Skip to content

expfmt: reject nameless samples after metric metadata - #988

Merged
roidelapluie merged 1 commit into
prometheus:mainfrom
roidelapluie:roidelapluie/reject_nameless_samples
Sep 16, 2026
Merged

roidelapluie merged 1 commit into
prometheus:mainfrom
roidelapluie:roidelapluie/reject_nameless_samples

Conversation

@roidelapluie

Copy link
Copy Markdown
Member

No description provided.

Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
@roidelapluie
roidelapluie merged commit 06f751e into prometheus:main Sep 16, 2026
8 checks passed
nadaverell added a commit to skyhook-io/radar that referenced this pull request Oct 7, 2026
One tested batch replacing this week's soaked, low-risk Dependabot PRs.
Dependency-only: no application source changes.

**Screening time:** 2026-10-07T08:38:20Z (reconciled again at 09:00Z;
rebased onto `1f05aee9`, which picked up #2015)
**72h cutoff:** published at or before 2026-10-04T08:38:20Z. Every
resolved version below clears it. The npm lockfile was regenerated with
`npm install --package-lock-only --before=2026-10-04T08:38:20Z`, so the
resolver could not float past the cutoff.

## Included

| PR | Dependency | Old → New | Published (UTC) | Soak | Risk conclusion
|
|---|---|---|---|---|---|
| #2000 | modernc.org/sqlite (+ modernc.org/libc) | v1.59.0 → v1.60.1
(libc v1.75.7 → v1.77.1) | sqlite 2026-09-29 08:41; libc 2026-09-21
23:07 | ~8d / ~15.4d | Low–moderate. SQLite stays 3.53.4. The libc range
is substantial on Linux (see notes), so I ran the sqlite-backed stores'
tests in a Linux container. |
| #1998 | github.com/prometheus/common | v0.71.0 → v0.72.0 | 2026-09-28
08:33 | ~9d | Low. The only change on Radar's path is a stricter text
parser (see notes). Its `go 1.26` requirement matches ours. |
| #2001 | github.com/google/go-containerregistry | v0.22.0 → v0.22.1 |
2026-09-04 00:07 | ~33d | Low. SSRF-guard hardening, an authn fix that
stops empty credentials overwriting `AuthConfig.Auth`, and sha512 digest
support. Radar uses `authn`, `name`, `remote`, `v1/google` for image
inspection. |
| #2004 | react-virtuoso | 4.18.13 → 4.18.16 | 2026-09-29 16:03 | ~7.7d
| Low. 4.18.14 adds custom `ul` List wrappers. 4.18.15 fixes
`initialItemCount` clamping, which Radar doesn't use. 4.18.16 stops
TableVirtuoso leaking `skipAnimationFrameInResizeObserver` onto the DOM;
Radar doesn't pass that prop. I smoke-tested the binary in a browser
anyway. |
| #2003 | typescript-eslint (+ 10 `@typescript-eslint/*`) | 8.70.0 →
**8.71.0** | 2026-09-28 17:12 | ~8.6d | Low. Adds a new rule that's
opt-in and not enabled here, plus rule fixes. The set of 444 lint
warnings is identical before and after (0 errors). **8.71.1 (2026-10-05)
is too fresh and was deliberately not picked up.** |
| #2002 | vitest (+ @vitest/mocker, @vitest/spy) | 5.0.2 → 5.0.3 |
2026-09-30 11:30 | ~6.9d | Low. Bug fixes in repeats/retry, cache
revalidation, jsdom Blob and the pool. Both test suites pass. |
| #1999 | prettier | 3.9.8 → 3.9.9 | 2026-09-23 06:31 | ~14d |
Negligible. A Markdown `$`-as-math fix; Radar formats only `.ts/.tsx`.
`prettier --check` reports the same 1089 files before and after. |
| #2015 | source-map-js | 1.2.1 → 1.2.2 | 2026-09-30 14:08 | ~6.8d |
Low. Build/test-time only (postcss, Tailwind, jsdom via css-tree). Fixes
a DoS from malicious indexed source maps (CVE-2026-93749) and a crash
under a no-`unsafe-eval` CSP. |

### Movement beyond the Dependabot PRs (reviewed, all soaked)
- **Go:** `modernc.org/libc` v1.75.7 → v1.77.1 (2026-09-21), the version
sqlite pins.
- Three `go.sum` entries also changed, `golang.org/x/tools` v0.50.0
(2026-09-08), `modernc.org/cc/v4` v4.29.7 (2026-09-12) and
`modernc.org/ccgo/v4` v4.36.1 (2026-09-21). `go mod why` shows each one
is reached only through `go-sdk/mcp.test` or `libc.test`, never a Radar
binary.
  - `pkg/go.mod` is untouched.
- **npm:**
- `ignore` 7.0.9 → 7.0.12 (2026-10-02, ~4.8d), nested under
`@typescript-eslint/eslint-plugin`. These are gitignore-semantics fixes
plus linear-time perf work (no release notes, so I reviewed the
[compare](kaelzhang/node-ignore@7.0.9...7.0.12)).
It's lint-time only, and the lint output is identical.
- `why-is-node-running` 3.2.2 → 3.2.1, a **downgrade** that vitest 5.0.3
pins on purpose
([vitest#11403](vitest-dev/vitest#11403),
avoids `ERR_PNPM_TRUST_DOWNGRADE`). 3.2.1 was published 2024-10-29.
  - Nothing else in `package-lock.json` changed.
- **Too-fresh versions deliberately excluded by the `--before` cutoff:**
- typescript-eslint / `@typescript-eslint/*` 8.71.1 (2026-10-05 17:08Z).
- `magic-string` 1.4.3 (2026-10-05 04:52Z), which a plain refresh would
have pulled under `@vitest/mocker`. It stays at 1.4.2, inside mocker's
`^1.2.3`.

## Held / excluded (these PRs stay open)

| PR | Update | Decision | Reason |
|---|---|---|---|
| #1776 | modelcontextprotocol/go-sdk 1.6.1 → 1.8.0 | **Exclude** | I
re-tested on today's main and it still fails
`TestInvestigationHandlerAnnotatesRealToolCallWithoutChangingPublicContract`
("private mount handshake did not mark the scope connected"). Radar's
private-mount handling needs adapting first; that's a source change. |
| #1930 | monaco-editor 0.55.1 → 0.57.0 | **Exclude** | Unchanged from
last week. The deep side-effect imports in
`packages/k8s-ui/src/components/ui/monacoRuntime.ts` don't resolve under
0.57 (TS2882), and that file hasn't changed since. The PR's CI still
fails the Frontend, k8s-ui and Settings jobs. Needs a source change. |
| #1939 | helmfile/helmfile-action v2.2.0 → v2.4.8 | **Exclude** |
Unchanged from last week. Since v2.3.0 the action installs helm-unittest
twice ([#648](helmfile/helmfile-action#648)), so
the Helm chart job fails. Needs a `ci.yml` change to how `helm-plugins`
is specified. |

## Radar usage and risk notes
- **modernc sqlite 1.60 / libc 1.77.1** ([sqlite
CHANGELOG](https://gitlab.com/cznic/sqlite/-/blob/v1.60.1/CHANGELOG.md),
[libc
compare](https://gitlab.com/cznic/libc/-/compare/v1.75.7...v1.77.1)):
- Radar imports only the `database/sql` driver: the SQLite timeline
store and `ai-runs.db`. It uses no `vfs`, no `pcache`, none of the
removed `lib` constants and no `_pragma`. `StrictPragmas` is opt-in.
- The 1.60.0 change that turns a WAL `-shm` read fault into
`SQLITE_IOERR_IN_PAGE` instead of a crash is a strict improvement.
- The libc range includes musl atomics translated to Go, stdio locking
fixes, and setenv/putenv mirrored into the Go environment on musl.
That's why I ran the targeted Linux run below.
- **Last week's blocker is resolved.** libc v1.77.0's recursive `Xnanf`
stack overflow
([cznic/libc#60](https://gitlab.com/cznic/libc/-/issues/60)) caused us
to hold sqlite 1.60. v1.77.0 is now retracted, and v1.77.1 adds the fix
and a `TestNaN` regression test.
- The issue is still open upstream, so I checked directly: in v1.77.1,
Linux `Xnanf` returns `X__builtin_nanf` (`math.NaN()`), and the issue's
own reproducer prints `NaN NaN NaN` on linux/arm64 with no stack
overflow.
- **prometheus/common 0.72**
([release](https://github.com/prometheus/common/releases/tag/v0.72.0)):
- Radar uses `expfmt.NewTextParser(model.LegacyValidation)` to parse
apiserver metrics in `internal/upgrade/collectors_live.go`, and `model`
in the MCP Prometheus tool.
- The only text-parser change
([#988](prometheus/common#988)) now rejects a
nameless `{}` sample that follows metric metadata. Kubernetes apiserver
exposition never emits one.
- The rest is OpenMetrics 2.0 encoding, which is experimental and unused
here, plus format-constant docs
([#992](prometheus/common#992), backward
compatible).
- **typescript-eslint 8.71.0**
([release](https://github.com/typescript-eslint/typescript-eslint/releases/tag/v8.71.0)),
**vitest 5.0.3**
([release](https://github.com/vitest-dev/vitest/releases/tag/v5.0.3)),
**prettier 3.9.9**
([changelog](https://github.com/prettier/prettier/blob/3.9.9/CHANGELOG.md#399)),
**react-virtuoso**
([releases](https://github.com/petyosi/react-virtuoso/releases)),
**source-map-js 1.2.2**
([compare](7rulnik/source-map-js@v1.2.1...v1.2.2)),
**go-containerregistry 0.22.1**
([release](https://github.com/google/go-containerregistry/releases/tag/v0.22.1)).

## Verification
All results are on the rebased branch (base `1f05aee9`).
- **Integrity:** `npm ci` passes, and `go mod verify` passes for root
and `pkg`.
- **Type-check:** `make tsc` passes.
- **Go tests:** `make test` (root `go test ./...`, 43 packages ok) and
`cd pkg && go test ./...` both pass. The first run failed with "package
unsafe is not in std" and missing `go-build` cache files because the Go
build cache was cleaned during the run; both pass on a clean re-run.
- **Frontend tests:**
  - `packages/k8s-ui` `npm test`: 218 files, 4160 passed, 1 skipped.
  - `web` `npm run test`: 158 files, 1865 passed.
- **Lint:** `cd web && npm run lint` gives 0 errors and 444 warnings,
the same warning set as main. I compared the normalized warning lists
from a clean `npm ci` of `origin/main`.
- **Prettier:** `prettier --check` on `web/src` and
`packages/k8s-ui/src` flags 1089 files before and after, so the bump
changes no formatting.
- **Build:** `make build` passes.
- **Linux targeted test:** `go test ./internal/timeline/ ./internal/ai/`
passes in `golang:1.26` (go1.26.8, linux/arm64) against libc v1.77.1.
Those are the SQLite timeline store and the `ai-runs.db` store.
- **Binary smoke test:** I ran the built binary against a live cluster
with the Pods table at ~100 pods.
- TableVirtuoso virtualized as expected: 5182px of scroll height
rendered 29 rows, then 43 after a mid-scroll.
- Rows rendered correctly mid-list, and the console showed 0 errors and
0 warnings.
- **visual-test:** I didn't run a full `/visual-test`. The only
rendering-relevant bump is react-virtuoso, and the targeted smoke test
above covers it.

Supersedes #2000, #1998, #2001, #2004, #2003, #2002, #1999, #2015

<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Low Risk**
> Dependency-only lockfile and manifest updates with no runtime code
changes; bumps are patch/minor and were screened with tests per the PR
description.
> 
> **Overview**
> This PR **batches soaked Dependabot dependency bumps** with **no
application source changes**—only `go.mod`/`go.sum`, root
`package-lock.json`, and version pins in `packages/k8s-ui/package.json`
and `web/package.json`.
> 
> **Go** updates include `modernc.org/sqlite` (v1.59.0 → v1.60.1) with
its `modernc.org/libc` transitive bump, `github.com/prometheus/common`
(v0.71.0 → v0.72.0), and `github.com/google/go-containerregistry`
(v0.22.0 → v0.22.1), plus related `go.sum` entries for sqlite/libc
toolchain deps.
> 
> **npm** updates cover `react-virtuoso` (4.18.13 → 4.18.16) in k8s-ui
and web, `typescript-eslint` / `@typescript-eslint/*` (8.70.0 → 8.71.0),
`vitest` (5.0.2 → 5.0.3) in k8s-ui, `prettier` (3.9.8 → 3.9.9), and
lockfile-only moves such as `source-map-js` 1.2.2 and vitest’s pinned
`why-is-node-running` downgrade.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
3716b5a. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants