feat(pivot): PR-B1 — port nccl_fr off internal selftel + lifecycle by trilamsr · Pull Request #184 · TraceCoreAI/tracecore

trilamsr · 2026-05-31T03:43:26Z

Summary

Port components/receivers/nccl_fr off internal/selftelemetry + internal/runtime/lifecycle. Helpers travel as siblings (selftel.go + lifecycle.go) co-located with the receiver, per RFC-0013 §migration PR-B1.
Receiver-scoped meter is acquired from set.Telemetry.MeterProvider. Instrumentation scope name = the receiver's Go import path (github.com/tracecoreai/tracecore/components/receivers/nccl_fr); will follow the receiver to module/receiver/ncclfrreceiver/ in PR-I.1.
Metric names + label shape preserved (tracecore.receiver.{errors_total,emissions_total,collection_latency_seconds,degraded_seconds_total,last_activity_unix_seconds} with component_id + kind), so dashboards / alerts do not regress. The cmd/tracecore integration test that asserts these names by string still passes.
Lifecycle sibling is intentionally slimmer than the internal helper: drops Add() + the post-Shutdown silent-no-op path because nccl_fr is single-source. PanicCallback(any) signature is preserved so the existing panic → IncError(kindPanic) + SetDegraded(true) wiring is untouched.

This is partial credit toward the internal/selftelemetry + internal/runtime/lifecycle deletes scheduled for PR-F. The package deletions wait on stdoutexporter, dcgm, kernelevents, and clockreceiver also porting off (or being deleted by) the wave.

Why receiver-local scope (vs reusing the internal scope name)

The adversarial review flagged the scope-name choice as the key decision. Picking the receiver's import path means:

Transitional fragmentation: tracecore_receiver_* series will appear under two otel_scope_name labels while old receivers still emit under internal/selftelemetry. Bounded — all other in-tree receivers either die (PR-K) or move to module/receiver/* (PR-I.1) within v0.2.0, and no operator alerts exist at v0.1.x that strip the scope label.
Aligns with OTel convention (scope = Go import path), which makes the PR-I.1 move re-stamp the scope cleanly.

Test plan

make check (gofumpt, golangci-lint, go vet, mod verify) — 0 issues.
go test ./... repo-wide — green, including cmd/tracecore integration tests that assert tracecore_receiver_* metric names.
New tests: selftel_test.go (6 tests covering noop, nil-MP, errors_total + kind label, emissions_total, scope-name pin, init_errors_total + the factory-fallback seam), lifecycle_test.go (5 tests covering happy path, idempotent Start/Shutdown, panic recovery, shutdown-deadline).
Adversarial review across 9 lenses (sibling-impl drift, lifecycle drift, test correctness, cardinality, doc-rot, ctx usage, race-safety, helper duplication, scope creep) — no findings.
grep "internal/selftelemetry\|internal/runtime/lifecycle" components/receivers/nccl_fr/*.go — only doc-comment mentions remain (no imports).

Follow-ups

PR-B2: port nccl_fr off internal/pipeline + internal/consumer (blocked on PR-A2).
PR-F: delete internal/{componentstatus,selftelemetry,runtime/lifecycle} once every consumer has ported or been deleted.

Binding doc: docs/rfcs/0013-distro-first-pivot.md §migration PR-B1.

Release notes

[CHANGE] nccl_fr receiver migrates its self-telemetry + lifecycle helpers to co-located sibling files; emitted metric names + labels are unchanged but the OTel instrumentation scope is now the receiver's Go import path.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Signed-off-by: Tri Lam <tri@maydow.com>

## Summary - Port `components/receivers/clockreceiver` off `internal/selftelemetry` + `internal/runtime/lifecycle`. Helpers travel as siblings (`selftel.go` + `lifecycle.go`) co-located with the receiver, per RFC-0013 §migration PR-B (mechanical mirror of PR-B1 #184 for nccl_fr). - Receiver-scoped meter is acquired from `set.Telemetry.MeterProvider`. Instrumentation scope name = the receiver's Go import path (`github.com/tracecoreai/tracecore/components/receivers/clockreceiver`). - Metric names + label shape preserved (`tracecore.receiver.{errors_total,emissions_total,collection_latency_seconds,degraded_seconds_total,last_activity_unix_seconds}` with `component_id` + `kind`). The `cmd/tracecore` integration tests that assert these names by string still pass. - Lifecycle sibling is slimmer than the internal helper: drops `Add()` + the post-Shutdown silent-no-op path because clockreceiver is single-source (same shape as the nccl_fr sibling). `panicCallback(any)` signature preserved so the panic → `IncError(kindPanic)` + `SetDegraded(true)` wiring is untouched. - `kind` enum is trimmed to the two values clockreceiver actually emits (`downstream`, `panic`) — no speculative kinds. ## Why port-before-delete clockreceiver is scheduled for deletion in PR-K, so a fair question is "why port it at all?" Three reasons: 1. **Unblocks PR-F earlier.** The `internal/selftelemetry` + `internal/runtime/lifecycle` deletes in PR-F are blocked on every consumer porting off or being deleted. Porting clockreceiver here means PR-F can land before PR-K, instead of waiting on the kernelevents / stdoutexporter / dcgm porting wave plus the clockreceiver delete. 2. **Mechanical mirror of PR-B1.** This diff is a sibling-rename of PR #184 — the review surface is the trimmed `kind` enum + the package-name swap. No new design choices. 3. **Anchors the pattern.** clockreceiver is the "canonical example receiver" per its package doc; even briefly carrying the wrong shape would signal that the migration is optional. Partial credit toward the `internal/selftelemetry` + `internal/runtime/lifecycle` deletes scheduled for PR-F. ## Test plan - [x] `make check` (gofumpt, golangci-lint, go vet, mod verify) — 0 issues. - [x] `go test ./...` repo-wide — green for the clockreceiver and `cmd/tracecore` packages, which assert `tracecore_receiver_*` metric names. (kernelevents `TestReceiver_SLIBudget` shows a known p99-latency overshoot under full-suite load; passes in isolation with `-count=3`; unrelated to this PR.) - [x] New tests: `selftel_test.go` (6 tests — noop safety, nil-MP, errors_total + kind, emissions_total, scope-name pin, init_errors_total + factory-fallback seam), `lifecycle_test.go` (5 tests — happy path, idempotent Start/Shutdown, panic recovery, shutdown-deadline). - [x] 9-lens adversarial review (sibling-impl drift, lifecycle drift, test correctness, cardinality, doc-rot, ctx usage, race-safety, helper duplication, scope creep) — no findings. Bonus catch: `r.logger` now falls back to `slog.Default()` when `set.Telemetry.Logger` is nil; the original would have nil-dereferenced. - [x] `grep "internal/selftelemetry\|internal/runtime/lifecycle" components/receivers/clockreceiver/*.go` — only doc-comment mentions remain (no imports). ## Follow-ups - PR-F: delete `internal/{componentstatus,selftelemetry,runtime/lifecycle}` once every remaining consumer (stdoutexporter, dcgm, kernelevents, containerstdout) has ported or been deleted. - PR-K: delete clockreceiver itself (replaced by hostmetricsreceiver heartbeat). Binding doc: [docs/rfcs/0013-distro-first-pivot.md](docs/rfcs/0013-distro-first-pivot.md) §migration PR-B. ## Release notes ```release-notes [CHANGE] clockreceiver migrates its self-telemetry + lifecycle helpers to co-located sibling files; emitted metric names + labels are unchanged but the OTel instrumentation scope is now the receiver's Go import path. ``` Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

Two cross-cut-reviewer findings on PR #188 before re-enabling auto-merge: 1. KindParse was speculative — the dcgm cgo client returns typed values, never raw bytes, so no call site ever emitted it. Removed from the canonical Kind* const block in selftel.go, the canonicalKindByName resolver map in docs_parity_test.go, and the noop-safety test's IncError sweep in selftel_test.go. The AST-walking TestCanonicalKindByName_CoversConstBlock guard stays green because both halves move together. 2. Mirror PR #184's nccl_fr register-failure pattern: added a failingDcgmMP test seam (wraps a real MeterProvider but fails every `tracecore.receiver.*` instrument registration) plus TestSelfTelemetry_NewReceiver_RegisterFailureReturnsErr that asserts newSelfTelemetry surfaces a wrapped errSyntheticDcgmFailure rather than returning a partially-wired impl. Pins the criterion-6 symmetry the cross-cut review flagged so a future refactor that reorders the constructor can't silently bypass the register- failure path on this receiver. TestRecordInitError_NilProviderIsSafe was already present from the initial port; no change needed. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Signed-off-by: Tri Lam <tri@maydow.com>

## Summary RFC-0013 PR-B2: ports `components/receivers/kernelevents` off `internal/selftelemetry` + `internal/runtime/lifecycle` to package-local sibling types, following PR-B1 (#184) for `nccl_fr`. One more receiver off the internal helpers; clears another blocker for PR-F (delete `internal/selftelemetry` + `internal/runtime/lifecycle`). ## Why `internal/selftelemetry` and `internal/runtime/lifecycle` are slated for deletion in RFC-0013 PR-F. Every in-tree user must migrate first. PR-B1 landed the pattern (`nccl_fr`); PR-B2 applies it to `kernelevents`. Receivers landing post-PR-F will own their selftel + lifecycle in their own submodule per OTel convention — this PR puts kernelevents on that footing. ## Key difference from PR-B1 PR-B1 slimmed nccl_fr's sibling lifecycle by dropping `Add()` (single source). kernelevents is multi-source (kmsg + journald) and registers one driver goroutine per source under the same WaitGroup via `Add()`, so this PR keeps `Add()` and ports it (with its post-Shutdown and pre-Start refusal guards) from `internal/runtime/lifecycle`. ## What changed - New `components/receivers/kernelevents/selftel.go` — local `selfTelemetry` interface, `kind` type, noop impl, real OTel-backed impl, `recordInitError`. Metric names + label shape preserved (`tracecore.receiver.errors_total{kind,component_id}` + siblings); scope name is the receiver's Go import path (OTel convention). Receiver-local kinds (`kmsg_oversized`, `kmsg_overflow`, `journalctl_crash`) sit alongside the canonical mirror set (`parse`, `downstream`, `panic`, `cardinality`); merges what used to live in `kinds.go` (deleted). - New `components/receivers/kernelevents/lifecycle.go` — local `lifecycle` type with `Start` / `Shutdown` / `Add`. TOCTOU-safe Add()/Shutdown via shared mutex; idempotent Shutdown that stashes the first error so deadline misses aren't swallowed; panic recovery routes through the same `onPanic` callback for both Start'd and Add'd goroutines. - New `selftel_test.go` + `lifecycle_test.go` — TDD-first, no `internal/telemetry` dep, OTel `sdkmetric.NewManualReader` for metric assertions. Covers noop safety, nil-MP error, errors_total + kind + component_id, scope-name pin, emissions monotonicity + negative-discard, init_errors_total, recordInitError nil-safety, Start/Shutdown happy path, Start-twice sentinel, idempotent Shutdown, panic callback fires, shutdown-deadline returns ctx err, Add() registers under same WaitGroup, Add'd panic fires callback, Add-before-Start no-op, Add-after-Shutdown no-op. - Rewires `kernelevents.go`, `source.go`, `kmsg.go`, `journald.go`, `factory.go`, `export_test.go`, `nullsource_test.go`, `source_template.go.example`, README — drop the `internal/*` imports, switch to the local `selfTelemetry` / `lifecycle` / `kind` symbols. `FakeTelemetry.ErrorKinds()` now returns `[]string` (was `[]selftelemetry.Kind`) — only consumer is in-package tests. - `runbook_kinds_test.go` AST gate rewritten to resolve local `kind` const decls (was: selector-walk for `selftelemetry.Kind...`); expected set unchanged so a future regression that drops a kind still fails the gate. - `internal/runtime/lifecycle/lifecycle.go` doc comment updated: kernelevents removed from the "Used by" list; dcgm is the last in-tree user, noted alongside the deletion plan. ## Root cause Not a bug fix — this is a structural migration to unblock PR-F. kernelevents-specific requirement: multi-source receiver needs `Add()` preserved across the migration (PR-B1's pattern dropped it for the single-source case). No workaround; full port. ## Test plan - [x] `make check` — 0 issues (golangci-lint + vet + mod verify) - [x] `go test ./...` — entire repo green - [x] `go test -race -run "^(TestSelfTelemetry|TestLifecycle|TestRecordInit|TestSourceInterface|TestKmsgSource)" ./components/receivers/kernelevents/` — race-clean - [x] `TestRUNBOOK_KindsMatchEmitted_PinsExpectedSet` still pins the same kind set after the AST walker rewrite - [x] Verified `kernelevents.FakeTelemetry` + helpers have no out-of-package consumers ### Known pre-existing flake (NOT caused by this PR) `TestReceiver_SLIBudget` (bench_test.go) flakes when p99 emit latency exceeds 5ms — reproduced on clean `main` at 12.9ms p99 without these changes. Tracking separately; not blocking. ## Release notes ```release-notes [CHANGE] kernelevents: receiver self-telemetry + lifecycle helpers moved from internal/selftelemetry + internal/runtime/lifecycle to package-local siblings (RFC-0013 PR-B2). Metric names, label shape, and observable behavior are unchanged. The OTel instrumentation scope name moves from `github.com/tracecoreai/tracecore/internal/selftelemetry` to `github.com/tracecoreai/tracecore/components/receivers/kernelevents`, matching OTel convention; dashboards that pinned the old scope name need updating. ``` 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

## Summary Port `components/exporters/stdoutexporter` off `internal/selftelemetry`. This is the exporter-side half of RFC-0013 §migration PR-B — sibling to the receiver-side PR-B1 (#184) that ported `components/receivers/nccl_fr`. Same shape as PR-B1: receiver-scoped self-telemetry helpers travel as siblings (`selftel.go` + `selftel_test.go`) co-located with the component, scope-name = the component's Go import path, metric names + label shape preserved bit-for-bit so dashboards / alerts do not regress. ## Root cause Not a bug — a planned dependency severance. v0.1.x stdoutexporter imports `internal/selftelemetry` for the `Exporter` interface, the `NewNoopExporter` / `NewExporter` constructors, the `Kind` type, and `RecordInitError`. RFC-0013 PR-F deletes `internal/selftelemetry` entirely (the runtime that consumes it dies in the same wave). Every component carrying that import has to migrate to a sibling helper before PR-F can land. PR-B1 did the receiver; this PR does the exporter. ## Changes - `components/exporters/stdoutexporter/selftel.go` (new): sibling `selfExporter` interface + `selfExporterImpl` + `recordInitError`. Emits `tracecore.exporter.calls_total{result,kind,component_id}` on a meter acquired from `set.Telemetry.MeterProvider`. Scope = `github.com/tracecoreai/tracecore/components/exporters/stdoutexporter` (OTel convention). - `components/exporters/stdoutexporter/selftel_test.go` (new): 7 tests — noop safety, nil-MP error sentinel, calls_total emission + `{result,kind,component_id}` label shape, scope-name standard, `init_errors_total` tick + nil-MP safety, factory noop-fallback wiring with synthetic register failure, and a compile-time sibling-types guard. - `components/exporters/stdoutexporter/stdoutexporter.go` (rewired): use sibling types; drop `internal/selftelemetry` import; drop `(*stdoutExporter).SelfExporter()`. ## ExporterCarrier removal — rationale PR-B1 (nccl_fr receiver) didn't expose any selftelemetry types to the runtime — it was a clean port. stdoutexporter is different: v0.1.x exposed `SelfExporter() selftelemetry.Exporter` so `cmd/tracecore/collect.collectFailureRateReaders` could feed `tracecore.exporter.failure_rate`. This PR drops that contract: - The runtime's reader-collection path silently skips components that don't implement `ExporterCarrier` — documented "no per-exporter signal" degraded mode. - stdoutexporter is the canonical debug / example exporter (writes JSON lines to stdout). Operators don't alert on its failure_rate. Real backends in `components/exporters/otlphttp` carry that contract instead. - `tracecore_exporter_failure_rate` still surfaces in scrape via the SLO observable gauge (reports 0 with no readers registered) — the M2 acceptance check in `cmd/tracecore/integration_telemetry_test.go` still passes. - `tracecore_exporter_calls_total` continues to surface because the sibling impl emits it on `set.Telemetry.MeterProvider` directly, just under a new scope. - PR-F deletes the entire `internal/selftelemetry` package, so the `ExporterCarrier` contract evaporates regardless. Rationale documented inline in `stdoutexporter.go` and the `selfExporter` docstring so a six-months-cold reader knows why the carrier is missing. ## Test plan - [x] `make check` (gofumpt, golangci-lint, go vet, go mod verify) — green - [x] `go test ./components/exporters/stdoutexporter/... -count=1` — 13 tests pass (7 new selftel + 6 existing) - [x] `go test -race ./components/exporters/stdoutexporter/... -count=1` — green - [x] `go test ./cmd/tracecore/ -run TestIntegration_TelemetrySurface_EndToEnd -count=1` — passes (asserts both `tracecore_exporter_calls_total` and `tracecore_exporter_failure_rate` appear in scrape) - [x] `go test ./... -count=1 -short` — full repo green; no name regression in `cmd/tracecore`, `internal/telemetry`, or `internal/selftelemetry` (which still has its own tests until PR-F) - [x] Scope-name pin: assertion against `github.com/tracecoreai/tracecore/components/exporters/stdoutexporter` in selftel_test.go - [x] Sibling-types compile-time guard: `asSelfExporter` helper forces compile break if the type ever moves back to `internal/selftelemetry` ## Pattern continuity References `#184` (PR-B1, nccl_fr) for the receiver sibling pattern. Test helpers (`newTestMeterProvider`, `collectRM`, `scopeOf`, `kvMatch`, `dumpNames`, `failingExporterMP`) mirror the nccl_fr equivalents so an M8+ exporter author can read either pair and infer the convention. ## Release notes NONE — this is a pivot-internal port. Metric names + labels are unchanged on the scrape surface. The unexported `SelfExporter()` method removal is not observable to any v0.1.x operator (only `cmd/tracecore`'s reader-collection path read it, and that path's "skip on absent" branch is documented behavior). ```release-notes NONE ``` --------- Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

## Summary Mirrors PR-B1's nccl_fr sibling pattern (#184) for `components/receivers/dcgm`. Adds two package-local files — `selftel.go` and `lifecycle.go` — and rewires the receiver + every test off `internal/selftelemetry` and `internal/runtime/lifecycle`. Both internal packages are slated for deletion in RFC-0013 PR-F; this PR removes dcgm as a blocker. - `selftel.go` owns the `Kind` type, the canonical `Kind*` const block, the receiver-local `kindWatch` / `kindMIG`, the `selfTelemetry` interface + OTel-backed impl, and `recordInitError`. Scope name pinned to the receiver's Go import path per the PR-B1 standard. - `lifecycle.go` is a single-source slim variant of the internal helper (no `Add()` since dcgm doesn't run auxiliary watchers). Exports `ErrAlreadyStarted` so the double-Start contract test continues to compile via `errors.Is`. - The dcgm-specific bits PR-B1 didn't cover: receiver-local Kind constants layered on the canonical set (same const block, no separate import), explicit `SetDegraded(true|false)` transition path exercised by `degraded_seconds_total` accumulation across cycles, and the `WithSelfTelemetry` test seam that now accepts the package-local interface. ## Why port before the planned delete dcgm is scheduled for removal in PR-F. The port still earns its keep because: - It collapses the PR-F critical path from two PRs (dcgm-delete → internal-pkg-delete) to one (PR-F can land both atomically once dcgm is no longer a dependent). - The work is a mechanical mirror of PR-B1 — `selftel.go` + `lifecycle.go` are near-line-for-line copies with the dcgm-specific delta isolated to the const block and the SetDegraded contract. - Per `feedback_no_bloat.md`, the only justification for adding code that's about to be deleted is "it unblocks a delete." That's exactly this PR's posture. ## Root cause RFC-0013's PR-B-series unblocks PR-F by decoupling every `components/*` receiver from `internal/*` packages that should not have receiver-level dependents. Each receiver owns its own self-telemetry surface + lifecycle helper; `internal/selftelemetry` + `internal/runtime/lifecycle` become deletable in PR-F. Not a workaround — the port is the actual fix. ## TDD coverage added in `selftel_test.go` - Noop safety across every canonical Kind + the receiver-local `kindWatch` / `kindMIG`. - `errNilMeterProvider` sentinel returned (not noop-substituted) when MP is nil. - `tracecore.receiver.errors_total` partitioned by `kind` with `component_id` label, for both canonical kinds AND local kinds (separate test). - `tracecore.receiver.emissions_total` monotonic + drops negative values. - OTel scope name pinned to `github.com/tracecoreai/tracecore/components/receivers/dcgm`. - `degraded_seconds_total` accumulates across multiple degrade→recover cycles (the dcgm-specific contract). - `init_errors_total` ticks with `kind=receiver` / `reason=instrument_register`; nil MP is silently safe. - AST-walk parity test that asserts `docs_parity_test.go`'s hand-maintained `canonicalKindByName` matches the const block in `selftel.go` — so a future `KindFoo` addition without a paired map entry fails at the moment of drift. ## Test plan - [x] `make check` (gofmt + tidy-check + golangci-lint + go vet + mod-verify) → clean. - [x] `go test ./...` → clean. - [x] `go test -race ./components/receivers/dcgm/...` → clean. - [x] `go build ./components/receivers/dcgm/...` → clean. - [x] New parity test `TestCanonicalKindByName_CoversConstBlock` self-validates against the const block. - [x] `TestRUNBOOK_KindsMatchEmitted` continues to pass — the AST walker resolves both the local-kind idents (`kindWatch`, `kindMIG`) and the now-bare canonical `KindFoo` idents. - [x] `TestReceiver_UsesLifecycleHelper` updated to assert the new `*lifecycle` field-type shape (Ident inside StarExpr) — a revert to the deleted `internal/runtime/lifecycle.Lifecycle` would fail here. - [x] `TestReceiver_M2WiringFromMeterProvider` + `TestReceiver_RecordsInitErrorWhenConstructorFails` continue to exercise the end-to-end Prometheus scrape path against the real `internal/telemetry` MeterProvider — noop-fallback observability preserved. ## Release notes ```release-notes NONE: internal refactor only. Receiver behavior + metric names + label shape unchanged. Operators see no scrape-side diff. ``` --------- Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

…189) ## Summary RFC-0013 PR-A2 (sequencing gate for PR-B2 / PR-F / PR-I): retire the hand-wired `./cmd/tracecore` entry point and adopt the OpenTelemetry Collector Builder (OCB) output at `./_build/tracecore` as the canonical binary. After this lands, all receivers register through OCB's generated `otelcol.Factories` instead of the bespoke `cmd/tracecore/components.go`. Big diff (-3,869 / +388 across 53 files) because PR-A2 is the load-bearing pivot point; PR-B2 / PR-F / PR-I that follow can land surgically. ```release-notes CLI surface change — `tracecore` now uses the upstream OCB-generated CLI: - `tracecore collect --config=…` → `tracecore --config=…` (collect was default; OCB main runs the collector by default) - `tracecore receivers list` → `tracecore components` (now shows receivers + processors + exporters + extensions + connectors) - `tracecore debug dump` → removed (no OCB equivalent; use `tracecore components` + the live config when filing issues) - `tracecore failure-inject {nccl-hang,pod-evict}` → use the standalone `tools/failure-inject` binary (already ships xid, nccl-hang, pod-evict, cpu-steal subcommands) - `--log.format=text` / `--shutdown.drain-budget=…` / `--version-short` → removed; OCB upstream uses `--feature-gates` + `--set` flags Chart-shape changes — operator-visible: - Default pipeline flips from clockreceiver→stdoutexporter (in-tree, not registered by OCB) to hostmetrics→debug (both upstream OCB-bundled). Fresh install on a no-GPU cluster boots and emits load-average metrics immediately, same as before. - `telemetry.listen` + `telemetry.paths.{metrics,healthz,readyz}` → `telemetry.metricsListen` + `telemetry.healthListen` + `telemetry.healthPath`. The legacy single-listener block is gone because upstream `service.telemetry` and `healthcheckextension` are two separate processes. Probes hit `:13133/` (healthcheckextension default) instead of `:8888/healthz`. Prometheus scrape port (8888) is unchanged. - Self-telemetry metric names rename `tracecore_*` → `otelcol_*` (upstream vocabulary). Dashboards on any `tracecore_receiver_*`, `tracecore_exporter_*`, `tracecore_queue_*`, `tracecore_component_*`, or `tracecore_build_info` must be rewritten — see `docs/migration/v0.1-to-v0.2.md` for the exact map. ``` ## Breaking changes — orphan components (PR-A2 → PR-J/K bridge) The OCB-assembled binary registers only the components in `builder-config.yaml`: 6 receivers, 4 exporters, 3 extensions, 4 processors. The chart's per-component toggles for the legacy in-tree set survive this PR so the values shape doesn't break for operators that pin them, but **enabling any of the following in chart values will cause the pod to fail at startup with an "unknown factory" error until PR-B2 / PR-J / PR-K rewire them**: | Component | Kind | Replacement (planned) | |---|---|---| | `clockreceiver` | receiver | `hostmetrics` (PR-E shipped; now default) | | `containerstdout` | receiver | `filelogreceiver` + container stanza + `file_storage` extension (PR-J) | | `dcgm` | receiver | `dcgm-exporter` DaemonSet + `prometheusreceiver` (PR-J) | | `k8sevents` | receiver | `k8sobjectsreceiver` + OTTL `k8s.event.hint` transform (PR-J) | | `kernelevents` | receiver | `journaldreceiver` + `filelogreceiver` (kmsg) + OTTL Xid transform (PR-J) | | `nccl_fr` | receiver | In-repo Go submodule via OCB `gomod:` (PR-B2 + PR-I) | | `pyspy` | receiver | Deferred until OTel Profiles GA | | `stdoutexporter` | exporter | `debug` (OCB-bundled; now default) | | `otlphttp` (in-tree clone) | exporter | `otlphttpexporter` (OCB-bundled; same `otlphttp` name in chart values, same field shape — `endpoint`, `compression`, `headers`, `tls.*`, `timeout`, `retry_on_failure`, `sending_queue`; pass-through render so any upstream field works) | To verify what's actually registered in the binary you're running: `./_build/tracecore components`. The chart's `NOTES.txt` surfaces a WARNING when an operator enables any of these, and the chart-render CI workflow now runs `tracecore validate` against the default + one-receiver-on fixtures so a chart edit that emits a non-OCB key trips CI before reaching `helm install`. ## What landed ### Deletions (3,032 LOC across 22 source + 7 test files) - `cmd/tracecore/` entire tree: `main.go`, `collect.go`, `validate.go`, `debug.go`, `receivers.go`, `signals.go`, `failure_inject.go`, `openflags_{linux,other}.go`, `receiver_variants{,_dcgm_cgo,_dcgm_stub}.go`, `components.go`, + every `_test.go` (`collect_test`, `debug_test`, `failure_inject{,_linux}_test`, `integration_test`, `integration_telemetry_test`, `main_test`, `receivers_test`) - `components.yaml` + `tools/components-gen/{main,main_test}.go` — superseded by `builder-config.yaml`; OCB owns codegen now - `components/receivers/kernelevents/runbook_test.go` — depended on the deleted `tracecore debug dump` subcommand; kernelevents itself is scheduled for deletion in PR-K ### Build path swap | File | Change | |---|---| | `Makefile` | `build` target now runs OCB (was: legacy `go build ./cmd/tracecore`); dropped `generate`, `generate-check`, `run`, legacy `-ldflags -X` version injection; `coverage` `-coverpkg` drops `./cmd/...` | | `install/kubernetes/tracecore/Dockerfile` | Builds via OCB (`make build` then copy `_build/tracecore`) | | `install/kubernetes/tracecore/templates/daemonset.yaml` | `args` drops the `collect` subcommand; probes hit the new `health` port (13133) at `healthPath` | | `install/kubernetes/tracecore/templates/_helpers.tpl` | `renderedConfig` emits upstream OTel shape — `service.telemetry.metrics.address` + `extensions.health_check` + `service.extensions: [health_check]` — instead of the legacy single-listener `telemetry:` top-level block | | `install/kubernetes/tracecore/values.yaml` | Default pipeline flipped to `hostmetrics → debug`; in-tree-only toggles kept with explicit doc-comments naming PR-J/K as their migration owner | | `.goreleaser.yaml` | Switched to `builder: prebuilt` against `./_build/{Os}-{Arch}/tracecore`; release.yml gains a per-platform pre-build step | | `.ko.yaml` | Builds from inside `./_build/` (the OCB submodule) via `KO_CONFIG_PATH=../.ko.yaml`; `main: .` | | `.github/workflows/ci.yml` | `package` job runs OCB; old `build-ocb` drift gate replaced by `smoke-test-binary` job consuming the package artefact | | `.github/workflows/release.yml` | New "Pre-build OCB binaries" step before goreleaser; ko-publish step runs from `cd ./_build/` so OCB submodule resolves | | `.github/workflows/chart.yml` | Restored the `tracecore validate` gate against the default + one-receiver-on chart renders; path triggers swap `cmd/tracecore/**` → `builder-config.yaml` | | `.github/workflows/install-bench.yml` | Path triggers swap `cmd/tracecore/**` → `builder-config.yaml` | | `builder-config.yaml` | `dist.version` bumped to `0.1.0-m9-alpha` to match `Chart.yaml` `appVersion`; `chart-appversion-check.sh` now reads it as source of truth | | `scripts/chart-appversion-check.sh` | Read `dist.version` from `builder-config.yaml` (was: `internal/version/version.go`) | | `scripts/smoke.sh` | Rewritten for OCB binary — hostmetrics → debug config, expects upstream lifecycle log lines | | `scripts/validator-recipe.sh` | `BIN` default now `./_build/tracecore` | | `scripts/{doc-check,no-autoupdate-check}.sh` | Drop `cmd` from scan paths | ### New integration seam - `internal/integration/ocb_scrape_test.go`: spawns `_build/tracecore` against a hostmetrics → debug config, polls the upstream `:NNNN/metrics` surface, asserts both `otelcol_process_uptime` and `otelcol_receiver_accepted_metric_points` are present, then SIGTERMs the subprocess and asserts clean exit. The test skips when `_build/tracecore` is absent so a fresh `git clone` + `go test ./...` stays green; `make build` is the prereq. This is the regression gate for the chart's operator-facing self-telemetry contract (RFC-0013 §3): if a future upstream OCB release renames either metric, the chart's `service.telemetry.metrics.address` advertisement breaks downstream dashboards silently — this test fires first. ### Integration recipe migration - `docs/integrations/examples/honeycomb.yaml` + `otel-backend.yaml`: `clockreceiver` → `hostmetrics` loadscraper (the OCB-supported equivalent per RFC-0013 PR-E). The other two recipes carry `pending-rfc-0013-pr-a` markers and are still skipped by `validator-recipe.sh`. ### Doc rot fixes - `docs/FAILURE-MODES.md`: rerouted 8 entries that referenced deleted in-tree tests to their upstream OCB owners. - `docs/FLAKY-TESTS.md`: moved the two in-tree integration flakes to Resolved. - `STYLE.md`: rewrote repo-layout + component-registration + CLI + build-release sections around OCB. - `PRINCIPLES.md`: dropped concrete-example reference to deleted file. - `install/kubernetes/tracecore/README.md`: ko local-build steps now run from inside `./_build/`. - `docs/rfcs/0013-distro-first-pivot.md`: PR-A2 entry rewritten as landed. - `docs/migration/v0.1-to-v0.2.md`: added rows for the self-telemetry metric-name rename (`tracecore_*` → `otelcol_*`) and the `telemetry.*` chart values key rename. ## Sequencing constraints honored - `components/receivers/{clockreceiver, containerstdout, dcgm, k8sevents, kernelevents, nccl_fr, pyspy}` survive as orphan-but-compiling code until PR-K deletes them along with the chart-fixture migration. - `internal/{pipeline, selftelemetry, telemetry, componentstatus, pipelinebuilder, consumer, fanout, runtime}` survive until PR-F (after PR-B1 lifted nccl_fr off `internal/selftelemetry` in #184). - `tracecore validate` gate in chart workflow is **restored** on the default + one-receiver-on fixtures (was temporarily disabled in the first push of this PR; the chart's renderedConfig template migration was completed in the same PR). ## Sequencing gate satisfied PR-B2 (`nccl_fr` import swap to upstream OCB types), PR-F (`internal/*` deletion), and PR-I (Go submodule extraction) are unblocked. The legacy boot path is gone; the OCB-driven boot path is live. ## Test plan - [x] `make build` → produces `./_build/tracecore` binary via OCB - [x] `./_build/tracecore --version` reports `0.1.0-m9-alpha` matching `Chart.yaml` - [x] `./_build/tracecore components` lists 6 receivers + 4 exporters + 3 extensions + 4 processors (the `builder-config.yaml` inventory) - [x] `./_build/tracecore validate --config=<rendered chart default>` exits 0 - [x] `./_build/tracecore validate --config=<rendered chart one-receiver-on fixture>` exits 0 - [x] `make smoke` passes (hostmetrics → debug, 1.5s window, clean shutdown) - [x] `make check` passes - [x] `go test ./internal/integration/...` passes (new OCB scrape test; ~1.2s) - [x] `helm lint install/kubernetes/tracecore` clean (1 chart, 0 failed; icon advisory only) - [x] `helm template demo install/kubernetes/tracecore --show-only templates/daemonset.yaml` renders `args: [--config=…]` + two ports (`telemetry`, `health`) + probes hitting `health` port at `/` - [x] `helm template demo install/kubernetes/tracecore --show-only templates/configmap.yaml | yq '.data["config.yaml"]'` renders upstream OTel shape: `service.telemetry.metrics.address`, `extensions.health_check`, `service.extensions: [health_check]` - [x] `conftest test --policy install/kubernetes/tracecore/policies/conftest/tracecore.rego /tmp/chart-render.yaml` — 51 tests pass - [ ] CI: `package` job builds amd64 + arm64 OCB binaries - [ ] CI: `smoke-test-binary` job runs `--version` + `components` on the package artefact - [ ] CI: `chart` workflow lints + templates + validate + yq + conftest pass - [ ] CI: `install-bench` workflow's kind-cluster install still rolls out (bench values updated to drop the obsolete `stdoutexporter` reference and align `otlphttp.endpoint` with the upstream `otlphttpexporter` schema — pass-through render so all upstream fields work without chart changes) --------- Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…block) (#196) ## Summary RFC-0013 PR-F unblock: ports `components/receivers/k8sevents` off `internal/selftelemetry` + `internal/runtime/lifecycle` to package-local sibling types, mirroring PR-B1 (#184, nccl_fr) and PR-B2 (#187, kernelevents). One more receiver off the internal helpers; clears another blocker for PR-F's deletion of those internal packages. ## Why `internal/selftelemetry` and `internal/runtime/lifecycle` are slated for deletion in RFC-0013 PR-F. Every in-tree user must migrate first. PR-B1 landed the pattern (`nccl_fr`); PR-B2 applied it to `kernelevents`; this PR applies it to `k8sevents`. Receivers landing post-PR-F will own their selftel + lifecycle in their own submodule per OTel convention — this PR puts k8sevents on that footing. ## Pattern choice (multi-source, keeps `Add()`) k8sevents is multi-source (Events informer + Node informer driver goroutines both join the receiver's lifecycle), so this PR follows the PR-B2 kernelevents sibling rather than the slimmer PR-B1 nccl_fr sibling — keeps `lifecycle.Add()` (with its post-Shutdown and pre-Start refusal guards, plus the mutex-guarded TOCTOU-safe `Add`/`Shutdown` interleave) so the SharedInformerFactory driver goroutine joins the same WaitGroup as the run-loop goroutine. ## What changed - **New** `components/receivers/k8sevents/selftel.go` — local `selfTelemetry` interface + noop + OTel-backed implementations + `recordInitError` fallback ticker. Instrumentation scope name pinned to k8sevents' Go import path per OTel convention (PR-I will move it to the submodule path when k8sevents goes external). Metric names (`tracecore.receiver.errors_total{kind,component_id}` and siblings) preserved so dashboards/alerts don't regress. - **New** `components/receivers/k8sevents/lifecycle.go` — local `lifecycle` helper (Start + Add + Shutdown + panic-recovery + TOCTOU-safe mutex-guarded `closed`/`Add` interleave). - **New** `selftel_test.go` + `lifecycle_test.go` — TDD-first, no `internal/telemetry` dep. Pin: noop safety across every kind, nil-MeterProvider error, errors_total kind+component_id labels, scope-name = receiver Go import path, init_errors_total tick, factory noop fallback, Start/Shutdown/Add contracts, Add refusal-modes, TOCTOU concurrent Add-during-Shutdown stress. - `factory.go`, `receiver.go` — drop internal imports, wire local `selfTelemetry` + `lifecycle`. `r.lc` retyped from `*lifecycle.Lifecycle` to `*lifecycle`; `r.telemetry` retyped from `selftelemetry.Receiver` to `selfTelemetry`. Exported `KindWatch` / `KindBackpressureDrop` / `KindNode*` constants retyped from `selftelemetry.Kind` to package-local `kind` alias (load-bearing for the `K8sEventsReceiverDegraded` alert label values, which stay byte-identical). Added re-exported `KindParse` / `KindDownstream` / `KindCardinality` / `KindPanic` so external `_test` packages can still partition assertions on canonical kinds without depending on the deleted internal package. - `export_test.go` — test-only `recordingTel` replaced by exported `CapturingTelemetry` (mirrors `selftelemetry.CapturingReceiver`, trimmed to the k8sevents surface). External `_test` packages get a stable shim that survives the internal/selftelemetry deletion. - `receiver_test.go`, `node_failure_modes_test.go` — drop `internal/selftelemetry` import; redirect through `k8sevents.CapturingTelemetry`. Per-test bodies otherwise unchanged (kept the `recordingTel` symbol as a thin wrapper). ## Verification - `make check` clean (golangci-lint 0 issues, vet clean, mod-verify) - `make verify` clean (incl. doc-check + alert-check + chart-check + no-autoupdate) - `go test -race -count=1 ./...` — entire repo green (incl. k8sevents + kernelevents + every internal package, fixture, tool) - `go test -race -count=1 ./components/receivers/k8sevents/...` — all green; lifecycle TOCTOU stress test passes deterministically under `-race` - Goroutine leaks: pre-existing `TestReceiver_GoleakNoLeakAfterShutdown` continues to pass after the lifecycle rewire ## Out of scope (deferred) - PR-F itself (deleting `internal/selftelemetry` + `internal/runtime/lifecycle`) — gated on the remaining 3 wave-1 consumers (pyspy, otlphttp, containerstdout) landing their ports. This PR is one of those four. ```release-notes NONE ``` Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

…k) (#194) ## Summary Ports `components/receivers/pyspy` off `internal/selftelemetry` + `internal/runtime/lifecycle` using the sibling co-location pattern established by PR #184 (nccl_fr reference impl). `selftel.go` + `lifecycle.go` now live next to the receiver source. Unblocks **RFC-0013 PR-F** deletion of those internal moats. ## What changed - `selftel.go` (new) — receiver-scoped `selfTelemetry` interface + `selfTelemetryImpl` (OTel-backed) + `noopSelfTelemetry` + `recordInitError`. Metric names + label shape preserved 1:1 with `internal/selftelemetry`: - `tracecore.receiver.errors_total{kind,component_id}` - `tracecore.receiver.emissions_total{component_id}` - `tracecore.receiver.collection_latency_seconds{component_id}` (12-bucket boundaries, matches v0.1.x) - `tracecore.receiver.degraded_seconds_total{component_id}` (observable counter) - `tracecore.receiver.last_activity_unix_seconds{component_id}` (observable gauge) - `tracecore.selftelemetry.init_errors_total{kind,component_id,reason}` (factory-fallback signal) - `lifecycle.go` (new) — slim single-source variant (no `Add()`). pyspy fans its three goroutines (scan loop, trigger cadence driver, trigger run loop) via its own `sync.WaitGroup` inside `runAll`, so lifecycle only needs `Start` + `Shutdown`. - `kinds.go` — `kind` is now a receiver-local type. `kindPanic` is declared locally (replaces `selftelemetry.KindPanic`). - `factory.go` — uses local `newSelfTelemetry` / `recordInitError` / `reasonInstrumentRegister`. - `pyspy.go` — fields/types switched to local sibling. - `pyspy_test.go` — uses sibling-local `fakeTelemetry` captor (replaces `selftelemetry.CapturingReceiver`). - `selftel_test.go` (new, RED-first TDD) — 7 tests covering noop safety, nil-MP, errors_total kind+component_id, scope-name pin, init_errors_total, nil-MP safety on recordInitError, factory fallback path. - `fake_telemetry_test.go` (new) — test-only captor implementing the `selfTelemetry` interface; mirrors the kernelevents `export_test.go` pattern. ## Scope-name standard Instrumentation scope pinned to the receiver's Go import path: ``` github.com/tracecoreai/tracecore/components/receivers/pyspy ``` When the receiver moves to `module/receiver/pyspyreceiver/` in PR-I.1, the scope name moves with it (matches OTel convention). ## Lifecycle pattern decision pyspy uses **single-source** lifecycle (like nccl_fr), NOT multi-source (like kernelevents). The receiver's `runAll` callback owns a local `sync.WaitGroup` and fans into three goroutines from within the single `lifecycle.Start(ctx, r.runAll)` call. No `Add()` needed → dropped from the slim sibling. ## Tests ``` $ go test ./components/receivers/pyspy/ -count=1 ok github.com/tracecoreai/tracecore/components/receivers/pyspy 0.735s $ go test -race ./components/receivers/pyspy/ -count=1 ok github.com/tracecoreai/tracecore/components/receivers/pyspy 1.745s $ make check 0 issues. ``` 13 targeted tests pass (7 new selftel + 6 existing receiver lifecycle tests). ## Wave context pyspy is 1 of 4 remaining unported consumers of `internal/selftelemetry` + `internal/runtime/lifecycle`. Sibling agents (otlphttp / containerstdout / k8sevents) are landing in parallel; this PR's scope is the pyspy receiver only — zero file overlap with the parallel work. ## Test plan - [x] `go test ./components/receivers/pyspy/ -count=1` — green - [x] `go test -race ./components/receivers/pyspy/ -count=1` — green - [x] `make check` — 0 lint issues, vet clean, modules verified - [x] Scope name asserted at `github.com/tracecoreai/tracecore/components/receivers/pyspy` - [x] Factory fallback ticks `tracecore.selftelemetry.init_errors_total` on instrument-register failure - [x] Noop hot-path methods never panic; nil-MP rejected with sentinel error - [x] Existing receiver tests (Start_EmptyUDSDir, Start_NonExistent, Start_TopLevelDisabled, Shutdown_HonorsContextBudget, Shutdown_Idempotent) still pass against the sibling ```release-notes NONE ``` --------- Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

## Summary - Port `components/receivers/nccl_fr` off the v0.1.x internal facades (`internal/pipeline`, `internal/consumer`, `internal/runtime/lifecycle`) onto upstream `go.opentelemetry.io/collector/{component,receiver,consumer}` v1.59.0 — the canonical types the OCB-generated `_build/main.go` already consumes for all third-party receivers. - Factory is now `receiver.NewFactory(componentType, createDefaultConfig, receiver.WithLogs(createLogs, component.StabilityLevelBeta))` instead of a hand-rolled struct implementing `internal/pipeline.ReceiverFactory`. Stability level (`Beta`) preserved across the swap so OCB-surfaced metadata doesn't regress. - Receiver struct renamed to `ncclFRReceiver` (avoids collision with the upstream `receiver` package name) and implements `receiver.Logs` via a `Start(ctx, component.Host) error` / `Shutdown(ctx) error` pair — the `pipeline.ComponentState` embed was dropped (upstream `component.Component` carries no equivalent mixin; the lifecycle bookkeeping the receiver actually needs lives in the sibling `lifecycle.go` helper added in PR-B1 #184). - Logger swapped from `*slog.Logger` → upstream's `*zap.Logger` (the type carried in `component.TelemetrySettings.Logger`). All log call sites converted to `zap.String/Int64/Duration/Error` fields; log messages and fields are byte-for-byte preserved so operator alerting on log content does not regress. ## Hard gate PR-I.1 (submodule extraction to `module/receiver/ncclfrreceiver/`) requires `grep -r 'internal/(pipeline|consumer|runtime/lifecycle)' components/receivers/nccl_fr/` to return zero hits. This PR clears it: ``` $ grep -rn 'internal/pipeline\|internal/consumer\|internal/runtime/lifecycle' components/receivers/nccl_fr/*.go (no matches) ``` (Two comment-only mentions remain in `lifecycle.go` and `factory.go` — historical context for the v0.1.x → v0.2.0 migration, not imports.) ## Predecessor PR-B1 #184 (merged 2026-05-30) ported the **self-telemetry + lifecycle** helpers into the package as siblings. This PR (PR-B2) handles the **pipeline + consumer + factory** layer — the last remaining `internal/*` imports. ## Test plan - [x] `go build ./...` — green - [x] `go test ./components/receivers/nccl_fr/... -race` — 12 tests pass - [x] `go test ./...` — green except pre-existing flake in `components/receivers/kernelevents/TestReceiver_SLIBudget` (verified flaky on stashed/PR-B2-not-applied tree) - [x] `make check` — golangci-lint + go vet + go mod verify — green - [x] Scope-name pin (`TestSelfTelemetry_ScopeNameIsReceiverImportPath`) still asserts `github.com/tracecoreai/tracecore/components/receivers/nccl_fr` - [x] Factory fallback contract (`TestFactory_FallsBackToNoopWhenMeterFails`) still surfaces `tracecore.selftelemetry.init_errors_total` when every `tracecore.receiver.*` instrument registration is synthetically failed ## Compatibility note `go.mod` now pins `go.opentelemetry.io/collector/{component,receiver,consumer} v1.59.0` — the v1.x stable line aligned with `pdata v1.59.0` the repo was already on. No transitive-dep churn beyond zap 1.24.0 → 1.28.0 (upstream component v1.59.0 requires it). ```release-notes NONE ``` ## Type-swap reference This PR is the first-of-kind upstream-API port; 7 more receiver/exporter ports (PR-F.2 series — clockreceiver, kernelevents, stdoutexporter, k8sevents, containerstdout, otlphttp, pyspy, dcgm) inherit this exact mapping. Use this table as reference when porting those components off `internal/pipeline` + `internal/consumer`. | Internal | Upstream | |---|---| | `internal/pipeline.Type` | `component.Type` | | `internal/pipeline.ReceiverFactory` | `receiver.Factory` | | `internal/pipeline.CreateSettings` | `receiver.Settings` (via `receivertest.NewNopSettings` in tests) | | `internal/pipeline.Config` | `component.Config` | | `internal/pipeline.Receiver` | `receiver.Logs` (`= interface{ component.Component }`) | | `internal/consumer.Logs` | `consumer.Logs` | | `*slog.Logger` | `*zap.Logger` | | `internal/pipeline.MustNewType` | `component.MustNewType` | | `internal/pipeline.MustNewID` | `component.NewIDWithName` | ## Deep-review cleanup (post-aec83be) A follow-up commit applies five reference-pattern fixes so PR-F.2 inherits the cleanest possible template: 1. Deleted dead `var Factory` + indirection wrapper — `NewFactory()` now constructs the factory directly, mirroring upstream `otlpreceiver` / `filelogreceiver`. The `tools/components-gen` driver that motivated the package-var was deleted in PR-A2 #168. 2. Fixed misleading comment claiming `receiver.Logs` carries a "LogsReceiver tag" — upstream defines `receiver.Logs` as `interface { component.Component }`; the type identity is a documentation marker only. 3. Switched tests to upstream `receivertest.NewNopSettings(componentType())` so test Settings auto-track upstream field additions (`BuildInfo`, `TelemetrySettings`). A thin `testSettings()` wrapper pins the ID to `nccl_fr/test` so selftel label assertions stay deterministic. 4. Renamed unexported `ncclFRReceiver` → `ncclfrReceiver` per Go acronym convention (lowercased) and aligned with the planned PR-I.1b package name `ncclfrreceiver`. 5. Appended this Type-swap reference section. --------- Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

#206) ## Summary Deletes the three internal moats and the in-tree DCGM receiver that RFC-0013 §migration step 8 promised — the payoff for the wave-3 sibling-port PRs (#184/#185/#186/#187/#188/#193/#194/#196/#197). **Net: -12,482 LOC across 92 files (78 deletions, 14 modifications).** ### What deletes | Path | LOC | Why safe now | |---|---|---| | `components/receivers/dcgm/` | 7,604 | cgo stub never shipped real code; #188's PR-B2-shaped dcgm sweep already removed the live port surface. | | `pkg/dcgm/` | 922 | Only consumer was the deleted receiver. Bonus cleanup. | | `internal/selftelemetry/` | 1,946 | Every consumer (containerstdout, clockreceiver, kernelevents, k8sevents, nccl_fr, dcgm, pyspy, stdoutexporter, otlphttp) ported onto receiver/exporter-scoped sibling `selftel.go` files. | | `internal/telemetry/` | 1,991 | Probes flow through upstream `healthcheckextension`; MeterProvider via upstream `service.telemetry`. Only remaining consumers were `internal/selftelemetry/*_test.go` (deleted together) + one orphan clockreceiver test. | | `components/receivers/clockreceiver/errors_integration_test.go` | 100 | Orphan from #185's PR-B1 clockreceiver port — bootstrapped via the deleted `selftelemetry.Receiver` interface but never migrated to the receiver-scoped sibling `selftel.go`. Covered behaviour ("errors_total surfaces on downstream failure") is now exercised through clockreceiver's sibling tests. | ### Pre-flight grep evidence (post-merge of origin/main) ``` $ grep -rn "tracecoreai/tracecore/internal/selftelemetry" --include="*.go" . (zero matches) $ grep -rn "tracecoreai/tracecore/internal/telemetry" --include="*.go" . (zero matches) $ grep -rn "tracecoreai/tracecore/components/receivers/dcgm" --include="*.go" . $ grep -rn "tracecoreai/tracecore/pkg/dcgm" --include="*.go" . (zero matches) ``` ### Tooling - Retire the `dcgm` build tag — `make build-tags` no longer vets `-tags dcgm` (kept as a hook for future build-tag-gated paths). - `make bench-check` loop drops both deleted package rows (`internal/telemetry`, `components/receivers/dcgm`). - `scripts/register-lint.sh` allowlist emptied (the two `internal/telemetry/{build_info,slo}.go` entries are gone with the package; allowlist comment notes the post-PR-F.1 state). - `go.mod` direct deps shrink — `github.com/prometheus/client_golang` and `go.opentelemetry.io/otel/exporters/prometheus` drop to indirect (they were used by `internal/telemetry/server.go`). ### Chart toggles intentionally retained Chart `receivers.dcgm` toggle + `templates/NOTES.txt` warning + `templates/_helpers.tpl` doc-comment list keep the `dcgm` symbol for the migration window. The toggle has been inert since PR-A2 — operators enabling `receivers.dcgm.enabled=true` already crashed at boot because the OCB binary doesn't register the factory. PR-K removes the toggle entirely alongside the chart-default flip from `clockreceiver` → `hostmetrics` and the v0.2.0 recipe migration. ### Doc sweep - `internal/runtime/lifecycle/lifecycle.go` doc-comment: drop the dcgm pointer; flag containerstdout as the sole remaining in-tree consumer; reschedule the package itself for PR-F.2 deletion once containerstdout ports off the helper or PR-K.2 deletes the receiver. - `docs/FAILURE-MODES.md` self-tel-surface rows rewired from `internal/telemetry/server_test.go::*` (deleted) to upstream-delegated wording. - `docs/patterns/{README,pattern-{1,3,4,5}}.md` replay-test pointers updated — the in-tree `components/receivers/dcgm/pattern_replay_test.go` is gone; pattern replay now flows through `docs/integrations/prometheus-scrape.md` (PR-J's upstream `dcgm-exporter` recipe). - `docs/README.md` per-component table: drop the deleted `internal/telemetry/{README,SECURITY}.md` rows + the deleted `components/receivers/dcgm/{README,RUNBOOK}.md` rows. - `STYLE.md` vendor-SDK section: drop the `pkg/dcgm/` reference + the `//go:build dcgm` example; explicit cross-reference to PR-F.1 in the integration-test build-tag note. - `CHANGELOG.md`: PR-F.1 landed entry under Unreleased; "Remaining v0.1.0 work" line updated to point at PR-F.2. - `docs/rfcs/0013-distro-first-pivot.md` §migration step 8: PR-F entry replaced with the PR-F.1/PR-F.2 split + the explicit rationale (componentstatus travels with pipeline; pipeline is out of PR-F's scope per line 240's original framing). ### Out of scope (PR-F.2 follow-up) - `internal/componentstatus/` — 5-line `ReportStatus` free function. Travels with `internal/pipeline` (its only non-test consumers are `internal/pipeline/runtime_test.go` + `internal/pipeline/pipelinetest/fixture_test.go`). Deletion lands when pipeline migrates to upstream `go.opentelemetry.io/collector/component/componentstatus`. ### Rationale links - RFC-0013 §migration step 8 — the PR-F entry now codifies the F.1/F.2 split in this branch's RFC update. - PR-B2 scope-discovery (#188) — established the "rename + slim, don't reshape" pattern for the dcgm sweep that retired the cgo path. - Wave-3 PRs that unblocked selftelemetry deletion: #184 (nccl_fr), #185 (clockreceiver), #186 (kernelevents), #187 (stdoutexporter), #188 (dcgm), #193 (otlphttp), #194 (pyspy), #196 (k8sevents), #197 (containerstdout). ```release-notes [CHANGE] internal/{selftelemetry,telemetry} packages deleted; components/receivers/dcgm + pkg/dcgm deleted. Operators using the v0.1.x in-tree `tracecore.*` self-telemetry metric names migrate per docs/migration/v0.1-to-v0.2.md. Third-party importers of internal/* (unlikely pre-1.0) lose the `selftelemetry.{Receiver,Exporter}` interfaces and the `telemetry.MeterProvider` wrapper; receiver/exporter authors now wire a receiver-scoped sibling `selftel.go` per the PR-B1 pattern. ``` ## Test plan - [x] `make verify` (lint + vet + tidy-check + mod-verify + license-check + generate-fixtures-check + build-tags + nccl-fr-rce-gate + register-lint + actionlint + zizmor + doc-check + no-autoupdate-check) — exit 0. - [x] `go test ./...` — all green (29 packages). - [x] `make build` (OCB) — `./_build/tracecore` produced. - [x] `./_build/tracecore --version` — `tracecore version 0.1.0-m9-alpha`. - [x] Pre-flight greps for all four deleted paths — zero external importers. - [ ] CI green on PR (linux/race matrix, chart render, install-bench, zizmor, govulncheck). - [ ] Operator verification that the chart's `dcgm` toggle remains inert post-merge (no behaviour change from main — already inert since PR-A2). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

## Summary Deletes the seven `internal/*` packages that RFC-0013 §migration step 8 PR-F.2 promised once the upstream-port wave (#201/#202/#203/#204/#205/#207/#208/#209) cleared every external caller of the in-tree pipeline runtime. **Net: -6,888 LOC across 56 deleted files, +80 LOC across 14 modified files. 70 files total.** This is the final cut of RFC-0013 §migration step 8 PR-F. ## What deletes | Path | LOC | Replacement | |---|---|---| | `internal/pipeline/` | 4,134 | `go.opentelemetry.io/collector/service` (OCB-generated `_build/main.go` consumes `builder-config.yaml`). | | `internal/pipelinebuilder/` | 1,282 | Same — assembly is upstream `service`. | | `internal/config/` | 718 | Upstream `confmap` providers (`file`, `yaml`, `env`). | | `internal/consumer/` | 87 | Upstream `go.opentelemetry.io/collector/consumer`. | | `internal/fanout/` | 366 | Upstream `internal/fanoutconsumer` (collector module). | | `internal/componentstatus/` | 16 | Upstream `component/componentstatus.ReportStatus` (same free-function shape). | | `internal/runtime/lifecycle/` | 505 | Per-receiver package-local `lifecycle.go` siblings — already ported during the PR-B1 wave (#184/#185/#186/#187/#194/#196/#197); the in-tree helper had no remaining non-test consumer after PR-F.1 + the wave-2 upstream-port PRs. `kernelevents/lifecycle.go` was inherited from k8sevents (#208). | ## Pre-flight grep evidence ``` $ grep -rn 'tracecoreai/tracecore/internal/(pipeline|consumer|pipelinebuilder|config|fanout|componentstatus|runtime/lifecycle)' --include='*.go' . (zero matches) ``` ## Tooling - `.golangci.yml` `ignore-interface-regexps` repointed at upstream `consumer.{Metrics,Traces,Logs}` + `component.Component`. The in-tree-only same-package-error-wrap exemption stays — the STYLE rule applies regardless of which interface is forwarded. - `.github/workflows/chaos.yml` drops the `chaos-pipeline-test` job (the in-tree `internal/pipeline/chaos_test.go` is gone; upstream `service` provides the equivalent panic-recovery contract). `harness-determinism` (failure-inject golden-SHA), `cpu-steal-mpstat`, `pattern-pod-evicted` jobs preserved. - `.github/workflows/install-bench.yml` drops the `internal/{pipeline,runtime,selftelemetry}/**` path-filter rows. - `go.mod` / `go.sum` unchanged. ## Doc sweep - `CHANGELOG.md` Unreleased: PR-F.2 landed entry replacing the "PR-F.2 deferred" sentence; "Remaining v0.1.0 work" line updated; one dead `internal/pipeline/README.md` link in Foundation block rewritten as "deleted at v0.1.0". - `docs/rfcs/0013-distro-first-pivot.md` §7 deletion table: both pipeline-internals and runtime/lifecycle rows updated from "v0.1.0 (audit first…)" / "v0.2.0 (with last consumer)" to "v0.1.0 (landed PR-F.2)". §migration step 8 reframed. - `docs/FAILURE-MODES.md` Lifecycle / Data flow / Shutdown timing / Backend tables rewired from in-tree `internal/{config,pipeline,fanout}/*_test.go::TestName` pointers to upstream-delegated wording matching the pattern PR-A2 established. - `docs/STRATEGY.md` "Post-RFC-0013 status" intro updated; "Stable interfaces in `internal/pipeline/`" graduation row rewritten to point at the upstream surface. - `docs/migration/v0.1-to-v0.2.md` `internal/*` section status banner flipped from "deferred, still present in RC builds" to "landed, deleted in v0.2.0 builds". - `MILESTONES.md` v0.1.0 deletions row extended with boot-path internals; M1 + M4b + M19 rubric details annotated with the PR-F.2 retirement. - `README.md` Contributor row repointed at upstream `go.opentelemetry.io/collector` package docs. - `AGENTS.md` "Self-telemetry internals" bullet split into "Self-tel internals" + "Pipeline / boot-path internals" with explicit deletion status. - `docs/README.md` table row for `internal/pipeline/README.md` dropped. - `components/receivers/kernelevents/README.md` lifecycle-sibling rationale updated to past-tense. - `tools/failure-inject/README.md` "Testing locally" section drops the `-tags=chaos ./internal/pipeline/...` invocation. ## Sequencing This PR is hard-gated on every upstream-port PR landing first: - #201 nccl_fr (PR-B2) - #202 stdoutexporter - #203 pyspy - #204 k8sevents - #205 clockreceiver (PR-B3) - #207 otlphttp - #208 kernelevents - #209 containerstdout - #206 PR-F.1 (selftel / telemetry / dcgm) All nine merged before this PR opened; this is the moat-deletion payoff. Remaining v0.1.0 work is PR-K (chart-default flip + `clockreceiver` + `stdoutexporter` + remaining receiver source deletions, coupled with test-fixture migration and the `telemetry:` values-key deprecation cycle). ## Test plan - [x] `make check` — golangci-lint 0 issues, go vet clean, go mod verify ok. - [x] `go build ./...` — clean. - [x] `go test -count=1 ./...` — green (excluding the known `kernelevents/TestReceiver_SLIBudget` flake called out in #205's body, which only triggers under heavy parallel `go test ./...` load; passes standalone). - [x] `grep` confirms zero non-internal callers of the deleted packages. - [x] Doc-check pre-push hook passes after the CHANGELOG dead-link fix. ```release-notes [CHANGE] internal/{pipeline,pipelinebuilder,config,consumer,fanout,componentstatus,runtime/lifecycle} packages deleted. The OCB-generated boot path off builder-config.yaml replaces them. Third-party importers of internal/* (unlikely pre-1.0; the packages live under internal/ and the Go compiler rejects external imports) lose the pipeline-assembly + lifecycle + config-loader surfaces; receiver authors now wire against upstream go.opentelemetry.io/collector/{component,receiver,consumer,pipeline} directly. See docs/migration/v0.1-to-v0.2.md "internal/* package deletion". ``` --------- Signed-off-by: Tri Lam <tri@maydow.com> Co-authored-by: Tri Lam <tri@maydow.com>

feat(pivot): PR-B1 — port nccl_fr off internal selftel + lifecycle

aac0012

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Signed-off-by: Tri Lam <tri@maydow.com>

trilamsr enabled auto-merge (squash) May 31, 2026 03:47

trilamsr merged commit b4b6e9b into main May 31, 2026
13 of 14 checks passed

trilamsr deleted the pr-b1-nccl-fr-port branch May 31, 2026 03:47

trilamsr mentioned this pull request May 31, 2026

feat(pivot): port clockreceiver off internal selftel + lifecycle #185

Merged

5 tasks

This was referenced May 31, 2026

feat(pivot): port stdoutexporter off internal/selftelemetry #186

Merged

feat(pivot): port kernelevents off internal selftel + lifecycle #187

Merged

feat(pivot): PR-B2 — port dcgm off internal selftel + lifecycle #188

Merged

trilamsr mentioned this pull request May 31, 2026

feat(ocb): PR-A2 — switch tracecore entrypoint to OCB-generated main #189

Merged

16 tasks

This was referenced May 31, 2026

feat(pivot): port pyspy off internal selftel + lifecycle (PR-F unblock) #194

Merged

feat(pivot): port k8sevents off internal selftel + lifecycle (PR-F unblock) #196

Merged

trilamsr mentioned this pull request May 31, 2026

feat(pivot): PR-B2 — port nccl_fr off internal pipeline+consumer #201

Merged

6 tasks

trilamsr mentioned this pull request May 31, 2026

feat(pivot): PR-F.1 — delete internal moats (selftel/telemetry) + dcgm #206

Merged

7 tasks

trilamsr mentioned this pull request May 31, 2026

feat(pivot): PR-F.2 — delete in-tree boot-path internals #215

Merged

5 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(pivot): PR-B1 — port nccl_fr off internal selftel + lifecycle#184

feat(pivot): PR-B1 — port nccl_fr off internal selftel + lifecycle#184
trilamsr merged 1 commit into
mainfrom
pr-b1-nccl-fr-port

trilamsr commented May 31, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

trilamsr commented May 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Why receiver-local scope (vs reusing the internal scope name)

Test plan

Follow-ups

Release notes

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

trilamsr commented May 31, 2026 •

edited

Loading