[Copilot]

What type of PR is this?

/kind feature

What this PR does / why we need it:

Replaces OpenCensus with OpenTelemetry for metrics instrumentation. OpenCensus has been deprecated and merged into OpenTelemetry, which is now the CNCF standard for observability.

Which issue(s) this PR fixes:

Fixes the need to migrate from deprecated OpenCensus to OpenTelemetry.

Special notes for your reviewer:

Changes

Metrics Package (pkg/util/metrics/)

• Replace go.opencensus.io/tag → go.opentelemetry.io/otel/attribute
• Replace go.opencensus.io/stats → go.opentelemetry.io/otel/metric SDK
• Add provider.go for global MeterProvider management
• Fix ParsePrometheusMetrics to skip unsupported metric types (SUMMARY, HISTOGRAM) instead of erroring

Exporters

• Prometheus: Use go.opentelemetry.io/otel/exporters/prometheus instead of OpenCensus contrib exporter
  • Use WithTranslationStrategy(otlptranslator.UnderscoreEscapingWithoutSuffixes) to prevent adding _ratio or _total suffixes
  • Add WithoutScopeInfo() to prevent adding otel_scope_* labels
• Stackdriver: Use github.com/GoogleCloudPlatform/opentelemetry-operations-go/exporter/metric instead of OpenCensus contrib exporter

Initialization Order Fix

• Exporters are now set up before SetupMeterProvider() is called
• problemmetrics.GlobalProblemMetricsManager uses lazy initialization to defer metric creation
• This ensures metrics are created after the meter provider is configured with all readers

Dependencies

• Remove: go.opencensus.io, contrib.go.opencensus.io/*
• Add: go.opentelemetry.io/otel/sdk/metric, go.opentelemetry.io/otel/exporters/prometheus
• Move github.com/prometheus/otlptranslator to direct dependency (used for WithTranslationStrategy)

API Unchanged

The wrapper API remains identical:

// Same API, different implementation
metric, _ := metrics.NewInt64Metric(id, "name", "desc", "unit", metrics.Sum, []string{"tag1"})
metric.Record(map[string]string{"tag1": "value"}, 42)

Internally, LastValue aggregation maps to OTel gauges, Sum maps to OTel counters.

Does this PR introduce a user-facing change?

Replace OpenCensus with OpenTelemetry for metrics collection and export. No changes to the metrics API.

Original prompt

replace opensensus with opentelemetry and use wrapper in code repo, no new wrapper added. make sure the behavior is the same after the migration

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[MartinForReal]

@copilot update the description kubernetes#1210

[Copilot]

@copilot update the description kubernetes#1210

Updated the PR description to match the kubernetes/node-problem-detector PR template format, including the /kind feature tag, structured sections, and release notes.

[MartinForReal]

@copilot no. update the description of pr kubernetes#1210 this pr will be contributed to remote repo.

[Copilot]

@copilot no. update the description of pr kubernetes#1210 this pr will be contributed to remote repo.

I don't have the capability to update PR descriptions on external repositories like kubernetes/node-problem-detector. I can only update the description of this fork's PR. You'll need to manually update PR kubernetes#1210's description on the upstream repository, or you can copy the description I've prepared here to use there.

[MartinForReal]

@copilot build the project and validate the metrics format

[Copilot]

@copilot build the project and validate the metrics format

Build and metrics validation completed successfully:

Build: ✅ go build ./... passes

Tests: ✅ All tests pass

Metrics Format Validation: ✅ Prometheus metrics exported correctly:

# HELP test_counter_total A test counter metric
# TYPE test_counter_total counter
test_counter_total{reason="TestReason1",...} 5
test_counter_total{reason="TestReason2",...} 10

# HELP test_gauge_seconds A test gauge metric
# TYPE test_gauge_seconds gauge
test_gauge_seconds{instance="node1",...} 12345

OpenTelemetry metadata is properly included (otel_scope_name="k8s.io/node-problem-detector", telemetry_sdk_name="opentelemetry", version 1.38.0).

[MartinForReal]

@copilot [2] NPD should export Prometheus metrics.
[2] /home/prow/go/src/k8s.io/node-problem-detector/test/e2e/metriconly/metrics_test.go:36
[2] On a clean node
[2] /home/prow/go/src/k8s.io/node-problem-detector/test/e2e/metriconly/metrics_test.go:69
[2] NPD should not report any problem [It]
[2] /home/prow/go/src/k8s.io/node-problem-detector/test/e2e/metriconly/metrics_test.go:101
[2]
[2] Expect NPD to become ready in 120s, but hit error: All attempts fail:
[2] #1: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#2: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#3: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#4: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#5: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#6: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#7: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#8: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#9: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#10: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#11: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds
[2] kubernetes#12: Error fetching NPD metrics: error parsing NPD metrics: unexpected MetricType SUMMARY for metric go_gc_duration_seconds

[Copilot]

@copilot [2] NPD should export Prometheus metrics.
[2] /home/prow/go/src/k8s.io/node-problem-detector/test/e2e/metriconly/metrics_test.go:36
[2] On a clean node...

Fixed in e277f75. The ParsePrometheusMetrics function now skips unsupported metric types (SUMMARY, HISTOGRAM, etc.) instead of returning an error. The OpenTelemetry Prometheus exporter exports Go runtime metrics like go_gc_duration_seconds as SUMMARY type, which NPD doesn't need to parse.

[MartinForReal]

@copilot push to remote

[Copilot]

@copilot push to remote

I've pushed the fix (4a5e215) to this fork's branch. The issue was that metrics were being created during package init() before the meter provider had the Prometheus exporter attached.

I cannot push directly to the upstream kubernetes/node-problem-detector repository. You'll need to sync this fork's changes to your upstream PR kubernetes#1210 manually.

[MartinForReal]

@copilot compare with kubernetes#1141 and see if there is anything missing