- Remove unused sinon import
- Remove unused DEFAULT_INTERVAL constant
- Use ternary in router per lint rule
- Break long lines under 120 char limit
- Add trailing commas per style

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Adds the mapped metric (nodejs.eventloop.utilization) and higher
percentiles. Fixes lint errors. Total: 13 OTel metrics.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add assertions for nodejs.eventloop.delay.p90, p99, and
nodejs.eventloop.utilization. Verify disabled state for
new event loop metrics. Matches existing runtime_metrics.spec.js
pattern of verifying every metric by name.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The rest of the tracer (agent writer) calls increment(), gauge(), etc.
on the runtime metrics module. When OTLP replaces DogStatsD, these
methods must exist as noops to avoid TypeError crashes.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Assert exactly 13 instruments
- Assert no DD-proprietary names (runtime.node.*)
- Matches .NET PR #8457 test pattern

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

dd-trace-js custom MeterProvider does not implement
addBatchObservableCallback. Switch to per-instrument addCallback
which is the pattern used by all existing OTel metrics in the repo.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Mirrors the .NET PR's SystemRuntimeMetricsFlowThroughOtlpPipeline test:
builds a real MeterProvider + PeriodicMetricReader, registers our 13
runtime instruments via otlp_runtime_metrics.start(), forces a flush,
and asserts the captured export contains all expected metric names,
the correct scope name, and positive observed values.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Audited each instrument against the upstream Node-specific OTel semantic
convention YAMLs (model/v8js, model/nodejs, model/process) and the
@opentelemetry/instrumentation-runtime-node community implementation.
Findings + fixes:

- 5 memory metrics had wrong instrument type. v8js.memory.heap.used,
  v8js.memory.heap.limit, v8js.memory.heap.space.available_size,
  v8js.memory.heap.space.physical_size, and process.memory.usage are
  updowncounters per spec (state metrics that can go up or down).
  Switched from createObservableGauge to createObservableUpDownCounter.
  This matches the spec; OTel contrib still uses gauge here, mirroring
  the same drift the .NET PR addressed.

- nodejs.eventloop.delay.* used unit 'ns' and multiplied by 1e6.
  perf_hooks.monitorEventLoopDelay returns nanoseconds, so the old
  conversion inflated values by 1e6. Switched to unit 's' and divide
  by 1e9, matching both the spec and OTel contrib's
  EventLoopDelayCollector.

- nodejs.eventloop.utilization reported the cumulative-since-startup
  ratio, which converges to an average rather than reflecting recent
  utilization. Now computes a delta vs the last sample, mirroring
  OTel contrib's EventLoopUtilizationCollector.

Tests now cover instrument types per metric, the unit fixes, and the
sum-vs-gauge wire shape after the OtlpTransformer. End-to-end flow
test is unchanged in shape but now passes a fourth assertion that
exercises the OTLP serialization path.

The Node.js OTel semantic conventions list 7 event loop delay metrics
(min, max, mean, stddev, p50, p90, p99). We were emitting 6, missing
stddev. The histogram already exposes the value, so this is a single
ObservableGauge in seconds, matching the unit and instrument type the
spec defines and the @opentelemetry/instrumentation-runtime-node
contrib package uses.

Bumps unit-test and pipeline-flow expected counts to 14.

Co-authored-by: Cursor <cursoragent@cursor.com>

…NERS

The previous layout put new test files under test/runtime_metrics/, a new
subdirectory that fell outside both the test:trace:core glob and any
CODEOWNERS rule. Existing convention is flat: src/runtime_metrics/foo.js
is tested by test/foo.spec.js, mirroring how runtime_metrics.spec.js
sits next to runtime_metrics.js's test root, not in a subdir.

- Move both spec files up one level (path adjusts in proxyquire/require).
- CODEOWNERS gets two specific entries paralleling the existing
  runtime_metrics.spec.js line, owned by apm-sdk-capabilities-js.
- Picks up packages/dd-trace/test/*.spec.js glob in test:trace:core
  automatically, so verify-exercised-tests stops flagging.

Replaces the per-file entry for runtime_metrics.spec.js (and would-be
extra lines for otlp_runtime_metrics.spec.js and .flow.spec.js) with a
single glob /packages/dd-trace/test/*runtime_metrics*.spec.js owned by
apm-sdk-capabilities-js. Net: -2 lines, same coverage.

…ec.js

Three things at once, all keyed off the existing repo trend rather than
adding new infrastructure:

1. Move the OTLP unit + flow tests into runtime_metrics.spec.js (the
   existing test file for the runtime_metrics module) instead of the
   separate test/runtime_metrics/*.spec.js files I had under a new
   subdirectory. Drops the new directory, no CODEOWNERS / package.json
   glob update needed — the existing entry covers everything.

2. Trim from 13 OTLP tests to 4 with the same coverage:
   - registers all 14 OTel-native metrics with spec-correct types,
     units, and callbacks (locks in the spec partition)
   - observes positive values + emits required attributes
     (v8js.heap.space.name, process.cpu.state)
   - skips event loop metrics when disabled and is restartable
   - flushes all 14 metrics with correct wire shape and scope name
     end-to-end through MeterProvider + PeriodicMetricReader +
     OtlpTransformer

   Each one consolidates a previously per-behavior assertion. Same
   spec coverage, ~210 fewer lines.

3. Add JSDoc to start(config) / stop() in otlp_runtime_metrics.js per
   CLAUDE.md.

Source directory was never owned (legacy runtime_metrics.js included).
Mergegate flagged the new otlp_runtime_metrics.js as unowned. One
directory entry covers both the legacy file and the new OTLP module,
parallels the existing test-side ownership rule, owned by the same
team that owns the corresponding test file.

Master invokes runtimeMetrics.start before initializeOpenTelemetryMetrics,
so when our OTLP runtime metrics module calls metrics.getMeterProvider(),
it gets the noop provider — instruments register against it and never
export. Move the OTel logs/metrics init blocks to fire before
runtimeMetrics.start.

track, boolean, histogram, count, gauge each had zero production
callers across the whole codebase. They were mirroring the legacy
runtime_metrics.js interface for safety — but per the no-fallbacks-for-
scenarios-that-cant-happen rule, leaving them in is just dead weight.
Only increment and decrement remain (17 + 6 real callers in
opentracing/span.js and opentracing/tracer.js).