## What is wrong `docs/dev/gitlab-com-infrastructure.md` states that the service has no dashboards and no alerting. That statement was correct when its author wrote it. [runbooks!11379](https://gitlab.com/gitlab-com/runbooks/-/merge_requests/11379) merged into `master` on 2026-08-31 as `0d9639ffe` and made it false. The claim is not one sentence. Four documents in this repository state it, cite it, or rest on it. A one-line edit corrects the sentence that names the problem and leaves the rest standing. What `runbooks!11379` created, read from its own diff at `0d9639ffe`: | Artifact | Where | | --- | --- | | Metrics-catalog entry | `metrics-catalog/services/artifact-registry-gke.jsonnet`, on the `runway-k8s-archetype` | | Service dashboard | `dashboards/artifact-registry-gke/main.dashboard.jsonnet` | | Rule files | 14 files under `mimir-rules/runway/artifact-registry-gke/`, all named `autogenerated-*.yml` | | Alert classes | `ApdexSLOViolation`, `ErrorSLOViolation`, `TrafficCessation`, `TrafficAbsent`, `KubeContainersWaitingInError`, and six `component_saturation_slo_out_of_bounds:*` records | | Slack route | `services/teams.yml` gains the team `artifact_registry`, with `slack_alerts_channel: f_artifact_registry_alerts` | | Documentation pointer | `services/service-catalog.yml` sets `documents.service` to `docs/dev/gitlab-com-infrastructure.md`, and names `artifact-registry-gke-main/artifact-registry-gke-overview` as the primary dashboard | The generated SLIs read Runway load-balancer metrics and Kubernetes metrics. They select no `gitlab_artifact_registry_*` series. The jsonnet overrides `lbSelector` on `url_map_name` and sets `disableOpsRatePrediction: true`. That fact decides whether an AR-specific rule still needs hand-writing, so a corrector states it rather than writing "the alerts exist". ## Why the false sentence costs something `runbooks!11379` points the service catalog at this exact file. The alerts it created route to `#f_artifact_registry_alerts`. An engineer paged by one of those alerts follows the service catalog to a page that says the service has no dashboard and no alerting. The same catalog entry names the dashboard that engineer needs. ## The sites, measured at one commit Measured at `30636aba3` on `origin/main`, 2026-08-31. Line numbers move. If a number does not match, search for the quoted text. | Site | What it says | Status | | --- | --- | --- | | `docs/dev/gitlab-com-infrastructure.md:168` | "The service's own dashboards and alerting do not exist yet." | **False.** | | `docs/dev/gitlab-com-infrastructure.md:169-174` | #354 tracks the metrics-catalog entries that generate the service dashboard with default SLIs and alerts, "plus AR-specific dashboards and alerting". | **Half true.** The catalog entry and the generated defaults landed. The AR-specific half is #354 Phase 3 and Phase 4, and both are open. | | `docs/dev/gitlab-com-infrastructure.md:175` | "That work depends on metrics collection working first." | **Discharged.** | | `docs/specs/S22-storage-accounting.md:35` | "no alerting surface exists to wire it against beforehand" | **False.** `mimir-rules/runway/artifact-registry-gke/` is that surface. | | `docs/specs/S22-storage-accounting.md:826` | "no `gitlab_artifact_registry_*` series reach the Mimir tenant from either environment … (see `docs/dev/gitlab-com-infrastructure.md`)" | **Dead citation**, and the Mimir half is unmeasured. See the two sections named `## Part of this correction predates runbooks!11379` and `## The Mimir clause needs a measurement nobody has taken`. | | `docs/specs/S22-storage-accounting.md:827` | "#354 … is itself blocked on collection working first." | **Discharged.** | | `docs/specs/S22-storage-accounting.md:1103` | "since no alerting surface exists to wire them against beforehand" | **False**, for the same reason as line 35. | | `docs/plans/2026-08-04-s22-storage-accounting.md:319` | Repeats the infrastructure page's two claims by citation. | **False** on the dashboards-and-alerting clause. The Mimir clause is unmeasured, and the citation is dead. | | `docs/plans/2026-08-04-s22-storage-accounting.md:373` | "no `gitlab_artifact_registry_*` series reaching Mimir from either environment" | **Unmeasured.** | | `docs/plans/2026-08-03-usage-data-emission-hardening.md:16` | "alert rules (blocked on the metrics pipeline reaching Mimir)" | **Unmeasured.** | | `docs/plans/2026-08-03-usage-data-emission-hardening.md:62` | "AR's app metrics do not reach Mimir yet" | **Unmeasured.** | | `docs/plans/2026-08-03-usage-data-emission-hardening.md:479` | "blocked on the AR metrics pipeline reaching Mimir" | **Unmeasured.** | ## Part of this correction predates runbooks!11379 `3c2cd7f42` (2026-08-12) removed the sentence "no `gitlab_artifact_registry_*` series reach the Mimir tenant from either environment" from `docs/dev/gitlab-com-infrastructure.md`. The ServiceMonitor paragraph replaced it. The spec clause landed at `2bc4d531b` on 2026-08-05, and the plan clause at `af842d383` on 2026-08-10. Both cite a sentence that the page has not carried since 2026-08-12. The citation is therefore dead whatever a fresh Mimir measurement returns, and this half of the correction is not `runbooks!11379`'s doing. ## What !1973 changes !1973 converts the S22 spec and plan into `docs/dev/storage-accounting.md`. While !1973 is open, the twelve sites in the table above are the whole list. Once !1973 merges, six of them are gone: it reduces `docs/specs/S22-storage-accounting.md` to a 26-line stub that carries none of lines 35, 826, 827 or 1103, and it removes `docs/plans/2026-08-04-s22-storage-accounting.md`. Once !1973 merges, five new sites exist in `docs/dev/storage-accounting.md`. Measured at !1973's head `bd563e14f`: | Line at that head | Text | Status | | --- | --- | --- | | 1302 | "**Wiring them is blocked outside this repository.**" | **Half true.** The AR-specific rules are still outside this repository, and "blocked" is now weaker. | | 1303 | "There is no alerting here … and no `gitlab_artifact_registry_*` series reach the Mimir tenant from either environment, so no rule has a series to select on." | First clause **true**. Mimir clause **unmeasured**. | | 1304 | "See [gitlab-com-infrastructure.md](gitlab-com-infrastructure.md)." | Cites the page this issue is about. | | 1305 | "#354 tracks the metrics-catalog work and is itself blocked on collection working first." | **Discharged.** | | 1381 | "The three paging alerts have nowhere to live" | **False.** `mimir-rules/runway/artifact-registry-gke/` is where they live. | At `bd563e14f` the merge request reports a conflict with `main`, so these five line numbers move again on a rebase. These five lines are !1973's own new text. Correcting them inside !1973 keeps that merge request's content true, which is author correctness rather than scope this issue takes. ## The Mimir clause needs a measurement nobody has taken Six sites above turn on one question: do `gitlab_artifact_registry_*` series reach the `Mimir - Runway` tenant? This issue does not answer it, and neither the run that filed it nor its dispatcher could take the measurement. #354's description claims that the Redis-lease-gated `reconciliation_backlog` gauge emits in production. `docs/dev/observability.md:93` names that gauge `gitlab_artifact_registry_reconciliation_backlog`. That claim is an assertion in a mutable issue field, not a measurement. What settles it: a query for any `gitlab_artifact_registry_*` series against the `Mimir - Runway` datasource (UID `mimir-runway`) in [Grafana](https://dashboards.gitlab.net), run once per environment. The scripted form needs a Grafana service-account token, which `docs/dev/gitlab-com-infrastructure.md:116-122` describes. Whoever corrects these six sites takes that measurement first, and records the date it was taken. One caution on the selector. Nobody has run `process_open_fds{type="artifact-registry-gke"}` or `count by (env) (process_open_fds)` against this tenant. A `type` label on the service's own `/-/metrics` series is therefore unverified. `docs/dev/gitlab-com-infrastructure.md:142-147` names `env` as the environment filter, and it names `artifact-registry-gke` as the Kubernetes namespace. Use `env` plus the Kubernetes namespace, and treat any `type` selector as unproven until a query returns rows for it. ## Sites that hold, so a correction does not go too far Measured at the same commit, `30636aba3`. | Site | Why it stands | | --- | --- | | `docs/specs/S22-storage-accounting.md:826`, first clause | "This repository contains no alerting whatsoever — no rule files, no `PrometheusRule`, no rules-testing job". A `PrometheusRule` grep over the tree returns only the S22 spec and the S22 plan, and the tree carries no rule file. The rules landed in `gitlab-com/runbooks`, not here. | | `docs/plans/2026-08-04-s22-storage-accounting.md:318` | The same claim, plus "no `deploy/`, `helm/`, or `runbooks/` tree", which holds for a top-level tree. `docs/runbooks/` carries operator procedures, not rules. | | `docs/plans/2026-08-04-s22-storage-accounting.md:371` and `docs/specs/S22-storage-accounting.md:828` | Both say the three alert rules land with #354. That holds while #354 Phase 4 is open. | | `docs/roadmap/closed-beta.md:44` | "Grafana dashboards and alerting split out to #354". A routing claim, not an existence claim. | ## Related items #354 owns the work that made the claim false. `runbooks!11379` is item 1 of its Phase 1 checklist, and #354 ticked that item on 2026-08-31. #354's description carries five phases and no item for correcting documentation in this repository, so this issue records a correction that #354 does not. #354 has no assignee on the day this issue is filed. !1973 writes five new copies of the same claim into `docs/dev/storage-accounting.md`, and it removes six of the twelve sites listed above. Which of the two merges first decides which list a corrector works from. ## Duplicate search before filing A search for `gitlab-com-infrastructure` over issue titles and descriptions returned no issue. A search for `alerting do not exist` returned 21 issues, and none of them is about this correction. A search for `11379` returned one issue, #354. A sweep of all 87 open merge requests by changed path returned two that touch any of the four files: !1973, and !2044, whose only hunk on the S22 plan sits near line 360 and corrects a different sentence. So no issue tracks this correction, and no open merge request fixes it. _This is a bot message 🤖 — /smurfit_