feat(monitoring): rename the gitborg_* metric prefix to bitborg_* in every producer #414

Sammanfogat

supernaut sammanfogade 4 incheckningar från feat/rename-7b-metric-prefix in i main

2026-08-10 18:31:57 +00:00

supernaut kommenterad

2026-08-10 18:25:33 +00:00

Ägare

ADR 0039 §7b. Stacked on #413 — base is the §7a branch so this diff shows only the producer rename. #413 must merge first; the dual-brand consumers are the safety net that makes this window-free.

166 renames across 23 files.

An allowlist, not a prefix sweep

A blanket gitborg_ → bitborg_ would have renamed two values that are not metrics and are load-bearing:

Token	What it actually is
`gitborg_web`	the live Postgres database and role, explicitly `LEGACY-PIN`ned by the §5 decision. Renaming the declaration creates a fresh empty database and the portal comes up blank.
`gitborg_ephemeral`	`controller.py`'s `METADATA_TAG` — the OpenStack metadata key deciding which VMs the runner-controller may DELETE. Renamed in code but not on live VMs, the controller stops recognising its own fleet.

Also excluded: gitborg_domain, gitborg_network_id (variable / tofu-output names) and gitborg_reconciler_image* (an Ansible var prefix). 59 renamed, 5 excluded, each classified by hand.

controller.py builds 23 metric names from _METRIC_PREFIX = "gitborg_runner_controller", which is why enumerating full names in the repo undercounts the surface. That one assignment moved and carries all 23.

Reconciler prose deliberately left on the old name

Seven gitborg_reconciler_* metrics are the only ones this repo does not produce — they come from bitborg-auth-reconciler/src/metrics.ts, a separate repo shipped as a container image. Moving them needs a code change, a release, an image-tag bump and an apply (§7d).

Their exprs are already dual-brand and match either name, so they need nothing. Their descriptions still say gitborg_reconciler_* because that is what production emits today, with a comment saying why. An operator grepping mid-incident for a metric name nothing emits is exactly the misdirection this plan keeps warning about.

Verified

vmalert:v1.147.0 -dryRun PASS; 46 alerts, set identical to pre-change, same for:/severity: counts
test_controller_logic.py — the suite CI runs, and it asserts on rendered metric output: 69/69 checks passed
gitborg_web still 2 hits in group_vars; gitborg_ephemeral still 7 in controller.py — both checked explicitly
dual-brand selectors intact: 46 dashboards, 35 alert-rules
dashboard JSON valid; ansible-lint 0/0 across 194 files at production; pinned shellcheck clean

Expected transient on apply

{__name__=~"[bg]itborg_x"} matches two series while retention holds the old one. They don't overlap in time except within one 5-minute staleness lookback at the switchover, so sum(...) panels can briefly read double. No alert aggregates over these metrics, so alerting is unaffected. for: clocks reset and increase() under-reports for one window — the documented discontinuity you chose.

ADR 0039 §7b. **Stacked on #413** — base is the §7a branch so this diff shows only the producer rename. #413 must merge first; the dual-brand consumers are the safety net that makes this window-free. 166 renames across 23 files. ## An allowlist, not a prefix sweep A blanket `gitborg_` → `bitborg_` would have renamed two values that are **not metrics** and are load-bearing: | Token | What it actually is | | --- | --- | | `gitborg_web` | the live Postgres **database and role**, explicitly `LEGACY-PIN`ned by the §5 decision. Renaming the declaration creates a fresh empty database and the portal comes up blank. | | `gitborg_ephemeral` | `controller.py`'s `METADATA_TAG` — the OpenStack metadata key deciding **which VMs the runner-controller may DELETE**. Renamed in code but not on live VMs, the controller stops recognising its own fleet. | Also excluded: `gitborg_domain`, `gitborg_network_id` (variable / tofu-output names) and `gitborg_reconciler_image*` (an Ansible var prefix). **59 renamed, 5 excluded**, each classified by hand. `controller.py` builds 23 metric names from `_METRIC_PREFIX = "gitborg_runner_controller"`, which is why enumerating full names in the repo undercounts the surface. That one assignment moved and carries all 23. ## Reconciler prose deliberately left on the old name Seven `gitborg_reconciler_*` metrics are the only ones this repo does not produce — they come from `bitborg-auth-reconciler/src/metrics.ts`, a separate repo shipped as a container image. Moving them needs a code change, a release, an image-tag bump and an apply (**§7d**). Their `expr`s are already dual-brand and match either name, so they need nothing. Their **descriptions** still say `gitborg_reconciler_*` because that is what production emits today, with a comment saying why. An operator grepping mid-incident for a metric name nothing emits is exactly the misdirection this plan keeps warning about. ## Verified - `vmalert:v1.147.0 -dryRun` **PASS**; 46 alerts, set identical to pre-change, same `for:`/`severity:` counts - `test_controller_logic.py` — the suite CI runs, and it asserts on rendered metric output: **69/69 checks passed** - `gitborg_web` still 2 hits in group_vars; `gitborg_ephemeral` still 7 in controller.py — both checked explicitly - dual-brand selectors intact: 46 dashboards, 35 alert-rules - dashboard JSON valid; `ansible-lint` 0/0 across 194 files at `production`; pinned `shellcheck` clean ## Expected transient on apply `{__name__=~"[bg]itborg_x"}` matches two series while retention holds the old one. They don't overlap in time **except within one 5-minute staleness lookback** at the switchover, so `sum(...)` panels can briefly read double. No alert aggregates over these metrics, so alerting is unaffected. `for:` clocks reset and `increase()` under-reports for one window — the documented discontinuity you chose.

supernaut lade till 1 incheckning

2026-08-10 18:25:33 +00:00

feat(monitoring): rename the gitborg_* metric prefix to bitborg_* in every producer

ci / ci (pull_request) Successful in 1m39s

Detaljer

b5961beed7

ADR 0039 §7b. Stacked on §7a — the dual-brand consumers must be live
first, so this must merge second.

166 renames across 23 files. Producers now emit `bitborg_*`; consumers
already accept either name, so there is no window where a panel is blank
or an alert selects something nothing emits.

## An allowlist, not a prefix sweep

A blanket `gitborg_` -> `bitborg_` would have renamed two values that are
not metrics and are load-bearing:

- `gitborg_web` — the live Postgres database AND role, explicitly
  LEGACY-PINned by the §5 decision. Renaming the declaration creates a
  fresh empty database and the portal comes up blank.
- `gitborg_ephemeral` — `controller.py`'s `METADATA_TAG`, the OpenStack
  metadata key that decides which VMs the runner-controller may DELETE.
  Renamed in code but not on live VMs, the controller stops recognising
  its own fleet.

Also excluded: `gitborg_domain` and `gitborg_network_id` (variable and
tofu-output names) and `gitborg_reconciler_image*` (an Ansible var
prefix). 59 names/prefixes renamed, 5 excluded, each verified by hand.

`controller.py` constructs 23 metric names from
`_METRIC_PREFIX = "gitborg_runner_controller"`, so enumerating full names
in the repo undercounts. That single assignment moved and carries all 23.

## Reconciler metrics deliberately NOT renamed in prose

Seven `gitborg_reconciler_*` metrics are the only ones this repo does not
produce — they come from `bitborg-auth-reconciler/src/metrics.ts`, a
separate repo shipped as a container image, so moving them needs a code
change, a release, an image-tag bump and an apply (§7d).

Their `expr`s are already dual-brand and match either name. Their
descriptions are left saying `gitborg_reconciler_*` because that is what
production emits today, with a comment explaining why, so nobody tidies
them into naming a metric that does not exist — an operator grepping for
a nonexistent metric name mid-incident is exactly the misdirection this
plan keeps warning about.

## Verified

- `vmalert:v1.147.0 -dryRun` PASS, 46 alerts in a set identical to
  pre-change, same `for:`/`severity:` counts.
- `test_controller_logic.py` (the suite CI runs, and it asserts on
  rendered metric output): **69/69 checks passed**.
- `gitborg_web` still 2 hits in group_vars, `gitborg_ephemeral` still 7
  in controller.py — both untouched, checked explicitly.
- Dual-brand selectors intact: 46 in dashboards, 35 in alert-rules.
- Dashboard JSON valid, `ansible-lint` 0/0 across 194 files at profile
  `production`, pinned `shellcheck` clean.

## Expected transient on apply

`{__name__=~"[bg]itborg_x"}` matches two series while retention holds the
old one. They do not overlap in time except within one 5-minute staleness
lookback at the switchover, so `sum(...)` panels can briefly read double.
No alert rule aggregates over these metrics, so alerting is unaffected.
`for:` clocks reset and `increase()` under-reports for one window — the
documented discontinuity.

supernaut ändrade målgrenen från feat/rename-7a-dualbrand-consumers till main

2026-08-10 18:29:06 +00:00

supernaut tvångsskickade feat/rename-7b-metric-prefix från b5961beed7

ci / ci (pull_request) Successful in 1m39s

Detaljer

till e14f63a240

ci / ci (pull_request) Successful in 1m44s

Detaljer

2026-08-10 18:29:14 +00:00

Jämför

supernaut sammanfogade incheckning 80d1700ebd till main

2026-08-10 18:31:57 +00:00