## Problem
RUM-to-traces correlation stopped working after the browser SDK 0.4.x
line. The SDK's identifiers became plain `BigInt`, so the unchanged
`toString(16)` serialization silently started dropping leading zeros.
Trace ids are UUIDv7-shaped (48-bit ms timestamp in the top bits), so
the leading hex nibble is always zero today — effectively every RUM
event stored a 31-char `_oo_trace_id`/`_o2_trace_id` while the traces
stream stores the padded 32-char `trace_id` (the `traceparent` header
was padded correctly, so backend traces have the full id). Exact-match
correlation could never hit, in either direction.
Verified against live data: for sampled traced requests, `'0' +
stored_id` found the backend trace every time; the stored id never did.
## Fix (frontend compatibility layer)
Works for the already-ingested short ids **and** for correct ids from
the fixed SDK:
- `utils/rum/fields.ts`: `normalizeTraceId` (pads to canonical 32-char
hex), `traceIdLookupVariants` (padded + legacy zero-stripped forms),
`rumFieldEqualsAnySql` (matches both namespace spellings x both id
forms), and a shared `RUM_CORRELATION_TRACES_STREAM` constant replacing
four scattered "default" literals.
- `useRumSpanBuilder`: trace→RUM lookup queries both id forms; canonical
id stamped on generated RUM spans.
- `useTraceCorrelation`: backend-span query now targets the real
`default` traces stream (was a nonexistent `_traces` stream) as
`page_type: "traces"` (was `"logs"`), with the padded id.
- `EventDetailDrawerContent`: trace navigation pads the id before
building the route.
- `PlayerTracesTab`: the session Traces tab canonicalizes ids before the
traces-stream metadata join (rows were silently dropped).
- `ErrorViewer`: error correlation card receives the padded id;
implausible (non-hex) ids hide the card instead of showing a misleading
empty state.
- `package.json`: browser SDK bumped to 0.4.2-beta.3, which serializes
trace/span ids as fixed-width canonical hex at the source (resolved
`browser-rum-core` verified at 0.4.2-beta.3).
## Testing
- New unit coverage: `fields.spec.ts` (16 tests) plus regression tests
in each touched spec (legacy 31-char id fixtures asserting padded
32-char behavior); full RUM/traces suites green (53 files, 1971 passed).
- Live-verified against real data: a trace from the broken window
renders the full RUM hierarchy (Session→View→Action→Resource + backend
spans) where the old code showed backend spans only; pre-regression
sessions still correlate (no behavior change for correct ids).
- `npm run lint` and `npm run type-check` clean.
## Problem
Every frontend build is red on `npm ci` in `web/`:
```
npm error `npm ci` can only install packages when your package.json and
package-lock.json ... are in sync.
npm error Missing: @noble/hashes@2.3.0 from lock file
```
## Root cause (not a feature change)
- `web/package-lock.json` was last regenerated with **npm 11**, whose
resolver **omits** the explicit `@noble/hashes@2.3.0` entry that
`jsdom`'s transitive `@exodus/bytes@1.15.1` **peer-requires** (`^1.8.0
|| ^2.0.0`).
- The **Playwright build pins `node-version: 22` → npm 10**
(`.github/workflows/playwright.yml`). npm 10's `npm ci` recomputes the
ideal tree, needs that entry, finds it missing, and hard-fails.
- Reproduced locally: `npm@10 ci` **fails**, `npm@11 ci` **passes**, on
the exact committed lock. So it's a **toolchain mismatch**, independent
of any PR — it breaks `main` and every open PR building the frontend.
## Fix
Regenerated the lock with `npm@10 install --package-lock-only` → adds
the nested `node_modules/jsdom/node_modules/@noble/hashes@2.3.0` entry.
**+15 lines, lockfile only**, `lockfileVersion` unchanged (3).
Verified `npm ci` now passes under **both npm 10 and npm 11**.
## Note
The durable fix is aligning the Playwright build to **Node 24 / npm 11**
(matching the AI-review workflow and the Node 20/22 runner deprecation).
This PR is the minimal unblock; the Node bump can follow separately.
## What
Deletes `web/scripts/design-debt-baseline.json` — **not by relaxing the
rules, by fixing the debt**. Every design-token category is now
zero-tolerance, SCSS is gone from the app, and 144 hand-written shadow
values collapse into one scale.
## The guard was under-reporting (three separate holes)
1. **Comment contamination.** `check-design-consistency.mjs` fed comment
text into its exemption checks, so a `keep()` comment that merely
*named* `:deep(` / `::after` / `@keyframes` silenced the rule after it —
and a `;` or `{` inside a comment corrupted the offsets, so stripping
comments afterwards couldn't help. Comments are now blanked
(offset-preserving) before the scan. **Uncovered 9 occurrences in 6
files.** Real debt was 110, not 101.
2. **`arbShadow` only scanned the template.** `box-shadow:` in a
`<style>` block and `boxShadow:` in a JS/`:style` object were invisible
— 40 literals sat there while the guard reported zero. It now scans all
three syntaxes across `.vue` and `.ts`.
3. **`rawVarInTemplate` missed the fallback form.**
`border-(--color-x,rgba(…))` and `bg-[var(--color-x,#fff)]` slipped
through. The token is always defined, so the fallback can never fire —
it only hid the read.
## stylelint had never run on 14 files
`postcss-scss` isn't installed, so `postcss-html` **silently skipped**
every `<style lang="scss">` block. The CI step named *"Lint styles
(stylelint — bans legacy --o2-* tokens)"* was a no-op for those files —
including `color-no-hex` and the `.body--dark` selector ban. All 14 are
plain CSS now and `vue/block-lang` errors on `scss`/`sass`/`less`, so it
cannot silently recur.
## What moved
| | before | after |
|---|---|---|
| `rawVarInComponent` (in `<style>`) | 110 | **0** |
| template `x-[var(--color-*)]`, utility exists | 225 | **0** |
| template reads, no utility | 27 | **0** |
| `style` / `:style` colour literals | 29 | **0** |
| `color-mix(token N%, transparent)` | 160 | **0** |
| hand-written shadows | 233 (144 distinct) | **0** |
| SFCs using SCSS | 14 | **0** |
Guard categories: 22 → 24. `rawVarInTemplate` and `arbShadow` are new;
`arbPx` was deleted because main's `local/no-hardcoded-px` ESLint rule
(#13541) now owns px across every file type.
## Shadows are a system now
Stripped of colour, the 144 values were the same handful of geometries
rewritten in px vs rem, with and without an explicit `0` spread.
```
--shadow-xs / sm / md / lg elevation
--shadow-sticky-left / -right directional (sticky columns)
--shadow-rail inset status rail
--shadow-glow-sm / -md / -lg coloured, 3 intensities
--shadow-scroll-top / -bottom / -both scroll affordance
--shadow-ring-hairline 1px ring where a utility can't reach
--shadow-ai-focus AI composer focus glow (flips in dark)
```
Colour is a **separate axis**. Tailwind rewrites a `--shadow-*` token as
`<geometry> var(--tw-shadow-color, <default>)` — verified by compiling a
probe, and it holds for multi-layer tokens — so a template writes
`shadow-md shadow-ai-accent/35`. A plain stylesheet can't add a utility,
so the role and glow tokens read `--glow-color`; that keeps the pipeline
canvas colour-coding nodes by type from CSS and lets the status rail be
painted per row from JS.
Also reclassified: `0 0 0 2px <colour>` is a **ring**, not a shadow, and
`0 1px 0 0 <colour>` is a **border**.
## color-mix, analysed by intent
All 331 occurrences were classified rather than treated as one problem:
- **160** mixed a token with `transparent` — only an alpha change, so
it's Tailwind's `/N` modifier written the long way (verified: `/N`
compiles to `color-mix(in oklab, <token> N%, transparent)`, and mixing
with `transparent` is premultiplied, so the interpolation space doesn't
matter)
- **81** blended two colours — a derived colour, so it's a token now
(~50 minted)
- **75** were shadow/ring geometry, handled by the scale
- **8 gradients + 1 `calc()`-driven percentage** genuinely have no
utility form and remain; the guard exempts gradient/mask/filter
contexts, matching what the style-block counter already exempts
## Merge with main
Main was running the same migration in parallel (**#13541 "px to rem"**
removed all SCSS, added `no-hardcoded-px`, and built its own shadow
scale), so the merge collided in 68 files / 155 hunks. Resolved by
taking main's source files and re-applying this branch on top, with the
guards driving convergence (429 → 0) rather than hand-resolving hunks.
On shadows, **main's structure wins** — geometry in `semantic.css`
`@theme`, colours as `--shadow-*-c*` sub-tokens, wrapped in its px
disable; its `--shadow-sm/md/lg`, `--shadow-scroll-*` and
`--shadow-glow` keep their definitions. This branch adds what main
didn't have: `xs`, the directional/inset roles, and the
elevation-matched coloured glows.
## New tokens
No new *colours* — each names a value a component already hardcoded,
because a registered palette utility (`text-orange-500`) is banned by
`rawPalette`, so "just register the ramp" was never available.
`--color-splitter-hover`, `--color-status-flaky` (+`-text`),
`--color-chat-bubble-user-border`, `--color-thread-avatar-bg`,
`--color-thread-accent-strong`, `--color-menu-badge-ring`,
`--color-progress-track-fill`, `--color-ingest-accent-*` (5), ~50
derived blend tokens, plus `--color-border-strong` /
`--color-border-subtle` / `--color-warning-300` registered for utility
use.
## Visual deltas to review
Everything else is value-preserving. These are not:
- **4 hover accents** move from a fixed `primary-600` to `accent` — 600
→ 400 **in dark only** (2 chevrons, an empty-state icon fill, one link)
- **sticky table header** in `TraceDetailsSidebar` was a fixed light
grey that never flipped; now `border-default`, so it themes
- **shadow canonicalization** snaps near-identical geometries together
and moves ~40 hardcoded `rgba(0,0,0,x)` onto the scale's token colour
Highest-surface single file: `AlertSettingsHelpDrawer.vue` — a whole
style block deleted with ~44 class names moved into template utilities.
## Verification
`lint:design` · `lint:tokens` · `lint:token-purity` · `lint:styles` ·
`format:check` · `lint:ci` · `vue-tsc` — all clean.
**94 spec files covering every file this branch touches: 4395 passing, 0
failing.**
Not run: the full 1236-file unit suite, Playwright/E2E, and any browser
visual check.
---------
Co-authored-by: ktx-vaidehi <vaidehi.akhani@kiara.tech>
Co-authored-by: Abhay Padamani <141122205+ktx-abhay@users.noreply.github.com>
Co-authored-by: ktx-abhay <abhay.padamani@kiara.tech>
Co-authored-by: ktx-vaidehi <134508096+ktx-vaidehi@users.noreply.github.com>
Co-authored-by: Bhargav <BJP232004@GMAIL.COM>
## Summary
- update vulnerable web transitive dependencies (`axios`,
`brace-expansion`, `fast-uri`, `http-proxy-middleware`, `js-yaml`,
`uuid`, and `websocket-driver`)
- remove the unused `node-polyfill-webpack-plugin`, eliminating the
unpatched `elliptic` dependency chain
- update Playwright to 1.55.1 and `node-forge` to 1.4.0 in the UI test
lockfile
- adapt `BaseImport` to Axios 1.18's stricter response-header type
This is expected to resolve 39 of the 41 currently open Dependabot
alerts.
## Remaining alerts
- `rmcp` remains on the project's custom `rmcp-openapi` fork. Updating
it safely requires a coordinated change to the enterprise
`Cargo.toml.openobserve`, which currently replaces the OSS manifest
during enterprise builds and pins the old API. The compatible
intermediate upstream releases also lack the fork's required
`allOf`/`anyOf` support.
- `thrift` 0.17.0 remains through `parquet` 58.4.0. The patched Thrift
line (0.23+) is outside Parquet 58's version constraint. Parquet 59
removes the Thrift dependency, but adopting it requires a coordinated
Arrow/DataFusion 59 upgrade.
## Validation
- `npm ci` and `npm audit --package-lock-only` in `web/` (0
vulnerabilities)
- `npm ci` and `npm audit --package-lock-only` in `tests/ui-testing/` (0
vulnerabilities)
- `npm run type-check:app`
- `npm run lint:ci -- --quiet`
- `npm run lint:design:strict`
- `NODE_OPTIONS=--max-old-space-size=8192 npm run build`
- `npm run test:unit -- --run` (41,917 passed, 404 skipped)
- `npx vitest run src/components/common/BaseImport.spec.ts --environment
jsdom` (41 passed)
- Playwright test discovery under 1.55.1 (2,076 tests in 265 files)
- `cargo fmt --all -- --check`
- `cargo test -p openobserve-mcp` (94 passed, 1 ignored)
- `cargo audit` (0 vulnerabilities; 2 allowed unmaintained warnings)
Design at autocomplete/autocomplete.md
Rewrites the query editor's autocomplete across every surface that has
one, and
follows the field-value path far enough to fix the bugs it was hiding.
53
commits, in six phases; each phase was written test-first, reviewed
adversarially, and checked in a browser before the next one started.
## What was wrong
The completion providers were wired inconsistently. Six surfaces bound a
resolver that was never defined, so their value completion could not
resolve at
all — SLO, dashboards and anomaly detection among them. Traces never
bound
`:suggestions`. Alerts never called `updateStreamKeywords`. PromQL ran a
second
editor runtime purely for its grammar, and asked
`/prometheus/api/v1/series`
— the heaviest endpoint available — for data the stream schema already
has.
## Phases
**1 — icons, staleness, quoting.** Correct `CompletionItemKind` glyphs,
stream
keywords refreshed when the stream changes, `spath` quoting.
**2 — the function catalog.** Custom functions get one tab stop per
argument,
org VRL functions stop appearing twice, column types read from
`dataType` as
well as `type`.
**3 — signature help, hover, async completion.** Providers return real
`CompletionList`s with `incomplete: true`, so a list that arrives late
is
re-queried rather than pinned.
**4 — numeric ranking.** Numeric columns sort first inside a numeric
aggregate.
**5 — values for streams nobody has searched.** The value cache only
ever held
what a search had already returned, so a stream you had not queried
offered
nothing. Values are now fetched on demand, bounded (10s timeout, 60s
cooldown,
one in-flight request per key), and shared between the editor and the
builder.
PromQL joins the same path: labels from the stream schema, values from
the
shared cache. `/series` now has zero production callers.
**6 — one way to fetch a value.** Item 22 only, in this PR.
## The dashboard filter 400
Clicking the ✕ beside "Select Field" in a filter's dropdown sets the
column to
`{}`, and the watcher on `condition.column` then asks for that column's
values.
`loadFilterItem` built `fields: [row.field]` out of `undefined`,
JSON.stringify
turned that into `null`, and the server answered
```
400 {"code":400,"message":"SerdeJsonError# invalid type: null, expected a string"}
```
`_values_stream` answers 400, not an empty result, for each way a piece
of the
payload can go missing — and every one is reachable: an unresolved
stream alias
(`.find(…)?.stream`), a stream not yet chosen (`""`), a range not yet
set. The
range fails differently: `meta.dateTime` starts as `{start_time: ""}`
and
`""?.toISOString()` does not short-circuit, so it throws and both
callers
surface "Something went wrong!" for a lookup the user never asked to
fail.
Guarded at the two callers, where the payload is built and where the
throw
happens. Not at `fetchFieldValues` — a guard there is too late to stop
the
throw and would duplicate validation the callers need anyway. The
trigger is
left alone: re-picking a field must still load its values.
## Notable removals
CodeMirror is gone — it was still bundled for the PromQL grammar alone.
Bundle 26.3 MB → 25.43 MB; the PromQL chunks 442.7 KB → 20.6 KB. The
PromQL
term catalog went from 7 entries to 113, generated from Prometheus's own
tables
and vendored with its MIT notice.
## Verification
- 76 files, +10.5k/−1.4k, ~9,100 tests passing across the touched areas.
- `vue-tsc` clean, `npm run build` exit 0.
- Every phase checked in a running browser against a live instance,
including
the 400 above: the ✕ now produces no request and no toast, adding a
filter and
re-picking a field each still fetch once, and the value list still
populates.
## Known, deliberately out of scope
- A filter still fetches twice (once on add, once on open) — that is
item 23,
the cache read, not this PR.
- `useMetricsExplorer.loadFilterItem` has the same unguarded
`toISOString()`
pattern on a different surface.
- `services/metrics.ts` and `services/search.ts` still carry dead
`get_promql_series` wrappers now that nothing calls `/series`.
Design at: alerts_2.md
Per-group and per-series alerting, SLO measurement, and the UI for both.
## Alerts
**Multi-alerts (Feature 3).** A grouped alert can opt in to evaluating
and paging **per group** rather than collapsing to one verdict. Opt-in
only (`multi_alert`), so no existing alert changes behaviour: the flag
cannot be present in JSON written before it existed. Per-group state,
transitions, silence fingerprints, disappearance/reaping and a
cardinality cap, with the count-gate rules that make "most severe group"
and the legacy collapsed level provably the same verdict.
**Per-series alerting for PromQL.** The same feature for the metrics
family, where a group is one returned **series** keyed by its full label
set — that is a series' identity in Prometheus, and the expression's own
`by (…)` clause is already where the label set is chosen, so a second
picker could only disagree with it. One additive nullable column; NULL
means off, so stored alerts are untouched.
**Alert detail page**, replacing the row-click side panel: stat strip,
evaluation chart with threshold marklines, per-group table and history.
Simple alerts get a real history too — evaluations (one record per run)
by default, level changes alongside.
## SLOs
Count / time-slice / alert SLIs, error-budget and burn-rate arithmetic,
rolling windows, per-group slices with an exact rollup, coverage gating,
backfill, and SLO-based alerts.
Two capabilities the model could not express, both added tests-first:
- **Metrics-native counting** (`CountSource::PromQl`) — counter metrics
have no rows for a predicate to classify, so "good" exists only as
arithmetic between series, and correct counter arithmetic is
`increase()`.
- **Freshness semantics** (`absent_is_bad`) — for a pipeline SLO,
silence IS the failure, which the default gap rule can never report. A
*failed* query still freezes rather than pages; only a query that PROVED
emptiness counts as bad.
**SLO form**: live good/bad preview so a wrong predicate is caught
before saving rather than days later, real query editors with field
autocomplete, folders shared with alerts, and a move endpoint.
## Notes for review
- The SLO and composite features are **deferred (TODO)**:
`ZO_SLO_ENABLED` stays default **false** and the SLOs menu entry is
hidden until they ship; the SLO backend and UI remain in the tree behind
the flag, and composites stay pure logic + tests.
- One migration (schema 61): a nullable column for the PromQL opt-in. No
backfill.
- Two fixes outside the feature that it surfaced: query-editor
completions were duplicated per mounted editor (Monaco registers
providers per *language*), and the alert detail chart had no PromQL
branch, so it charted row counts against the series-count gate.
- Lands together with the matching o2-enterprise PR — `QueryType::Slo`
is new here and the enterprise dedup matches are exhaustive over it.
- `ratelimit::test_add_batch_missing_rule_id` fails intermittently on
`main` as well; unrelated to this branch.
---------
Co-authored-by: Shrinath Rao <shnath@openobserve.ai>
Co-authored-by: hengfeiyang <hengfei.yang@gmail.com>
Co-authored-by: sai nikhil kethe <nikhil@openobserve.ai>
Applies the **Calm Signal** colour language and summary stat strips
across the monitoring + AI-evaluation pages.
- Colourful summary stat strips (`OStatStrip`) with selectable filter
tiles + left row rails / exception washes on **Alerts, Incidents, Eval
Jobs, Scorers, Score Configs**.
- Tile tones match their `OTag` chip colours; critical-first (left)
ordering; re-click a tile to clear the filter.
- AI Observability: consistent section headers, detail drawer, and
per-session signals.
- Migrate `ScorerTypeDialog` to the standard `ODialog`; token-compliance
+ vertical separators on the eval form pages.
- Table perf: lazy tooltips, delegated row shortcuts, virtualization.
dependabot[bot]
Ashish Kolheandgithub-actions[bot]
Prabhat Sharma
IluckySi
Omkar Kesarkhane
Abhay PadamaniandHengfei Yang
