics, dashboards, and alerts work before upgrading production.

Q: Where can I get help?

A: File an issue at https://github.com/tektoncd/pipeline/issues or ask in Slack (#tekton channel).

9. Quick Reference Checklist

Infrastructure Metrics Update

Core Metrics Update

• ✅ No changes needed - Core metrics are backward compatible
• If you enable metrics.count.enable-reason, add sum by(le) aggregation to duration queries

Configuration Update

• Change metrics.backend-destination → metrics-protocol
• Add OTLP endpoint if using gRPC/HTTP protocols
• Configure tracing if desired (new capability)

Dashboard Update

• Replace workqueue metric names
• Replace Go runtime metric names
• Replace K8s client metric names
• No changes needed for core Tekton metrics (PipelineRun/TaskRun)
• Test all panels

Alert Update

• Update workqueue metrics in alert rules
• Update Go runtime metrics in alert rules
• Update K8s client metrics if used
• Verify thresholds still appropriate
• Test alerts fire correctly

Additional Resources

• Tekton PR: https://github.com/tektoncd/pipeline/pull/9043
• OpenTelemetry Documentation: https://opentelemetry.io/docs/
• Tekton Metrics Documentation: https://tekton.dev/docs/pipelines/metrics/
• Tekton Slack: https://kubernetes.slack.com - #tekton channel
• GitHub Issues: https://github.com/tektoncd/pipeline/issues

For questions or clarifications, please refer to PR #9043 or contact the Tekton team.