#### Summary
> You can see the sbshot diff at:
https://pr-docs.viniciusl.com.br/signoz-v3-pr13087-1118
Closes https://github.com/SigNoz/uforge-pod/issues/6284
This PR moves the rest of the app onto the reworked SigNoz UI library
(0.4.0-rc.4) and the new design tokens. Inputs, selects, dialogs,
drawers, popovers, toasts and text now all come from the library, so a
later design change happens in one place instead of in overrides on each
page.
Pages should look and work as they do on main. If something looks
different and is not mentioned below, treat it as a bug.
#### What looks different
Text now uses Inter from the design tokens, so numbers and a few letters
look slightly different, and monospace text uses the system code font.
Truncated text shows its full value on hover. Error toasts stay on
screen until you dismiss them.
A few flows changed on purpose. Saving or renaming a view opens a dialog
instead of a small popover. The Kubernetes entity drawer has a
breadcrumb, so you can go back to the entity you came from.
#### How to review
The commits are grouped by area and named after the pod that owns it,
such as pulse-dashboards or events-traces, so each team can review only
its part. Shared components, the query builder and styles come first,
and the tests sit in the last three commits. Only the final commit
builds, because the library upgrade breaks every page at once.
On your pages, the things most likely to be off are field heights,
spacing, text that now wraps or truncates, and popups opened from inside
a drawer or dialog. Most pages have Storybook stories from #13493, which
this PR is stacked on.
#### What changed in the components
Each block below shows one component, with a before and after taken from
this PR.
<details>
<summary>Typography</summary>
Typography is now a single component. Instead of choosing between Text
and Title, you pick a variant (paragraph, label, uppercase or code), a
size and a weight. Capitals only come from the uppercase variant, and
"muted" is now "secondary". Text cut with an ellipsis shows the full
value in a tooltip.
```diff
- <Typography.Text className="total"> of {total}</Typography.Text>
+ <Typography variant="paragraph" size="base" weight="regular" color="inherit" as="span" textOverflow="wrap">
+ of {total}
+ </Typography>
```
</details>
<details>
<summary>Dialog and AlertDialog</summary>
Dialogs take a single close callback instead of an open-change handler,
and footers are split into a left and a right slot. Confirmations moved
from ConfirmDialog to AlertDialog. Library dialogs now sit above antd
modals.
```diff
- <ConfirmDialog
- onOpenChange={(next): void => {
- if (!next) {
- handleDiscardCancel();
- }
- }}
- confirmColor="danger"
+ <AlertDialog
+ color="danger"
```
</details>
<details>
<summary>Drawer</summary>
Drawers close the same way as dialogs. A drawer has to say whether it
blocks the page (a task) or sits beside it (an inspector), and how it
stacks over other drawers. A drawer can also show a breadcrumb trail,
which the Kubernetes entity drawer uses to step back.
```diff
- <DrawerWrapper
- onOpenChange={(isOpen): void => {
- if (!isOpen) {
- onClose();
- }
- }}
+ <Drawer
+ type="task"
+ stack="overlay"
+ closePosition="start"
+ trail={trail}
+ onTrailSelect={handleTrailSelect}
+ onClose={onClose}
```
</details>
<details>
<summary>Popover</summary>
Popovers take their trigger as a prop instead of being assembled from
parts. The content unmounts when the popover closes.
```diff
- <Popover>
- <PopoverTrigger asChild>{button}</PopoverTrigger>
- <PopoverContent align="end">{content}</PopoverContent>
- </Popover>
+ <Popover nativeButton side="bottom" align="end" trigger={button}>
+ {content}
+ </Popover>
```
</details>
<details>
<summary>Input</summary>
Inputs no longer take a container class name. Wrap them, or use the
width and size props. The test id and data attributes now sit on the
outer frame, and the input element gets the same id with a "-field"
suffix. Number inputs use Input.Number, which hands back a number or
null, so the string parsing at call sites is gone.
```diff
- <Input
- testId="panel-type-search"
- containerClassName={styles.search}
- />
+ <div className={styles.search}>
+ <Input testId="panel-type-search" />
+ </div>
```
</details>
<details>
<summary>Select and Combobox</summary>
SelectSimple and ComboboxSimple are now Select and Combobox. They size
with a width prop instead of a class name, and take a placeholder. The
roles picker moved from antd Select to Combobox.
```diff
- <SelectSimple
- className={styles.filtersBarSource}
+ <Select
+ width={160}
+ placeholder="Filter by source"
```
</details>
<details>
<summary>Toast</summary>
Toasts come from the library toast module. An error toast is now called
danger, and it needs an action because it stays until dismissed.
```diff
- toast.error('Failed to copy link');
+ toast.danger('Failed to copy link', { action: { label: 'Dismiss' } });
```
</details>
<details>
<summary>Callout</summary>
Callouts use color names (primary, danger, warning), take their text as
children and render links with Callout.Link.
```diff
- <Callout type="info" showIcon title="Store the key securely." />
+ <Callout color="primary" size="sm" icon={<SolidInfoCircle />}>
+ Store the key securely.
+ </Callout>
```
</details>
<details>
<summary>AlertStrip</summary>
The no-auth banner moved from our own persisted banner to the library
AlertStrip, which remembers when it was closed.
```diff
- <PersistedAnnouncementBanner type="warning" className={styles.banner}>
- <a href={docsLink} rel="noreferrer">
+ <AlertStrip.CloseablePersisted color="warning" side={bottomSlot ? 'bottom' : 'top'}>
+ <AlertStrip.Link href={docsLink}>
```
</details>
<details>
<summary>Resizable</summary>
Resizable takes its panes as an items array instead of panel and handle
parts, and saves the layout under a storage key.
```diff
- <ResizablePanelGroup id="panel-editor-v2">
- <ResizablePanel minSize="75%" maxSize="80%" defaultSize="80%">
+ <Resizable
+ orientation="horizontal"
+ storageKey="panel-editor-v2"
+ storage={layoutStorage}
+ items={[{ value: 'main', label: 'Preview and editor', minSize: '75%', maxSize: '80%' }]}
```
</details>
<details>
<summary>Breadcrumb</summary>
Breadcrumb takes an items array instead of list, item, link and
separator parts.
```diff
- <Breadcrumb>
- <BreadcrumbList>
- <BreadcrumbItem>
- <BreadcrumbLink icon={<LayoutGrid size={14} />} href={dashboardPageLink}>
+ <Breadcrumb
+ testId="dashboard-page-breadcrumb"
+ items={[
+ { title: 'Dashboard', icon: <LayoutGrid size={14} />, href: dashboardPageLink },
+ { title },
+ ]}
+ />
```
</details>
<details>
<summary>Kbd</summary>
Kbd takes the key name as a value, and named keys render as glyphs
(Escape shows ⎋).
```diff
- <Kbd>Esc</Kbd>
+ <Kbd value="Escape" testId="pie-tooltip-escape-key" />
```
</details>
<details>
<summary>Smaller changes</summary>
Tabs and Toaster no longer accept a class name, so the styles moved to a
wrapper. Avatar needs a name, which it uses for the initials. List
headers show their count in a Badge, with a Skeleton while the count
loads.
</details>
#### Patterns in the code
<details>
<summary>Details</summary>
Most of the diff is text styling. About 2,500 font declarations left the
SCSS files. They were replaced by Typography props or, where a
stylesheet still needs them, by the new text-style tokens described in
the CSS modules guide. The old font size, weight and line height tokens
no longer exist in design tokens 2.7.0, and the legacy card background
is now surface-2.
```diff
- font-size: var(--font-size-sm);
- font-weight: var(--font-weight-normal);
+ font-size: var(--paragraph-base-400-font-size);
```
Tests find inputs through the "-field" id. They pick combobox options by
typing in the search box and pressing Enter, since jsdom cannot click
the new options. The e2e drivers have shared helpers for inputs,
comboboxes and toasts, described in tests/e2e/AGENTS.md.
```diff
- await user.click(await screen.findByLabelText('Roles'));
- await user.click(await screen.findByTitle('signoz-editor'));
+ await user.click(await screen.findByTestId('member-roles-select'));
+ await user.type(
+ await screen.findByTestId('member-roles-select-search'),
+ `${managedRoles[1].name}{Enter}`,
+ );
```
</details>
#### New lint rules
Text style now belongs to Typography, and two rules keep it there. Both
warn for now, because about 2,700 font declarations remain in SCSS.
- signoz/no-text-css (oxlint) flags font and text properties in style
props, CSSProperties objects and styled-components.
- Stylelint flags the same properties in SCSS.
You can disable either rule where Typography cannot render the text,
such as canvas or chart labels. The disable must say why after "--", or
it fails as an error.
<details>
<summary>Example</summary>
```tsx
// oxlint-disable-next-line signoz/no-text-css -- canvas text, not DOM
ctx.font = '12px Inter';
```
```scss
/* stylelint-disable-next-line property-disallowed-list -- antd table header, no Typography slot */
font-weight: 500;
```
</details>
---------
Co-authored-by: Vinícius Lourenço <vinicius@signoz.io>
#### Description
This PR adds the stories we wrote while migrating to the new version of
`@signozhq/ui` (#13087), but running against the components we have on
main today.
I'm splitting them out so the screenshot diff of the migration has a
"before" for every story. Without this, the new stories only exist on
the migration branch and there is nothing to compare them with.
They cover the states the migration touches: open selects, dialogs,
drawers and popovers, more toast variants, the command palette, the
license and trial banners, the K8s drawer opened a few levels deep, SSO
attribute mapping and the query builder add-ons. A few existing plays
also moved into their `*.play.ts` files. No app code changes, only
stories and the mocks they need.
#### Screenshots / Screen Recordings
Every new story, in dark and light:
https://claude.ai/artifact/TeBTXRkB9puh7W6LnGXFQ8
#### Additional Information
#13087 sits on top of this one, so its diff only shows the migration.
---------
Co-authored-by: Manika Malhotra <manika@signoz.io>
#### Description
- `GET /api/v1/service_accounts` returns a page (`serviceAccounts`,
`total`, `reservedKeywords`) with `limit`, `offset`, `sort` and `order`.
- `query` is a SigNoz filter expression over `name`, `email`, `status`,
`created_at` and `updated_at`; a bare token matches name or email.
#### Issues Closed by this PR
Closes: https://github.com/SigNoz/keystone-pod/issues/221
---------
Co-authored-by: hardik-signoz <hardik.saini@signoz.io>
#### Description
- Switching resources under "Viewing · Resource" no longer clears the
filter. Clauses the new resource can filter by are kept and the rest are
dropped, the same rule the drawer already uses, so the quick filter
checkboxes stay in step.
- The drawer's locked filter shows every clause instead of the last one
plus a `+N` count, so the scope reads without hovering. The existing
"Infra monitoring/Drawer overview tab" story already renders this state.
- Switching resources no longer changes what a filter means. A filter
with an OR, or a NOT over a group, is not carried at all, because no
subset of its clauses means the same thing. A clause with a leading NOT
is left out and the rest are carried. Before, `a = 1 OR b = 2` arrived
as `a = 1 AND b = 2`, and `NOT a = 1` arrived as `a = 1`.
- Stacked on #13405, which fixes the clause cutting this relies on.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/535
#### Description
- The Kafka and Celery messaging queue pages load their graphs, tiles
and drawer tables with the v3 and v4 `query_range` APIs. The backend
plans to remove those APIs.
- The pages now load from v5 `query_range` and show the same data. The
Kafka group, topic and partition dropdowns read `/api/v1/fields/values`
instead of the v3 autocomplete.
- Chart clicks did nothing on the Kafka Consumer Lag graph, the Celery
state bars, the worker graphs and the Overview drawer graphs. The new
time series and bar panels dropped the page's own click handler. The
panels call it again when drilldown is off.
- The Drop Rate view is removed. It read the legacy `span/evaluation`
endpoint.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/157
#### Screenshots / Screen Recordings
Main vs this branch, diff view:
https://claude.ai/artifact/BpiYPvdSVXPAD7goQdfVhh
| Screen | On `main` | On this branch |
| --- | --- | --- |
| Kafka overview (3 stories) | Five view cards, including Drop Rate View
| Four view cards. Drop Rate View is gone. |
| Celery task (Default, Failing Tasks, Failed Tasks Only, No Flower
Metrics) | Error% by worker draws a single point on a 0–2000% axis |
Error% by worker draws one line per worker (30–80%) |
| Celery task, State Drawer (new story) | Not reachable. A chart click
does nothing on `main`. | A click on the All state graph opens the
drawer with the slowest tasks for that state |
- Each side is a static Storybook build, shot in dark and light. The
`main` side uses its own v3/v4 story mocks.
- 14 pairs changed and 20 are pixel-identical (Kafka Detail and the
queue Overview).
- The Error% difference comes partly from the mocks: on `main`, the
story's v4 mock answers the formula as a single point.
#### Additional Information
- Part of https://github.com/SigNoz/events-pod/issues/156.
- Stories serve v5. A new `StateDrawer` play opens the Celery drawer
from a chart click. The Kafka detail state is already covered by the
story's `selected` arg.
**FAQ (v4/v3 → v5)**
**Celery traces queries carry `aggregations: [{ expression }]`**
v5 builds a traces aggregation only from the expression. Without it,
`aggregateOperator` is ignored and the query becomes `count()`.
**Celery queries order by `p99(duration_nano)`, not `#SIGNOZ_VALUE`**
`#SIGNOZ_VALUE` was a v4 alias for the value column, and v5 rejects it.
The order key has to name the aggregation expression. This affects the
latency tabs and the drawer tables.
**Grouped Celery queries filter on `<key> exists`**
_Where:_ Messaging Queues › Celery › All state graph and Task Latency
v4 dropped spans that lack the group-by key. v5 returns them as an extra
group with an empty label, which shows as a blank bar on All and an
extra line on Task Latency. API Monitoring got the same fix (`http_url
exists`).
**Kafka metrics graphs send no step**
_Where:_ Messaging Queues › Kafka › Metric view, on ranges over 5 hours
v5 clamps a metrics step to at least range/300, with a 60s floor. A
fixed 60s on a 1-week range shows "requesting aggregation interval 60
seconds…". Sending no step lets v5 pick one. Service details (#13617)
does the same.
**Celery formulas moved to `queryFormulas`**
_Where:_ Messaging Queues › Celery › Error% by worker; Messaging Queues
› Overview › row drawer › Error %
v4 took them inside `builderQueries`. The v5 payload builder reads
formulas only from `queryFormulas`, so in the old place they went out as
metrics builder queries.
**State tiles show bigger numbers**
_Where:_ Messaging Queues › Celery › All Tasks / Failed / Retry /
Successful; Messaging Queues › Overview › row drawer tiles
v4 value panels showed one bucket of a time series (the last minute). v5
scalar requests aggregate the whole window. The numbers match ClickHouse
for the selected range.
**Graphs show the most recent data**
_Where:_ Messaging Queues › Celery › the last bar or point of each graph
v4 traces queries rounded the end down to a whole step, which dropped up
to one step of recent data. v5 keeps the partial bucket.
**Kafka group/topic/partition dropdowns call `/fields/values`**
_Where:_ Messaging Queues › Kafka › Consumer Lag view › group / topic /
partition selects
They used the v3 autocomplete endpoint. `/fields/values` takes the
metric name and filters `searchText` on the server.
**Chart clicks call the page's handler again**
_Where:_ Messaging Queues › Kafka › Consumer Lag graph; Messaging Queues
› Celery › state bars and worker graphs; Messaging Queues › Overview ›
row drawer graphs
When the panels moved to the new time series and bar panels, a page's
own `onClickHandler` stopped being called with drilldown off. The Kafka
lag graph, the Celery state bars and the Overview drawer graphs ignored
clicks. The handler is called again from `usePanelContextMenu`, so API
Monitoring endpoint graphs get their View Traces click back too. A click
outside a series no longer opens a Celery drawer.
**Drop Rate removed**
_Where:_ Messaging Queues › Kafka › view cards and the view selector
Agreed with Pandey to remove it rather than port it. It read the legacy
`span/evaluation` endpoint, which the backend serves only behind the
experimental `kafka_span_eval` flag. The flag is off by default, so the
view returned an error unless `KAFKA_SPAN_EVAL=true` was set. Old
`?mqServiceView=dropRate` links open Consumer Lag.
**Warning icon on empty metric graphs**
_Where:_ Messaging Queues › Kafka › Metric view
v5 reports a metric that was never ingested, and the graph shows that.
Issues found on `main` while testing are filed in
https://github.com/SigNoz/events-pod/issues/280. The ones this PR fixes
are ticked there.
#### Description
- Opening a pod or container from a list filtered by a quick filter
broke the drawer's overview tab. A list filter like `k8s.cluster.name in
['cluster-a']` arrived in the drawer as `k8s.cluster.name in
['cluster-a'`, which does not parse.
- Each carried clause is now cut from its key to the end of its value,
which for `in [...]` is the closing bracket.
- The same cut also dropped the key from a `NOT IN` clause, and pulled
every dropped clause in front of a kept one into the result. Both are
fixed and covered in `relations.test.ts`.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/756
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
A rule's alerts carry the lowercased name of the threshold that fired as
their severity; the rule's own `severity` label is a plain label since
#13612. The v3 list row's new `thresholds` field reports exactly that:
each distinct threshold name, ranked critical, error, warning, info,
then the rest.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/653
---------
Co-authored-by: Ashu Ghildiyal <ashu.ghildiyal@signoz.io>
#### Description
- `ModelMatchesAnyRule` and pattern validation now call the collector's
`pkg/llmpricing` (SigNoz/signoz-otel-collector#973) instead of a copied
regex, so a pattern matches the model exactly or followed by snapshot
suffixes (`-20260921`, `-latest`, `-v2:0`, `@20260921`), and a trailing
`*` matches
any suffix. Priced snapshots no longer show up as unmapped.
- Validation rejects a non-trailing `*`, surrounding whitespace and
control characters; other glob characters are literal in the collector
now, so rejecting them was wrong.
- `PatternEditor` shows the stored pattern instead of appending a
cosmetic `*`, and the help text describes the actual matching.
- Companion to SigNoz/zeus#606 for nerve-pod#517.
#### Issues closed by this PR
Closes https://github.com/SigNoz/nerve-pod/issues/517
#### Screenshots
Before:
<img width="541" height="531" alt="image"
src="https://github.com/user-attachments/assets/2a0cde21-ab05-40c2-be25-2a759ac01426"
/>
After:
<img width="541" height="531" alt="image"
src="https://github.com/user-attachments/assets/a50d8a62-1096-4c89-9871-019d9f6cb26d"
/>
#### Additional Information
- `signoz-otel-collector` is pinned to `v0.144.17-rc.1`, the first tag
with `pkg/llmpricing`; bump to `v0.144.17` once it's released.
- No story change: the `ModelCostDrawer` story already renders the
pattern chips.
- Rules saved with glob syntax keep working as literal patterns,
matching the collector.
---------
Co-authored-by: Vishal Sharma <makeavish786@gmail.com>
#### Description
- Removes `/api/v1/explorer/views` (list/create/get/update/delete) along
with its legacy request/response converters and `v3.SavedView`.
- Folds the `*V2` handlers into plain `Create/Get/Update/List`;
`/api/v2/saved_views` is unchanged (OpenAPI spec has no diff).
- The UI and the signoz-mcp-server already use v2. Nightswatch is still
on older version which uses v1 APIs
Closes https://github.com/SigNoz/signoz/issues/13651
#### Additional Information
Breaking: external API clients still calling `/api/v1/explorer/views`
must switch to `/api/v2/saved_views`.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- adding audit value for source enum and generating openapi spec
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
contributes to https://github.com/signoz/keystone-pod/issues/214
#### Description
- Adds a page story for the Trace Details **Thread** tab, mocking `GET
/api/v1/traces/{id}/thread` with controls for scenario (tool-calling
conversation, every part type, roles and finish reasons, raw attributes
only, no AI spans), Formatted / JSON view, opened-on span and paging.
- Covers loading, next page pending, span not found (404), and a span
dropped for having no messages.
- Adds `SpanMessages` stories for its edge cases: empty state, role-less
output, empty parts, and large JSON behind Show more.
- Moves the existing story to `Trace Details/Overview` so Thread sits
beside it; Thread reuses its mocks for the header and span panel.
#### Issues closed by this PR
https://github.com/orgs/SigNoz/projects/59/views/18?pane=issue&itemId=308646781&issue=SigNoz%7Cnerve-pod%7C646
#### Screenshots / Screen Recordings
https://github.com/user-attachments/assets/69a26f1f-1c20-4b74-872e-0e5e1a5af3c4
#### Additional Information
---------
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
#### Description
Stacked on #13598; merge after it.
- Pages/Noz stories for the UI #13598 changed: `Thinking`,
`StreamingSteps`, `DraftWhileStreaming`, `ScrolledUp`, `SingleStep`,
`PickedContexts`, `SentContexts`.
- The thread fixture gains picked contexts on the opening question, a
one-step answer and a `thinking` agent state; the default thread shows
the first two.
- `AddContext`'s play no longer waits for an idle agent, since the input
stays editable while Noz replies.
#### Issues closed by this PR
https://github.com/orgs/SigNoz/projects/59/views/18?pane=issue&itemId=308646781&issue=SigNoz%7Cnerve-pod%7C646
#### Screenshots / Screen Recordings
#### Additional Information
- `Default`, `AddContext` and `PickedContexts` sometimes open at the
thread's top: AppLayout's deferred `OverlayScrollbars` re-parents the
page after the thread has scrolled to its end, resetting it without a
scroll event. Pre-existing; the `ScrolledUp` play allows for it.
---------
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
#### Description
- The resource attribute filter on the Services list, Service details
and Service Map loads its suggestions from the v3 autocomplete API. The
backend plans to remove that API.
- The filter now reads `/api/v1/fields/keys` and `/api/v1/fields/values`
for the `signoz_calls_total` span metric. The key list, the URL
parameter and the chips are unchanged.
- The v3 client, its types and its story builders are removed.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/151
#### Screenshots
https://github.com/user-attachments/assets/39f1f9bb-6f96-4319-bb29-39678ebbf25ahttps://claude.ai/artifact/Lxma5VmYjdfywyyyv4gY7Z?sk=vynxojQhQ6bzovWbhPFWpQ
#### Additional Information
- Part of https://github.com/SigNoz/events-pod/issues/156.
- Stories serve `/fields/*`. The existing Service Map plays already open
the attribute search, the values, the Environment options and the empty
state, so no new story is needed.
- Tested on a local backend with span metrics from two seeded
environments. The key list matches main exactly, in traces mode and in
metrics mode.
**FAQ (v3 → v5)**
**Value suggestions are not limited to the selected time range**
_Where:_ Services, Service details, Service Map › Environment dropdown
and attribute values
v3 returned values reported since 00:00 UTC today. The v5 call sends no
time range, so it also returns older values that the metadata table
still holds. The QB v5 search bar sends no time range either. Metric
keys are not filtered by time on the backend, so the key list is
unchanged.
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
Stacked on #13597.
- While Noz works, its steps fold into one row naming the running step,
with a shimmer; it opens to the steps as they happen, and once done
reads "Worked through N steps".
- The @ context picker and context pills follow the design; a sent
message shows the contexts picked for it.
- Reply links render as pills, tables scroll sideways in a striped
panel, and text uses the design's type.
- Suggested actions stack as chips; approval and clarification cards
share one layout.
- The loading view mirrors the thread, at the input's width on the full
page.
- Sidebar chat titles use the design's neutral grey (vanilla-400) in
dark mode; light mode keeps `--l2-foreground`, since vanilla-400 doesn't
follow the theme.
- Selected text on the Noz page and in the chat uses the input's accent
highlight, from one shared `_selection.scss`.
- The thread no longer scrolls on its own while Noz replies; only a send
brings the end into view. A "↓ Latest" pill shows when there's more
below and shimmers while the reply streams.
- Only the streaming row re-renders as a reply arrives; earlier messages
are memoized.
- The input stays editable while Noz replies or waits on an approval or
question; only sending is blocked. Snake and thread loading still lock
it.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
<!--If applicable, include screenshots or screen recordings that clearly
show the behavior before the change and the result after the change. -->
#### Screenshots / Screen Recordings
https://github.com/user-attachments/assets/2452ecb8-3de2-4c26-a1f2-9e2258190867
<!--Anything reviewers should keep in mind while reviewing -->
#### Additional Information
- Pages/Noz Default, Streaming, Awaiting Approval, Awaiting
Clarification, Loading Thread and Add Context reach the changed states;
Latest shows on scrolling up in Default.
- The thread renders 1500px above the view (`increaseViewportBy`) so
Virtuoso measures rows before they scroll in; without it, scrolling up
jumps as it corrects estimated heights.
- Departs from Figma: "Worked through N steps" instead of "Worked for N
seconds", since step timings aren't persisted; the picker's "Active
Resource" row waits on a defined action.
---------
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
Stacked on #13596.
- A minimize button on the full page reopens the floating panel on the
page Noz was expanded from, or on home after a deep link.
- Full-page replies render as plain text instead of a bubble; errors
keep their callout, and the activity line and reply actions follow the
design (in the panel too).
- The conversation list follows the design: search line, full-width "New
conversation", uppercase group labels, and rows with a status dot and
short time.
- Conversation rows are flat, with a restyled ⋮ menu.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
<!--If applicable, include screenshots or screen recordings that clearly
show the behavior before the change and the result after the change. -->
#### Screenshots / Screen Recordings
https://github.com/user-attachments/assets/e05be601-b7a6-4438-85ff-5e38c51b6520
<!--Anything reviewers should keep in mind while reviewing -->
#### Additional Information
- Covered by the existing Pages/Noz Default and Streaming stories and
Bottom Strip NozPanel; `AIAssistantPage.test` covers minimizing back
after switching chats.
- The dot only marks a chat Noz is still replying in; the design's other
dot colours need a definition.
- The list keeps its date groups and replies keep their timestamp, where
the design differs.
---------
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
Stacked on #13552.
- New-chat screen per the design: "Noz · BETA" intro and suggestion
pills, with fixed-size placeholders while they load.
- An aurora dot grid above it that glows under the cursor and turns into
Snake on the Konami code; it follows the light and dark theme. (YES WE
ARE ADDING AN SNAKE GAME 🐍 )
- On first send, the new chat fades out and the first message floats up.
- New user bubble style and an L2 panel surface.
- Full-page layout: a faint pane, a 768px input over its edge and a
672px message column.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
<!--If applicable, include screenshots or screen recordings that clearly
show the behavior before the change and the result after the change. -->
#### Screenshots / Screen Recordings
https://github.com/user-attachments/assets/203121b7-e0d9-41ce-98ad-1ecea72690ea
<!--Anything reviewers should keep in mind while reviewing -->
#### Additional Information
- Covered by the existing Pages/Noz NewConversation and Default stories
and Bottom Strip NozPanel; Snake and the send transition are transient,
so no story holds them.
- The "New chat" header and Skills button are follow-ups.
---------
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
#### Description
- An alert with re-notify off was notified again every 120 hours while
it stayed firing.
- Re-notify off is a repeat interval of 8760h, and alertmanager keeps
the notification log entry for 2x the repeat interval, capped by the
nflog retention. The retention was 120h, so the entry was dropped and
the alert looked new.
- The default nflog retention is now 2x the re-notify off interval,
derived from a shared `NoRenotifyInterval` constant so the two cannot
drift.
- Rules with a normal re-notify interval are not affected: their entries
still expire at 2x their own interval.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/751
#### Additional Information
- Entries written before the upgrade keep their 120h expiry, so an alert
already firing at upgrade time is notified one more time.
- Entries of re-notify off rules now stay for up to 2 years, one small
entry per alert group and receiver.
- Deployments that set `alertmanager.signoz.nflog.retention` explicitly
keep their value.
#### Description
- Switches every screen from antd's Flex, Row, Col, Space, Layout, Empty
and List to our own versions from #13542.
- Stylesheets that targeted antd's internal classes (like `.ant-row`)
now use classes of their own.
- Adds Storybook stories for the app layout, the changelog modal and the
Services list column filter.
#### Screenshots / Screen Recordings
Before and after, dark and light:
https://claude.ai/artifact/X4c8YXVYe1kgdSpAnT1eVv
- No layout changes. 203 of 250 screenshots are identical.
- 17 differ in colour only: dark mode links and the light onboarding
background now use design system colours.
- The rest are capture noise (timestamps, scroll position) or new
stories.
#### Additional Information
- Merge #13542 first; this PR targets its branch.
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
Stacked on #13549. One commit per change:
- Restyles the chat input to the new design: a card on the full page,
flush with the panel's edges inside the panel. Add Context becomes an
`@` button beside send; send turns primary once there is something to
send. The placeholder reads "Ask Noz…".
- Removes voice input: the mic button, push-to-talk, the speech hook and
its two analytics events.
- The textarea draws its own blinking robin-500 caret, with a short
trail on arrow-key jumps, and a robin-500 selection.
- While Noz replies, a gradient sweeps the input's top edge, timed from
the stream's start so a remount does not restart it.
- In an empty conversation, Noz climbs over the input's top-right corner
and his eye follows the cursor.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
<!--If applicable, include screenshots or screen recordings that clearly
show the behavior before the change and the result after the change. -->
#### Screenshots / Screen Recordings
Visual diff (sbshot, dark + light):
https://claude.ai/artifact/3h4e3kqX7i67DEwN4HSE4qhttps://github.com/user-attachments/assets/352121ec-c7c9-4d5f-a869-7511dfa1995dhttps://github.com/user-attachments/assets/2e6d0c40-4796-4ad6-8763-3cf6cbbadabfhttps://github.com/user-attachments/assets/0e2db3a8-a75b-49a5-adc2-cfc41c670a4f
<!--Anything reviewers should keep in mind while reviewing -->
#### Additional Information
- Skills, queueing, steering and the new context-chip style are
follow-ups.
---------
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
#### Description
- Adds our own versions of seven antd layout components: Flex, Row, Col,
Space, Layout, Empty and List. They lay out exactly like antd's, so we
can move off antd without layout changes.
- Colours and fonts come from our design system tokens, so they switch
with light and dark mode and match the rest of the app.
- Each component has a Storybook page showing antd's version and ours
side by side, with an automatic check that sizes, spacing and positions
match.
- Checked in light and dark: every story lays out the same as antd.
#### Additional Information
- This PR only adds the components. Switching existing screens over to
them comes in a follow-up PR.
- Colours intentionally differ slightly from antd, because they now
follow the design system. Links use the same colour as the design
system's Typography.Link: identical to antd in light mode, slightly
brighter in dark mode.
- A few antd features nobody in the app uses were left out, for example
Layout's sidebar (`Sider`) and List's pagination.
#### Description
- The component migration (#13087) uses sbshot diffs of the stories as
its regression check. Several events-pod states that render
`@signozhq/ui` had no story.
- This PR adds stories for the logs, traces and shared surfaces. The
commits are split by area: logs explorer, Live Tail, Pipelines, trace
details, traces explorer and funnels, query builder, quick filters and
saved views.
- New controls reach states the old mocks could not:
- Logs: "Linked filter" and "Infra resource".
- Trace details: "Analytics".
- Saved views: "Writes" and "View queries".
- Some mocks now match the backend more closely:
- When nothing matches, the logs frequency chart and the table count
show no data.
- Trace details highlights only the failing spans.
- The exception event has a stacktrace.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/279
#### Screenshots / Screen Recordings
Every new story, in dark and light:
https://claude.ai/artifact/BMo2zjbqdwmddK7Hsmw1uZ
`FormatMenuOpen` (logs explorer) and `AddSeverityParserProcessor`
(pipelines) are not in it, because #13493 adds the same two stories and
[its shots](https://claude.ai/artifact/TeBTXRkB9puh7W6LnGXFQ8) cover
them.
#### Additional Information
- Some stories match #13493 on purpose: `FormatMenuOpen`,
`AddSeverityParserProcessor`, and the span drawer and funnel helpers.
The second PR to merge drops its copy.
- The AI thread states need #13493's Tab control and thread mock. Those
states go in after #13493 merges.
- `Tooltips`, `TooltipsInFilterBar` and the Events tab screenshots
change. The highlight mock and the stacktrace cause this.
- Some older Saved Views plays stay in the mocks module to limit
conflicts with #13087. They move to the play file after #13087 merges.
- These paths are dead code and have no story: the OptionsMenu fields,
`LogsPanelTable`, the ResizeTable Date and Time, the waterfall Error and
NoData, and the funnel details error branch.
- test-storybook passes 195 of 195 stories in the touched areas. cc.
@H4ad
#### Description
- With span metrics on, the services list and the home services widget
load with the v4 `query_range` API. The backend plans to remove the v4
APIs.
- On `main`, both show 0.00 for every P99, error rate and ops/s. The
reader looks for `newResult`, which v4 table responses do not carry.
- Both now load from v5 `query_range` and read the v5 table, so they
show real values.
- Span metrics store latency in ms, but three places treated it as ns:
- the home P99 column read 0.00;
- the services list header said "(in ns)";
- the service details latency axis showed µs.
All three now show ms in span-metrics mode. Traces mode does not change.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/148
#### Screenshots / Screen Recordings
<img width="3400" height="1582" alt="pages-home--default--dark"
src="https://github.com/user-attachments/assets/5da53065-b470-4064-8769-5c6fc02e7b60"
/>
<img width="3400" height="1364"
alt="pages-services-detail--span-metrics--dark"
src="https://github.com/user-attachments/assets/3d3bd054-fb74-4973-8c06-b7ecb0f9ff72"
/>
<img width="3400" height="1364"
alt="pages-services-list--span-metrics--dark"
src="https://github.com/user-attachments/assets/aa7f72f2-a41a-4865-afa6-4863bcf2a258"
/>
https://claude.ai/artifact/EKYK1oeQfy3RsCw7fDHhoh
(feat/service-details-v5 vs this branch)
#### Additional Information
- Part of https://github.com/SigNoz/events-pod/issues/156. This PR
stacks on #13617.
- #3196 added span metrics to screens built for traces, which read
`duration_nano` in ns. Those screens kept the ns unit for span metrics,
but `signoz_latency` is stored in ms. Nobody saw it because the list and
home showed 0.00.
- Span-metrics P99 and traces P99 differ for the same service. Span
metrics estimate p99 from histogram buckets for each step and average
the steps. Traces take the exact p99 over the spans. The unit is the
same.
- The services list query now sends no step, so v5 picks it, like the
service details metrics queries.
- A failed refetch now shows one row for the service. Before, the
service showed twice: zeros and the old values.
- Follow-up: an operation name that starts with `$` (for example
`$connect`) breaks that service's row on v5. The shared filter converter
treats the whole list as a dashboard variable. Service details has the
same issue. A fix follows in its own PR.
- These are the same on `main` and stay out of this PR:
- The resource attribute filter has no effect on the span-metrics
services list.
- The home widget's time picker changes only the operations window. The
metric queries use the global time.
- The services list row link drops the time range.
- Stories: the Home story's Services source control and the Services
Detail SpanMetrics story already reach the changed states. Both story
mocks now serve v5 and send P99 in ms.
- I tested it on a local backend with span metrics. The home widget and
the services list show the same values for the same window.
#### Description
- The service details page loads its graphs and the span-metrics key
operations table with the v4 `query_range` API. The backend plans to
remove the v4 APIs.
- The graphs and the span-metrics key operations table now load from v5
`query_range` and show the same data. The queries set their v5
aggregations explicitly, so v5 does not fall back to the default avg,
sum or `count()`.
- On v5, the Apdex graph drew two wrong lines. v5 treats
`signoz_latency.bucket` as a histogram, so B and C keep their `le`
label. The formula cannot join series with different `le` values. B and
C now count the requests at or below the threshold with
`comparisonSpaceAggregationParam`.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/150
#### Screenshots / Screen Recordings
https://claude.ai/artifact/KeUB5YZafsDGCSf6uPZMME (main vs this branch,
diff view)
Not covered:
- Services list and Home render `ServiceMetricsQuery`, which this branch
touches. That query stays on v4 and sends the same step as before, so it
has no visual change. No shots were taken for them.
#### Additional Information
- Part of https://github.com/SigNoz/events-pod/issues/156.
- The backend team confirmed that the `count` aggregation is meant for
the Apdex query:
https://signoz-team.slack.com/archives/C089D1B5516/p1794900404107019
- In span-metrics mode, the key operations table on `main` shows No
data. For a v4 table panel, `GetMetricQueryRange` returns the response
without `newResult`, and the table reads only `newResult`. The table now
reads the v5 result, so it shows its rows.
- Formula panels show No Data instead of a flat 0 line when the service
has no calls of that kind. This affects DB and external avg duration and
error %. v4 returned a series of zeros for empty inputs.
- The traces-mode key operations table stays on
`/v2/service/top_operations` and `/v2/service/entry_point_operations`,
as the ticket says.
- Metrics queries now send no step, so v5 picks it. This removes the
step warning at ranges of 1 day and more. Traces queries keep the card
step. The services list is still on v4 and keeps its step until
https://github.com/SigNoz/events-pod/issues/148.
- Known: on the metrics cards, View in Logs and View in Traces open a
60s window at ranges over about 5 hours. The shared
`usePanelContextMenu` hook uses 60s when the request has no step. This
already happens on any v5 panel without a step. A fix in the hook
follows in its own PR.
- The calls go through the existing `useGetQueryRange` wrapper, like the
other v5 migrations.
- Stories: the service details story serves v5 `query_range`. A new
Source control and a Span Metrics story show span-metrics mode.
#### Description
- `licensing.GetActive` now reads the org's active license from the
shared cache (`cache.Cache`, 5 minute TTL) instead of querying the
`license` table on every call. Every audited request in the otlphttp,
kafka and file auditors calls it, as do authz and gateway checks.
- "No active license" is cached too, as most self-hosted enterprise
installs run without a license. Store and parse errors are never cached.
- A cached license that has expired is reloaded from the store on read.
- `Activate`, `Delete` and both write paths of `Refresh` clear the org's
entry after a successful write.
#### Issues closed by this PR
ClosesSigNoz/keystone-pod#3
#### Additional Information
- Any new code that writes the `license` table must clear cache entry,
otherwise the change is served for up to 5 minute.
- A request that read the old license before a `Refresh` write can put
it back into the cache after the delete. The TTL bounds this to 1
minute, which is why the TTL is not 0.
#### Description
- An unparseable query on the alert rules list put the server message
under the search box in `--danger-foreground`, which is white: invisible
in light theme. The list body only said "No matching alert rules".
- The list now shows the dashboards list's Invalid query state: the
cleaned-up server message and Contact Support, in place of the list.
- That state moved to `components/ListFilter/InvalidQueryState`, shared
by both lists. `FilterZone` loses its `error` prop, which nothing else
used.
#### Issues closed by this PR
ClosesSigNoz/pulse-pod#745
#### Screenshots / Screen Recordings
[Visual diff](https://claude.ai/artifact/6NUY2jcvjyfzLPEbugYPVt),
merge-base vs branch, dark and light: `Pages/Alerts/Rules/Invalid Query`
changed; `Pages/Dashboards/List/Invalid Query` pixel-identical.
#### Additional Information
Storybook already covers both states (`Invalid Query` on each page);
only the alert story's description changed.
#### Description
- Dashboard settings get a "Default time range" select on the Overview
tab, saved to `spec.duration`. Picking "Default (Last 30 minutes)"
clears it.
- When a viewer opens a dashboard with no time range in the URL and no
saved choice for that dashboard, the URL gets `relativeTime=<duration>`
before the time picker and panels mount. A time in the URL or a saved
choice still wins.
- When the default changes while a viewer is still on the old one, they
move to the new one. A viewer who picked a time keeps it.
- The JSON editor and JSON import reject a `spec.duration` that is not a
whole number of minutes above zero (`1s`, `90s`, `0m`), since the time
picker has no seconds. The editor footer names the line of an invalid
duration or image and shows the full error on hover.
- A value set through the API can still carry seconds. It opens rounded
up to whole minutes: `15s` opens as `1m`, `1h30m` as `90m`. An empty or
zero duration is ignored, and the picker behaves as before: the viewer's
saved choice, else 30 minutes.
- Settings show a stored value that is not one of the options with how
it opens, e.g. "Last 1s (opens as 1m)" or "0m (not applied)".
- The API description of `spec.duration` now says how the UI applies it.
It said "Stored, but not applied by the UI".
- The Prometheus duration parser behind Jira reopen durations moves to
`utils/prometheusDuration.ts`, shared with the dashboard code. Units
must now come largest first and each only once, as the API requires, so
`30m1h` is rejected in the JSON editor, on import, and in the Jira form.
- Carries the frontend part of
[#12861](https://github.com/SigNoz/signoz/pull/12861), moved onto the
`lib/router` facade. Its backend validation is left out: `spec.duration`
is already checked when the spec is decoded.
- The dashboard stories get a "Default time range" control and four new
states: the settings select open, a stored value off the list, the
dashboard opening on its default, and the JSON editor rejecting `1s`.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/695
Closes https://github.com/SigNoz/signoz/issues/12860
#### Screenshots / Screen Recordings
[Visual diff](https://claude.ai/artifact/5mSCBVGEirWfkiC29ZbnEo) against
the merge-base, dark and light: 38 shots, 8 new, 15 changed, 15
identical. `default-time-range-card` is the new card on the settings
Overview tab (12), `new-stories` the four new states (8), and `noise` a
focus ring left on a clicked tab or button in the before shots (3).
#### Description
- The hotkeys handler threw `t.closest is not a function` when a script
dispatched `keydown` on `document` itself. The `Document` node has no
`closest`.
- The handler now ignores events whose target is not an `Element`. Real
key presses always target an element, so shortcuts work the same.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/179
#### Description
- With no alert rules, the empty state sat in a `min-height: 0` flex
column with no overflow, so "Or get started with these sample alerts"
and its cards were clipped below the fold with no way to scroll to them.
- The empty state is now its own scroll container; `margin: auto` keeps
it centered when it fits and top-aligned when it doesn't.
- Only the empty state's stylesheet changes; the rule list is
unaffected.
#### Issues closed by this PR
ClosesSigNoz/pulse-pod#744
#### Screenshots / Screen Recordings
[Visual diff](https://claude.ai/artifact/Y3cH28AULH1vvj62uNebgP) of
`Pages/Alerts/Rules/No Rules`, merge-base vs branch, dark and light: 2
changed, 0 new.
#### Additional Information
Storybook already covers this state with `Pages/Alerts/Rules/No Rules`;
no story edit needed.
#### Description
- AI queries get their own slot (`Query.builder_ai_query`), like
ClickHouse and PromQL.
- Composites from the API map AI queries into it; builder-only checks
accept AI queries.
- The provider and V2 editor edit and render the active slot.
- Nothing changes on the wire: AI still goes out as `builder` with
`builder_ai_query` envelopes.
#### Additional Information
- 2/4, stacked on #13483.
Closes https://github.com/SigNoz/pulse-pod/issues/582
#### Description
- Adds Storybook stories for the top list panel: default, rank and
share, thresholds, the share-of-p99 warning, no data, and the editor's
Appearance section. Their plays check rows, rank and share columns,
threshold colours and the data warning.
- Drilling into an `(empty)` top list row now filters the missing group
as `NOT EXISTS`. Before, the null value was dropped and the explorer
showed every record (Greptile P1 on #13084).
#### Screenshots / Screen Recordings
Visual diff against the merge-base, dark and light:
https://claude.ai/artifact/QqtPhnEJQsksa3reCWV12d
401 pairs identical, 12 new (top list stories), 16 changed (Top List
card in the panel picker), 43 unrelated movers (clock tick, focus rings,
toast fades).
#### Additional Information
- Stacked on #13084.
- The pie and table panels drop null group values the same way; left for
a follow-up.
#### Description
- Panel capabilities and definitions key on `EQueryType` instead of a
separate `QueryMode`.
- No behavior change.
#### Additional Information
- 1/4 of a stack splitting #13222.
## Description
- `/api/v1/ai_observability/fields/keys` returned the `trace.`
aggregates as `float64`, unlike every other numeric key, so
`fieldDataType=number` dropped them.
- Aggregates are now declared `number`, and
`FieldKeySelector.MatchesKey` treats `number`/`int64`/`float64` as one
family. The static-fields matcher in `telemetrymetadata` now reuses it
instead of keeping its
own copy.
- Frontend default AI trace columns switch to `number` so their column
ids still match the keys endpoint. Otherwise the field selector offers
them again as not added.
## Issues closed by this PR
ClosesSigNoz/nerve-pod#639
## Additional Information
Concerns
* If the user have removed one of these columns in select fields, it
will appear again, user has to remove it once more.
* Since saved view is not released so not a problem if we merge this
before that goes out.
#### Description
- Raise `tokenizer.opaque.token.max_per_user` default from 5 to 20.
- MCP clients now authenticate through regular SigNoz sessions, so a
user with a few browsers and MCP clients exceeded 5 and had their oldest
session silently evicted.
- Bump `lastObservedAt` cache expected entries to match (1000 users ×
20); `MaxCost` already fits ~21.8k entries.
Co-authored-by: Vishal Sharma <makeavish786@gmail.com>
#### Description
- The first ruler test of a session can fail: its rule fires before the
new org's alertmanager is registered (the next 5s sync), `PutAlerts`
fails silently and the ruler only resends after 1m, past the test's 30s
wait.
- `create_webhook_notification_channel` now waits for the org
alertmanager with `wait_for_alertmanager_sync`, as the alertmanager
suite already does.
#### Additional Information
- Seen in
https://github.com/SigNoz/signoz/actions/runs/44092243746/job/134070088185
(`test_threshold_above_at_least_once`).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
#### Description
- Adds `signoz/TopListPanel`: ranks the top N groups by one value as
horizontal bars. Spec mirrors `NumberPanelSpec` (visualization,
formatting, comparison thresholds) plus `appearance` (rank, share, bar
color).
- Allowed query kinds: builder, composite, formula, trace operator and
ClickHouse SQL. PromQL is rejected, including inside a composite.
- Frontend panel and regenerated Orval types follow in #13086.
- Artifact:
https://claude.ai/artifact/QqtPhnEJQsksa3reCWV12d?sk=S0XuyygvJdC1QYsz7grKvA
---------
Co-authored-by: Naman Verma <naman.verma@signoz.io>
#### Description
- add a resource def to `PUT /api/v1/service_accounts/me`, taking the
service account id from the caller's claims
- why? so the auditor records a service account renaming itself; it
wrote no audit event before
- a self update and an admin update of the same account record the same
resource id
#### Issues closed by this PR
contributes to https://github.com/SigNoz/keystone-pod/issues/7
#### Description
- OverlayScrollbars animates its scrollbars with `ScrollTimeline`, and
those animations never finish. Storybook waits up to 5s for animations
after every story, so each story with a scroll pane took 5s longer to
capture. `preview-head.html` now removes `ScrollTimeline`, and
OverlayScrollbars falls back to script.
- The `sbshot-visual-diff` skill now says to capture from `pnpm
storybook:build` and to turn off a11y checks during capture. Together
the dev server and the checks added 7 to 12s per story.
#### Additional Information
- CI time does not change: the test runner already skips the animation
wait.
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
A rule label named threshold.name or severity replaced the value the
evaluator stamped. Rule Labels aren't supposed to be treated in a
special way like this.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
Part of https://github.com/SigNoz/pulse-pod/issues/653
#### Description
- Adds `members.spec.ts` TC-04 to TC-07, each running as a member
holding only a custom role:
- TC-04: `list:user` without `create:user`, invite button denied
- TC-05: no `list:user`, list shows "not authorized" and search is
denied
- TC-06: invited member without `factor-password:list`, "Invite Link
Unavailable"
- TC-07: self edit, adds then removes a role (needs `attach` and
`detach` on user and role)
- Drivers: `guestSettings` fixture (settings pages on the signed-in
guest page), `createGrantedUser` seed, members locators, `addRole` and
`removeRole`.
- Unit, story and integration tests already cover the rest of Users FGA,
see the plan on the issue.
#### Issues closed by this PR
ClosesSigNoz/keystone-pod#212
#### Additional Information
- Each spec was checked by granting the missing permission and
confirming it fails.
- Run locally against hosted staging backend with the local frontend
(`--no-deps`). Not run on the pytest bootstrap stack.
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- Adds resource defs to the three `/api/v1/org/preferences` routes so
org preference updates emit audit events.
- Moves them from `AdminAccess` to `CheckResources`, scoped to
`org-preference:list`, `org-preference:read` and
`org-preference:update`. Admin keeps sole access.
- Migration 150 backfills the admin `org-preference` tuples for orgs
created before the grant existed. The registry and stored roles already
carry the grant, so it writes tuples only.
- The preference name is recorded as the resource ID with a wildcard
selector, since names are not UUIDs.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
contributes to https://github.com/SigNoz/keystone-pod/issues/10
contributes to https://github.com/SigNoz/keystone-pod/issues/43
#### Description
- Drop Sentry events whose throwing frame is `<anonymous>` (code loaded
without a script URL, such as extension or automation scripts).
- The existing `hasNoAppFrames` filter missed them: Sentry's bundled
`setTimeout` wrapper counts as an app frame.
- Matching the top frame instead of the error message keeps real bugs
reportable.
#### Issues closed by this PR
ClosesSigNoz/keystone-pod#218
#### Description
- New `fileauditor` suite: starts SigNoz with the file auditor and reads
`/tmp/audit.log` from the container.
- One test per audited verb (create, update, delete, attach, detach),
plus a failed request that writes a failure event.
#### Description
Clears every high finding from `pnpm audit` in the frontend (zod,
follow-redirects, markdown-it, joi) and most of the moderate and low
ones.
- Bumped axios, zod and i18next-http-backend directly.
- Raised or added override floors for fast-uri, joi, markdown-it, moment
and uuid. The uuid floor moves `@storybook/test-runner` from uuid 8 to
11, which still exports the `v4` it imports.
Left alone on purpose: react-router (the fix needs React 19), glob and
minimatch (the fix only exists in new majors that drop APIs jest and
eslint use), and deepmerge (no patched release). The e2e package only
has the minimatch finding, so it is unchanged.
To verify: `pnpm audit` in `frontend` shows 0 high, down from 7.
Closes https://github.com/SigNoz/uforge-pod/issues/6454
#### Description
> TLDR: You can take a look at
https://claude.ai/artifact/LbFG6NcXyeRCm5MSnCKANc
- Upgrades react-router v5 (plus `react-router-dom-v5-compat` 6.30 and
`history@4`) to `react-router` 7.18.4. Follows #13173: pages and
components keep calling `src/lib/router`, so outside the router code the
only changes are render-order and nuqs write-order fixes.
- The app mounts a data router (`createBrowserRouter`) under a pathless
root route that renders `AppProviders`. The base path becomes the
router's `basename`. `lib/history`, `history`, `@types/history` and
`@types/react-router-dom` are removed.
- nuqs moves to its `react-router/v7` adapter, and the `react-router <7`
override goes, which held nuqs's copy below the fix for
GHSA-wrjc-x8rr-h8h6.
- The route table is static. A `RouteGate` renders `NotFound` for routes
the plan, license or flags turn off, which is what the old dynamic
`<Switch>` fell through to. `exact` is dropped (v7 matches exactly) and
the 4 prefix routes become `nested` (`pattern/*`).
- `useNavigationBlocker` moves to v7 `useBlocker` through a new
`useAppBlocker`, which deletes the history@4 replay code from #13173.
What replaced what:
| Before | After |
| --- | --- |
| `<Router history>` + `<CompatRouter>` in `AppShell` |
`createAppRouter` + `RouterProvider` in `index.tsx` |
| `<Switch>` over the enabled routes | static table in
`AppRoutes/routeTable.tsx`, `<Outlet context={enabledRouteKeys}>` |
| `lib/history` singleton behind `navigate()` | `attachRouter`:
`navigate()` drives the data router attached last |
| `blockNavigation` + `retryTransition` | `useAppBlocker` (`proceed` /
`reset`) |
| `nuqs/adapters/react` | `nuqs/adapters/react-router/v7` |
| `MemoryRouter` / history-backed `TestRouter` | browser data router
over jsdom's history, `HostRouter` |
| `lib/history` alias in Storybook | memory data router per story with a
guarded `navigate` |
#### Issues closed by this PR
Closes https://github.com/SigNoz/uforge-pod/issues/2544
#### Additional Information
Reviewing: read by commit. Only the tip builds, since the package swap
cannot land halfway.
- Read closely: `feat(router): back the facade with react-router v7`
(`navigation.ts`, `matchRoute`, `buildRoutePath`, `AppLink` /
`Redirect`), `feat(router): mount the app on a data router`
(`app/AppRouter.tsx`, `AppRoutes/routeTable.tsx`, `Private.tsx`), the
blocker commit and the test router commit (`tests/router.tsx`,
`router-utils.ts`, `lib/router/HostRouter.tsx`).
- Read closely: the 8 `fix(router)` commits. `useShareBuilderUrl` and
`QuerySearchV2Provider` published their query onto the next route while
a route-driven tab was leaving, and the provider then wiped an
expression a child runs on mount (the K8s drawer's Overview tab); two
lazy routes of one module (logs explorer, pipelines) reused the previous
page until the gate was keyed by route entry; `stripBasePath` now
matches case-insensitively, as the router does.
- Three fix nuqs writes that a navigation dropped or overwrote.
`nuqs/adapters/react-router/v7` patches `history.pushState` and
`replaceState`, and any write it did not make clears the writes it still
has queued. On K8s a row needed two clicks to open its drawer (a no-op
`replace` from `useSafeNavigate`), a category switch with a filter kept
the old category, run query kept the old page, and a stored page size
was lost on mount. Alert history rebuilt URLs from the router's stale
location, and the adapter adopted them, dropping the search expression.
Fixed at the call sites; the tests run the real adapter, and
`routing/url-state.spec.ts` TC-05 to TC-08 cover K8s. #13336 has the 2
K8s commits, not the alert history one. The FAQ below has the full nuqs
picture.
- Skim: deps, jest environment, Storybook and lint commits.
Behavior to know:
- Route transitions run with `useTransitions={false}`. Under a
transition React keeps the old screen, so a route whose chunk is not
cached would show no loader.
- The post-login hop to the stashed route still pushes
(`replace={false}`). Back from it now ends on `/home` instead of the
stashed route, because the stash is already cleared; TC-15 asserts it.
- `navigate()` and `getCurrentLocation()` throw when no router is
attached. Tests render first or call `attachTestRouter`.
Testing:
- The 2 K8s nuqs commits came after those runs: `tsgo`, lint and `pnpm
build` clean on the tip, and the 46 related jest suites pass. With the
fixed files reverted, 5 of their 7 new tests fail.
- E2E chromium, run before the rebase onto main, routing specs from
#13078 included: 415 passed, 4 failed. The 4 (alert history expression
filter TC-10 to TC-12, token highlight dark theme) are load flakes:
repeated 3 times at 6 workers, those specs fail 15 of 50 runs on the
#13173 build and 12 of 50 here.
- Bundle, against the #13173 build: the entry chunk grows by 20 KB gzip
(+0.78%). Router libraries in it go from 33 KB minified (v5, v6 compat,
history) to 95 KB (`react-router` 7). CSS unchanged, one lazy chunk
fewer.
FAQ:
<details>
<summary>Why a data router and not declarative mode
(<code><BrowserRouter></code> +
<code><Routes></code>)?</summary>
`<BrowserRouter>` creates its history inside the component and gives
nothing outside React a handle on it. 88 files navigate through
`lib/router/navigation`, some from plain modules (`api/utils.ts`,
`utils/app.ts`). In declarative mode the only way to reach them is
`unstable_HistoryRouter` over a `history@5` object we own, which is what
#13336 does. A data router is an object created before the first render,
so `navigation.ts` calls `router.navigate` on it directly. `useBlocker`
needs one too: in 7.18.4 it throws under `<Routes>` or
`unstable_HistoryRouter`.
</details>
<details>
<summary>Why remove <code>history</code> and
<code>lib/history</code>?</summary>
A data router creates and owns its history; v7 has no prop that accepts
one. Keeping the `lib/history` singleton would leave a second writer to
`window.history` that the router never hears: the router only listens to
`popstate`, so a `history.push` from the singleton moves the address bar
and renders nothing. What the singleton did (navigate from anywhere,
`listen`, `hasInAppHistory`) now lives in `navigation.ts` on top of the
attached router.
</details>
<details>
<summary>Why a new React context in <code>HostRouter</code>?</summary>
It is for tests and stories only; the app does not use it. A data
router's routes, elements included, are fixed when the router is
created. `TestRouter` creates its router once per mount, so a `rerender`
with new children would keep showing the first ones. `HostRouter` puts
the children in a context and the route element (`HostedTreeOutlet`)
reads it, so a re-render shows the new tree.
</details>
<details>
<summary>Why do routes read the enabled set from outlet
context?</summary>
The route table is built once, at module load, before any hook runs, so
it cannot ask the plan, license or flags which routes are on. `App`
still computes that set the way it did and passes it down as `<Outlet
context>`. `RouteGate` reads it and renders the page or `NotFound`,
where the old dynamic `<Switch>` fell through. Flags loading later never
recreates the router.
</details>
<details>
<summary>Why do the providers render inside the router now?</summary>
A data router renders everything under `RouterProvider`, and the only
place for a layout above the pages is the root route's element. The nuqs
adapter calls `useNavigate` and `useSearchParams`, so it has to be
inside a router. `AppProviders` and `App` became that root element.
</details>
<details>
<summary>Why does <code>navigate()</code> keep a stack of routers, and
throw without one?</summary>
Tests and stories mount their own data router. `attachRouter` pushes it
and its detach pops it, so `navigate()` moves whichever router is on
screen and hands back to the one below. With nothing attached there is
no router to hear the navigation, so it throws instead of silently
writing the URL.
</details>
<details>
<summary>Why is <code>RouteGate</code> keyed by route entry?</summary>
`<Outlet>` renders the matched element without a key. The logs explorer
and pipelines routes lazy-load the same module, so React reused the old
instance and showed the previous page. The key remounts the page when
the route entry changes.
</details>
<details>
<summary>Does any page load data through a loader now?</summary>
No. No route has a `loader` or `lazy`, so every page fetches through
react-query as before and navigations commit synchronously. The first
`loader` makes navigations through that route asynchronous:
`getCurrentLocation()` right after `navigate()` would then return the
old location.
</details>
<details>
<summary>Why did query builder files change if pages only call the
facade?</summary>
A route-driven tab keeps the leaving pane mounted until its animation
ends. On v7 that pane saw the new location and published onto it:
`useShareBuilderUrl` wrote the old tab's `compositeQuery`, and
`QuerySearchV2Provider` restored a dropped `expression`. Both now
publish only for their own page, and the URL owns the expression. That
last change wiped an expression a child runs on mount (the K8s drawer's
Overview tab starts from the list filter), so the provider skips its
first run when the URL has none.
</details>
<details>
<summary>Is the nuqs de-sync fixed?</summary>
It is two problems, and only one is fixed.
The router still lags nuqs, as on main. A nuqs write is shallow: it
calls `history.replaceState` or `pushState` itself and does not tell the
router, so `useLocation().search` stays behind the address bar until the
next router navigation. Event handlers read the real URL through
`getUnstableCurrentSearchParams()`. Removing that lag means `shallow:
false` writes, a separate change with known regressions (the role editor
blocker, AI assistant chips, Back behavior, a re-render per write).
The adapter swap is new. `nuqs/adapters/react-router/v7` patches
`pushState` and `replaceState`. A write it did not make, which means
every router navigation, does two things: it aborts the nuqs writes
still queued (they wait up to 50 ms), and it becomes nuqs's state.
Main's `nuqs/adapters/react` does neither, so router writes to a nuqs
param do not reach its hooks there. That gave two kinds of bug:
| Kind | Example | Status |
| --- | --- | --- |
| A navigation drops a queued write | K8s: two clicks to open a drawer,
category switch keeps the old category, run query keeps the old page,
stored page size lost | fixed: navigate before the nuqs setters,
`useSafeNavigate` skips a no-op target, the default query redirect
renders before the list |
| A URL rebuilt from the stale router location is adopted back | alert
history: a state tab, the top contributors drawer or a graph zoom
dropped the search expression | fixed: build from
`getUnstableCurrentSearchParams()` |
| Same, not confirmed | members `?invite=true`, dashboard `variables`
(`useViewPanel`, `usePanelInteractions`), planned downtime, routing
policies | open: needs a contrast test against main's adapter |
Jest tests run the real adapter, and `routing/url-state.spec.ts` TC-05
to TC-08 check the K8s first clicks. For new code: navigate first and
call nuqs setters after, and build a URL from
`getUnstableCurrentSearchParams()`, not from the router location.
</details>
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- Adds resource defs to `GET`/`PUT /api/v2/orgs/me` so org updates emit
audit events.
- Moves both routes from `AdminAccess` to `CheckResources`, scoped to
`organization:read` and `organization:update`. Admin gets both; editor
and viewer stay blocked.
- Migration 149 backfills the admin grants and refreshes managed roles
for existing orgs.
- The org ID comes from the caller's claims, since the update handler
ignores the body `id`.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
contributes to https://github.com/SigNoz/keystone-pod/issues/10
contributes to https://github.com/SigNoz/keystone-pod/issues/43
<!--Anything reviewers should keep in mind while reviewing -->
#### Additional Information
- The organization object has no org prefix, so the managed `*` grant
(`organization:organization/*`) covers every org. Safe here because both
routes take the org from claims. A route that takes an org ID from the
request needs an org-scoped grant first.
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- adding integration tests for audit querier
- adds tests for querying ingested audit logs
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
contributes to https://github.com/SigNoz/keystone-pod/issues/13
#### Description
- The pipeline filter preview read its sample logs and its match count
from the v3 `query_range` API. The backend plans to remove the v3 APIs.
- The sample logs and the match count now come from v5 `query_range`.
The preview sends the filter as a v5 filter expression.
- On `use_json_body` tenants, v5 returns the log body as an object. The
pipelines preview API accepts only a string body, so "Simulate
Processing" failed with a 400. The preview now sends an object body as a
JSON string, as v3 did.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/274
#### Screen recording
https://github.com/user-attachments/assets/229ded0a-fc40-4f79-b593-d9387a43f3c7
#### Additional Information
- Part of https://github.com/SigNoz/events-pod/issues/156. This is the
`query_range` follow-up that https://github.com/SigNoz/signoz/pull/13207
mentioned.
- The preview calls v5 through the existing `useGetQueryRange` wrapper,
like the other v5 migrations.
- reproduced the 400 on a staging tenant with `use_json_body`. With the
fix, the simulation returns the processed logs.
- Stories: the Pipelines story serves v5 `query_range` for the sample
logs and the count. The story no longer serves the v3 suggestion APIs,
because the filter uses the v5 fields APIs since #13207. The simulation
fix renders nothing new, so the story is unchanged for it.
#### Description
- Add `user` and `factor-password` to the role permission schema and
panels, and teach the `generate authz` command to emit both.
- Gate the Members page on `user:list` (list, filter, search) and Invite
member, including the modal's submit, on `user:create`. Attach failures
show as the inline API error after click.
- Gate the member drawer: details on `user:read`, name and save on
`user:update`, delete/revoke on `user:delete`, and the password reset
link on `factor-password:create` + `user:attach`.
- Show the Members settings entry to every user; permissions decide what
it renders.
- Update unit tests, Members stories (new `Authz` folder, one story per
permission) and the e2e settings smoke spec for the new gating.
#### Issues closed by this PR
ClosesSigNoz/keystone-pod#202
#### Screenshots
[Before/after diff, dark and
light](https://claude.ai/artifact/XXY1xvkAMZoR6GS2cMmDdZ?sk=Yz4QIeLw8gEJ2ia5Ti5nbA):
110 pairs, 48 identical, 18 new, 44 changed (30 Roles permission cards,
6 settings nav, 8 sub-0.1% Members noise).
---------
Co-authored-by: vikrantgupta25 <vikrant@signoz.io>
#### Description
- Add a `relatedLinks` template func and `signoz.relatedLinks.slack` /
`signoz.relatedLinks.markdown` / `signoz.relatedLinks.html` blocks so
link selection, labels and formatting live in Go, not in stored channel
templates.
- Go-rendered notifiers (email, Google Chat, Jira, Slack buttons) build
their buttons from the shared `BuildRelatedLinks`.
- FE and Go defaults for Slack/MS Teams text, incident.io/JSM Ops
description and the Telegram message call the blocks instead of inlining
`.Annotations.related_logs/related_traces`.
- Migration 148 rewrites the shipped link snippet in
`alertmanager_config` receivers and `notification_channel` data/spec to
the block call. Sections that differ from the shipped defaults are left
as is and logged.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/587
#### Additional Information
- Labels change to "View Related Logs" / "View Related Traces" on
incident.io, JSM Ops and Telegram; Slack text now renders `<url|View
Related Logs> · <url|View Related Traces>` with no header lines when
links are absent.
- Telegram's FE default intentionally carries no `message` (the backend
fills `DefaultTelegramMessageTemplate`), so only the Go constant changes
for it. If #13119 ever ships a literal `message` default, that string
needs the html block call too.
- incident.io tests switch to `alertmanagernotifytest.CreateTmpl`
because upstream's helper lacks the SigNoz template blocks.
#### Description
- Adds `signoz/GeoMapPanel`: query results on a map as shaded
countries/states, bubbles or lat/lon points, in Mercator, Equal Earth or
Natural Earth.
- `location` names the group-by columns that place a row (country,
region, lat/lon); `map.colorBy` / `map.sizeBy` pick value columns. Both
are checked against the query, and all queries must share a group by,
since rows join by location (as for Scatter).
- Reuses Table formatting (`columnUnits`), Legend and the Heatmap
palette/scale enums. Thresholds are the Number panel's minus `format`: a
match always sets the color.
- Nested objects reject unknown fields; bubble radius bounds are checked
against the layer's defaults when only one is set.
- Allowed query kinds: builder, composite, formula and ClickHouse SQL.
No PromQL, and no trace operators: their input queries have no group by,
which breaks the shared group-by rule (same as Scatter).
- Public dashboards join GeoMap results into one table, as for Table and
Scatter.
#### Additional Information
- Stacked on #13030. Chart layer is #13350; the panel (#13351) carries
the regenerated Orval types.
- Replaces #13138, closed when the stack was reordered to put Geo Map
below Top List.
Closes https://github.com/SigNoz/pulse-pod/issues/694
---------
Co-authored-by: Naman Verma <naman.verma@signoz.io>
#### Description
- Stop redirecting `admin@signoz.cloud` to `/onboarding`. The first-user
check excluded that email from the user count but not from the redirect,
so support was locked out of every page on a new deployment until the
customer finished the questionnaire.
- Move the email to a `SIGNOZ_ADMIN_EMAIL` constant, shared by the user
filter and the redirect check.
#### Issues closed by this PR
ClosesSigNoz/keystone-pod#217
#### Description
- Adding the `safe-to-e2e` label now starts e2e right away. Before,
`e2eci` ignored label events, so the label only took effect on the next
push or reopen.
- Other labels get their own concurrency group and report as
`ignored-label`, so they never cancel a running e2e or hide its `e2e`
result. They still add a short run of skipped jobs to the checks list.
- The merge queue now runs the full test suite and reports a real `e2e`
result, instead of passing without running anything. Full runs take 17
to 27 minutes, inside the 60-minute queue timeout.
- `fmtlint` is skipped in the merge queue. It only checks `tests/e2e/`,
which each PR's own run already covers.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/736
#### Description
- The funnel step filters fetched keys and values from the v3
autocomplete APIs. The backend plans to remove those APIs.
- `QueryBuilderSearchV2` now fetches keys and values from the v5 fields
APIs. The pills and the saved filter shape stay the same, so saved
funnels and pipelines load as before.
- The funnel step Service and Span name dropdowns fetch values from v5
too, scoped to the `resource` and `span` contexts. They share a hook
with the Celery filters, so those move to v5 as well.
- Celery "Worker Online" counts workers from a v5 `query_range` over the
selected time range. It used to count the worker dropdown options, which
had no time range and a cap of 50.
- Pills carry the v5 context as the key type. The pipelines filter
converts it back to the legacy type when the filter changes (`attribute`
→ `tag`, log fields → top-level), so the filter preview, the save and
the stored pipelines match main.
- Span and log fields are marked as columns, so a funnel `EXISTS` on a
field like `name` reads the column.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/146
Closes https://github.com/SigNoz/events-pod/issues/145
#### Screen recordings
Funnels, Exceptions and pipelines on v5.
https://github.com/user-attachments/assets/217a0a85-984b-4f94-bf9e-a5f21696c3a6https://github.com/user-attachments/assets/19ba4648-7466-447e-9cf2-e50700022cf9https://github.com/user-attachments/assets/6dd6e984-d2e6-4a03-b639-a82181618d40
#### Additional Information
- Part of https://github.com/SigNoz/events-pod/issues/156. The same
changes move these filters off v3:
- the exceptions filter, https://github.com/SigNoz/events-pod/issues/145
- the pipelines filter, https://github.com/SigNoz/events-pod/issues/25
- the Celery dropdowns, https://github.com/SigNoz/events-pod/issues/157
- UI change: the key dropdown badge shows the v5 context, `attribute`
instead of `tag` and `span` for span fields. Neither has a badge colour.
Visible on the funnel step Where filter, the exceptions filter and the
pipelines filter.
- The pipelines filter no longer shows example queries. The v5 fields
API has no equivalent.
- These filters stay on `QueryBuilderSearchV2` with v5 lookups. The move
to `QuerySearch` is left to the page revamps.
- Stories: the Funnels, Celery and Exceptions page stories serve the v5
fields APIs, and the Celery story serves the Worker Online query.
- another PR will be raised to move query range to v5 for logs pipeline
#### Description
- Raw lists sorted by timestamp (logs: `timestamp, id`) fetch `offset +
limit` rows per chunk and drop the first `offset` in Go. Every dropped
row was first decoded into a `RawRow` (body + attribute maps), so deep
pages loaded every skipped row into memory and could OOM signoz-0.
- `readAsRawSkipping` now steps past the skipped rows with `rows.Next()`
without `Scan`, so they are never decoded. Memory stays at about one
page plus one ClickHouse block.
- `executeWindowList` passes the remaining offset to each chunk and
counts the skipped rows against `need` and `offsetLeft`.
#### Issues closed by this PR
Fixes: SigNoz/events-pod#270
Part of: SigNoz/events-pod#269
#### Additional Information
- Skipped rows still cross the wire. Pushing `OFFSET` into ClickHouse is
point 1 of the issue.
- `TestReadAsRawSkipping` uses `clickhouse-go-mock` rows (v0.15.1,
bumped in SigNoz/signoz#13580), whose `Next()` advances like the real
driver. Each skipped row is marked with `RowError`, so the test fails if
a skipped row is ever scanned.
- New integration tests `test_traces_list_offset_across_chunks` and
`test_logs_list_offset_across_chunks` page a 4h list with offsets
landing in, and crossing into, each of its three time chunks.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- add resource defs to session routes
- why? - so that auditor can pick up these routes for auditing
- session routes are pre-auth so audit events will be dropped at the
zero org id condition
- rotate and delete session (when session is active) routes are audited
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
contributes to https://github.com/SigNoz/keystone-pod/issues/10
#### Description
- Bumps `github.com/SigNoz/clickhouse-go-mock` v0.14.0 → v0.15.1. Its
`Rows.Next()` now advances like the real driver (skipping a row with
`Next()` and no `Scan()` works), and the new `Rows.RowError` fails
`Scan` of a chosen row (SigNoz/clickhouse-go-mock#14).
- v0.15.1 requires clickhouse-go v2.47.0, so this also raises
**production** clickhouse-go 2.44.0 → 2.47.0 and ch-go 0.71.0 → 0.73.0,
with transitive bumps (zap 1.28.0, otelhttp 0.69.0, gopsutil 4.26.5,
lz4, orb, brotli, purego, go-version, go-sysconf, numcpus, plan9stats).
- Kept apart from SigNoz/signoz#13564 so the driver upgrade can be
reverted on its own.
#### Issues closed by this PR
Part of: SigNoz/events-pod#270
#### Additional Information
- clickhouse-go 2.45–2.47: positional binding now skips `?` inside
string literals, quoted identifiers and comments (`--`, `/* */`, `#`)
(ClickHouse/clickhouse-go#1879, #1888). Our builders emit no SQL
comments and no `?` inside quotes, and values inlined by
`pkg/clickhousesql` that contain `?` were previously bound as
placeholders; now they stay literal. `\?` handling is unchanged. Other
fixes: JSON column serialization version (#1850), DateTime/DateTime64
time zones (#1882), connection eviction on wrapped net errors (#1869).
- ch-go 0.73.0 drops the obsolete nested-exception flag
(ClickHouse/ch-go#1159, marked breaking): `proto.Exception.Nested` is
gone. We only use ch-go's error-code constants, and clickhouse-go's own
`Exception` is unchanged between 2.44.0 and 2.47.0.
- Tested: `go build ./...`, `go test ./...` (all pass), `go vet ./...`
(same 38 findings as `main`, none new). Integration suites
`querierlogs`, `queriertraces` and `querier_json_body` against a server
built with these versions: 624 pass (5 first hit ClickHouse memory
limits in the shared local test container while inserting fixtures, and
passed on rerun).
- Follow-up: #13564 switches `TestReadAsRawSkipping` to `cmock` +
`RowError` and drops its mockery driver mocks once this merges.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
#### Description
- Noz's explorer chips carried the query but not its time window, so the
Explorer opened on the current picker value (e.g. Last 6 hours) instead
of the window the answer was computed for.
- Chips now apply the action's `start`/`end` as an absolute range:
off-page via `startTime`/`endTime` on the Explorer URL; on an
already-open Explorer by also setting the global time and dropping the
stale `relativeTime`.
- Without a valid range nothing changes: query and filters hand over,
time stays as is.
#### Issues closed by this PR
ClosesSigNoz/nerve-pod#174
#### Screenshots / Screen Recordings
Prev -
https://github.com/user-attachments/assets/9d54125f-ed32-486e-acc2-2d1c59963fa5
Now -
https://github.com/user-attachments/assets/1ce75dc0-d62a-430c-b458-ca1036459955
#### Additional Information
- Stacked on #13557, which carries the `redirectWithQueryBuilderData`
change this relies on.
- The Explorers' global time is Redux-only, so `useDispatch` goes behind
the existing `no-restricted-imports` escape hatch.
- Time params are not written with nuqs: the Explorers' time is Redux +
react-router URL writes, and a second nuqs writer adds a history entry
or leaves react-router with a stale `relativeTime` (see
`utils/timeUrlParams.ts`). Once the Explorers move to
`store/globalTime`, the chip can call `setSelectedTime` and drop its URL
handling.
---------
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- using signoz otel collector image for ch migration
- adding auditquerier in test suite in ci
- added v0.144.16-rc.1 temporarily
#### Description
- `redirectWithQueryBuilderData` deletes a search param passed as `null`
instead of writing `"null"`, so one navigation can swap params (e.g.
`relativeTime` for `startTime`/`endTime`).
- Its trailing `newTab` flag becomes `navigateOptions: { newTab, replace
}`, passed straight to `safeNavigate`.
#### Issues closed by this PR
#### Screenshots / Screen Recordings
#### Additional Information
- Split out of #13363. Stacked on #13556 only to keep #13363's diff
clean; it doesn't depend on it.
- The `source: null` reset in `CreateAlertV2/QuerySection` now removes
`source` instead of writing `source=null`; its readers only compare
against `DASHBOARDS`, so behavior is the same.
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- add resource defs to the self-service user routes: get and update my
user, change my password, forgot password, reset password, verify reset
token
- why? so that the auditor can pick up these routes for auditing
- update my user and change my password are audited today
- forgot and reset password are pre-auth, so their audit events are
dropped at the zero org id check until the principal is resolvable
- routes stay open access, no API or OpenAPI spec change
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
contributes to https://github.com/SigNoz/keystone-pod/issues/10
#### Description
- Fixes `ruler/08_rule_validation.py` failing on main after #13253,
since the rule authz check runs before the handler's validation.
- Malformed rule JSON returned 500 from the channel extractor, it now
decodes strictly like the handler and returns 400.
- Channel attach checks skip names that don't exist, so the handler's
`validateChannels` returns its clearer error (e.g. channel referenced by
name instead of display name). Existing channels still need attach.
- A malformed rule id now gets the generic `invalid_selector` 400 like
every other FGA route, test updated.
#### Description
- Schema migrator `v0.144.13+` (migration 1002) seeds the traces
`attributes` JSON rollout at install, as it already does for resources.
Tests that wrote their own rollout rows broke on it: `[map]` cases read
JSON, and the seed teardown deleted the migrator's row
(`test_summary_ai_trace[map]` failing on #13517).
- Bumps the schema migrator to `v0.144.15` (CI, tests, devenv). Tests no
longer write `column_evolution_metadata`; rollout-straddle tests anchor
to the migrator's time via `get_attribute_evolution_time`, mirroring
`get_resource_evolution_time`.
- Map/JSON dual run kept: `[map]` anchors spans and query window an hour
before the rollout, where the querier reads only the legacy maps, the
same SQL as with `use_trace_attributes_json` off. `queriertraces` raises
the flux interval so those past windows aren't served from each other's
cached buckets.
#### Additional Information
- Unblocks #13517, which switches migrations to the otel-collector
image.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: swapnil-signoz <swapnil@signoz.io>
#### Description
- `safeNavigate` replaced the history entry whenever the target added a
param the current URL lacked, even when the caller passed `replace:
false`. An explicit `replace` now wins; without one, behavior is
unchanged.
- No caller passes `replace: false` today, so nothing changes until one
opts in.
#### Issues closed by this PR
#### Screenshots / Screen Recordings
#### Additional Information
- Split out of #13363, whose Noz chips need a history entry Back can
return to when they add `startTime`/`endTime` on an open Explorer.
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
#### Description
- Move `/api/v2/rules*`, `/api/v3/rules` and `/api/v2/rules/test` from
role gates to `CheckResources` on rule verbs.
- Create, update, patch, clone and test also check attach and detach on
the notification channels the rule names, diffed against the stored
rule. Unknown channel names are rejected with 400.
- Create, update, patch, clone and test also check read on the telemetry
the rule's query touches, so a rule cannot be used to read signals the
caller cannot query directly. Managed roles already hold these reads,
only custom roles are affected.
- PATCH checks telemetry read only when the body carries
`condition.compositeQuery`, so disabling or renaming a rule needs no
read on its stored query. This adds an opt-in `Optional` flag on
`TelemetryResourceDef` for that one route, every other telemetry def
keeps failing closed.
- Detach of a channel name that no longer exists is skipped instead of
rejected, so a rule whose channel was removed can still be updated.
Attach stays strict.
- Integration tests cover viewer, editor and a custom role on every
gate. No migration, rule tuples come from #13251 and channel links from
#13249.
#### Issues closed by this PR
ClosesSigNoz/pulse-pod#286ClosesSigNoz/pulse-pod#355
#### Additional Information
Fifth of five stacked PRs splitting #12927. Stacked on #13252.
#### Description
- Merges to `main` now build and release the staging base
(`overlays/staging/*/signoz/base`), which every staging overlay
inherits.
- `staging:<deployment>` labels on PRs still release only that
deployment's overlay.
- Releases go to `signoz.staging.cl.gcp.deployments` with a `{path,
image, tag}` payload instead of `{deployment, signoz_version, source}`.
- The deployment label is checked against
`^[a-z0-9]([a-z0-9-]*[a-z0-9])?$` before it goes into the path glob and
the JSON payload.
- `checkout`, `create-github-app-token` and `cache` are pinned to
release SHAs.
#### Description
- The plays on the alert rules page, the dashboards list and the
metrics, traces, logs and pods explorers now check the args they depend
on: whether rows are listed, which permissions are held, whether the
data loaded, which tab, view or format is open, and how a write answers.
Run once against every control value, 167 of 969 alert rules runs, 336
of 1,372 dashboards list runs and 212 of about 2,200 explorer runs
failed on `main`. All of them pass now.
- The explorer pages' plays moved out of their story files into play
files, as the skill asks. Plays that had become the same play under a
different control value are now one.
- `grantsAll` in `src/storybook/access/access.ts` answers a permission
check from a story's access controls, the same way the `authz/check`
mock does.
- The `signoz-page-story` skill gets a **Plays** section listing the
cases these plays missed and how to guard them with the page's own code,
and `.claude/rules/storybook.md` points at it. CI runs each play only
with its story's own args, so these rules are what keeps a new play from
missing the same cases.
- No screen renders differently: only plays, their mock helpers and the
skill changed.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/727
#### Description
- The generated API client now sends array query params as repeated keys
(`states=firing&states=pending`) instead of axios' default `states[]=`.
The backend binds only repeated keys, so a `states` filter sent through
a generated hook was ignored and every rule came back.
- The alert list page and the home page alerts card drop their own
requests that worked around this.
- `states` on `GET /api/v3/rules` is the only array query param in the
generated and AI assistant clients today, so no other request changes.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/719
#### Description
- `ListAlertRules.views.test.tsx` now gives each test 15 s instead of
Jest's default 5 s. Other slow page tests use the same timeout.
- "deleting the active view falls back to All" is the slowest test in
the file, about 0.8 s locally. CI runs the file about 4x slower (48 to
52 s against 13 s locally), which pushes it past 5 s. It failed the
merge queue for [#13476](https://github.com/SigNoz/signoz/pull/13476)
and [#13301](https://github.com/SigNoz/signoz/pull/13301), and two
unrelated PRs.
- Each step in it finishes well inside the 1 s `waitFor` limit. Only the
total runs over, so this is load, not a stuck wait.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/726
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- using X-Forwarded-For header for client.address as audit log attribute
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
closes https://github.com/SigNoz/keystone-pod/issues/213
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- Audit query_range reads the new `signoz_audit` schema: `body`,
`scope`, `attributes` and `resource` are JSON columns, and the old typed
maps are gone.
- Audit skips the attribute and resource key tables, which the new
schema no longer has. Keys resolve from the intrinsic fields plus a
fixed list of the resource keys the audit writer emits, so bare resource
keys (e.g. `signoz.resource.kind`) work without the `resource.` prefix
until keys move to `signoz_metadata`.
- Raw audit rows return `resource` with flat dotted keys
(`resource.attributes["service.name"]`) instead of nesting at every dot.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
closes https://github.com/SigNoz/keystone-pod/issues/201
#### Description
- The All Endpoints and Endpoint Stats pages use `QuerySearch` for the
filter and `GroupByFilter` for the group-by, the query builder's own
components. This removes the last v3 autocomplete calls in API
monitoring.
- Each drawer tab keeps its filter in `QuerySearchV2Provider`, with its
own URL param (`apiMonitoringAllEndpointsExpression`,
`apiMonitoringEndpointStatsExpression`), the same way infra monitoring's
tabs do. The payload builders and the row-click drill-down work on the
expression directly, so the items conversion is gone. The filter applies
on Ctrl+Enter or when focus leaves the box.
- The group-by page no longer fetches keys or caches them to resolve
picks. `GroupByFilter` returns full keys, so about 80 lines of page
logic are gone.
- A Run button sits next to the filter on both pages. It applies the
filter the same way leaving the box does, and shows a spinner while that
tab's queries load.
- The filter shows the drawer's locked scope as a chip in front of the
text, like infra monitoring: `http_host = '<domain>' +1`, plus
`http_url` on Endpoint Stats. The chip is a new opt-in `scopeClauses`
prop on `QuerySearch`.
- Endpoint Stats and Top 10 Errors pick the endpoint from a searchable
select above the filter, not from text in the expression. The endpoint
is added to the request filter when the request is built.
- Removed the dead v4 fallback on the stats page and an unused
zero-state payload builder. Both have had no effect since November 2025.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/143
Closes https://github.com/SigNoz/events-pod/issues/144
#### Screenshots
sbshot:
https://claude.ai/artifact/YLourNcAvCqsmVbtnGhH2L?sk=bVmSGZMwZ7VSmSukZLyNAwhttps://github.com/user-attachments/assets/bf6e9930-4d14-40f7-ba66-fe36045eb4ef
url now carries filter exp
https://github.com/user-attachments/assets/3310000a-d03a-4a54-9119-dde748aa6dc7
#### Additional Information
- The filter applies on Ctrl+Enter or blur, not on every completed chip.
Free-form expressions (OR, NOT, functions) now work.
- Group-by keys store the v5 field context (`attribute` / `resource`)
where v3 wrote `tag`. Keys restored from the URL carry the name only; v5
resolves them by name.
- The group-by has no clear-all cross and no `+N` collapse because the
shared component has neither. Every selected key stays visible with its
own ×.
- The filter row on the stats page now grows with a multi-line
expression instead of overflowing.
- Links saved with the old items-shaped filters in `apiMonitoringParams`
still open. The list page moves those filters into the tab params once
and drops the old keys.
- The drawer starts from the list's filter when you open it by clicking
a domain. A drawer opened from a link uses only the filters in the link.
A cleared tab filter stays cleared across tab switches and reloads.
- The user's filter is joined to the page's base filter as `base AND
(filter)`, so an `OR` in the user's filter stays scoped to the domain
and endpoint.
- Clicking a grouped row whose group value is empty now drills down with
`key NOT EXISTS`. It used to send `key = null`, which matched nothing.
- A hand-typed `http_url = …` in the All Endpoints filter now carries
into Endpoint Stats as part of the user's filter. Before, opening a row
dropped it, so "All endpoints" showed every endpoint. Now the clause
still narrows the data and the endpoint list. Other `http_url` operators
behave as before.
- Fixed a URL bug: writes to `apiMonitoringParams` now build on the live
URL. Before, several writes in one render kept only the last one. On
main, picking a time range in the drawer dropped `selectedInterval` from
the URL.
- `QuerySearchV2Provider` now follows URL changes after mount. Before,
it read its URL param once and wrote its stored filter back over later
changes, so Back and Forward did not restore a filter. Using it here
exposed this. Infra monitoring's Logs, Traces, Events and Overview tabs
share the provider and get the same fix.
- The `QuerySearch` changes are opt-in props (`showRunButton`,
`isLoadingQueries`, `scopeClauses`). Its other callers, infra monitoring
included, render as before.
- Storybook: the existing drawer stories (Domain Endpoints, Endpoint
Stats, Service Filtered, Port Domain, Failing Domain, Top Errors)
already show the picker, the Run button and the chip. No story shows the
Run button busy.
- Part of https://github.com/SigNoz/events-pod/issues/156.
#### Description
Visual cleanup of the alert rules list, to match the dashboards list:
- Removed the gap between the tab bar and the list.
- Rows use one black background instead of alternating colors.
- Rule labels are neutral grey instead of red, so only firing state and
severity draw attention.
- Column header dividers show only on hover.
#### Additional Information
- The existing `Pages/Alerts/Rules` stories already render these states;
no story change needed.
#### Description
- Add Pixeltable and Claude Cowork cards to the onboarding data sources.
Pixeltable opens the trace-first guide. Claude Cowork opens the
log-event guide.
- Add a Pixeltable logo (`Logos/pixeltable.svg`). Pixeltable has no
vector logo, so I redrew its dot-matrix "P" on a 24x24 grid.
- Add the `python celery` search keyword to the Celery card. Searching
`python celery` returned no results before.
- Fix the built-in and AWS integration card links. The old
`/integrations?integration=...` links only opened the integrations list.
They now use `/integrations/builtin-<id>` and
`/integrations/aws?service=<id>`.
#### Issues closed by this PR
ClosesSigNoz/signoz.io#4487ClosesSigNoz/signoz.io#4482ClosesSigNoz/signoz.io#3279
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
We don't want external clients like Terraform or AI Agents to be sending
these fields anymore cuz they are useless. That's why we need to remove
these from the schema. But sadly, we probably have users who send these
in their scripts, and our rules APIs don't accept fields not in the
struct, so these fields stay in the struct.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/731
#### Description
- The Terraform provider is generated from our OpenAPI spec. #13424
described the Jira channel `auth` as a `oneOf` (cloud or self-hosted)
with no discriminator, which the generator cannot map, so the
notification channel resource stopped building.
- `deployment` sits next to `auth`, not inside it, so it cannot act as
the discriminator.
- This PR describes `auth` in the spec as one object with optional
`email`, `apiToken` and `pat`. Each description says which deployment
needs it.
- The Go type now matches the spec: one auth struct replaces the two
per-deployment structs, and the fields are validated against
`deployment`.
- The accepted payload, the responses and stored channels stay the same,
so no migration is needed.
#### Issues closed by this PR
ClosesSigNoz/pulse-pod#728
#### Additional Information
- Problem and options doc:
https://app.notion.com/p/415fcc6bcd1980a0869fd5ed2e19ca3c
- The spec no longer marks the auth fields as required per deployment.
The server still rejects a wrong or missing combination with a 400.
- A field of the other deployment (for example `pat` on a cloud channel)
now gets an explicit error message instead of the unknown field one. It
is rejected only when it is not blank.
- The generated spec and API types are in a separate commit.
#### Description
- `QuickFiltersLayout` and `QuerySection` in the AI Explorer tests.
Neither is under test, and rendering them made the tab-switch test take
~1s; it now takes ~0.1s.
- Assert the trace view and its URL columns before switching tabs, so
the cleared-columns check can't pass when the columns were never set.
- After the switch, assert the list panel type is in the URL and the
trace view unmounts.
#### Issues closed by this PR
#### Screenshots / Screen Recordings
#### Additional Information
Co-authored-by: Gaurav Tewari <tewarig@users.noreply.github.com>
#### Description
- API calls that nobody catches reach Sentry as unhandled rejections.
Most of these failures are expected: a 403 for a user without a viewer,
editor or admin role, a 401 for an expired session, or a network error
from a dropped connection.
- Since the axios 1.21.0 bump (#13339), these errors carry our call site
in the stack. Sentry opens a new issue for each call site, and each new
issue pages. SIGNOZ-UI-5N4 paged twice this way.
- `beforeSend` now drops axios errors with status 401 or 403, or with
code `ERR_NETWORK`. Sentry still gets 400, 404 and 5xx.
#### Issues closed by this PR
Closes https://github.com/SigNoz/events-pod/issues/261
#### Additional Information
- The filter matches raw axios errors only. A call that goes through
`ErrorResponseHandlerV2`, `ErrorResponseHandlerForGeneratedAPIs` or
`convertToApiError` rethrows an `APIError`. Sentry still reports that
error when nobody catches it.
- The filter drops all 403 and `ERR_NETWORK` errors. A missing
permission check in the UI, or a CORS or base-path regression, no longer
shows in Sentry. This is a deliberate choice.
- The filter also drops 401 errors from the token refresh retry path.
- A separate PR adds an error state for the empty key suggestions in the
query search bar.
#### Description
- `PUT /api/v2/quick_filters/{source}` (and the legacy v1 update) now
returns 400 for fields whose values are unique per span/log, so a quick
filter on them has nothing to pick from: trace/span/parent-span IDs
(incl. legacy `traceID`/`spanID`/`parentSpanID`), `timestamp`, log
`body`, `gen_ai.tool.call.id`, free-text payloads
(gen_ai/OpenInference/Langfuse/Traceloop messages,
`exception.message`/`stacktrace`, `db.statement`/`db.query.text`), token
counts (`gen_ai.usage.*`, `llm.token_count.*`, `ai.usage.*`, …) and the
per-span `signoz.gen_ai.*.cost` keys. The list is hard-coded in
`quickfiltertypes`.
- `trace_id` is removed from the default traces quick filters, including
the seeds in migrations `030`/`031` (separate commit), so `147` is a
no-op on fresh installs.
- Migration `147` strips these fields from every stored quick filter;
other entries are kept byte-for-byte.
#### Issues closed by this PR
Closes#13157
#### Additional Information
- Follows the approach agreed in SigNoz/platform-pod#3282, replacing the
frontend-only blocklist from #13182.
- Changelog: `trace_id`, `span_id` and other high-cardinality fields
(token counts, costs, message payloads) are no longer supported as quick
filters and are removed from existing quick filter settings; filter by
them from the search bar instead.
- The quick filter settings panel still lists these fields as
suggestions; adding one now surfaces the API error. Hiding them is out
of scope.
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
#### Description
- The alert rules State filter now uses the same select as the Created
by filter: checkbox options, one pill plus "+N", and the same background
and width. The placeholder reads "State" instead of "Filter by state".
- The Created by select moves into a shared `MultiSelectFilter` in
`components/ListFilter`, and both filters render through it. Created by
looks and behaves as before on the dashboards and alert rules lists.
- The State filter moves from `@signozhq/ui` `ComboboxSimple` to antd
`Select`, to match the antd Created by and Updated filters beside it.
`ComboboxSimple` can't run the query once on close or show checkbox
options yet (https://github.com/SigNoz/components/issues/483).
- New story on the alert rules page: the State dropdown open.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/720
#### Screenshots / Screen Recordings
Visual diff in dark and light:
https://claude.ai/artifact/UwA1RZA6VTDMEEbGpyVsvg
- 10 changed, all inside the State filter box on the alert rules list.
- 2 new: the State dropdown open.
- 8 pixel-identical: the dashboards list, where Created by is unchanged.
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- Get and delete check the path id. Bulk upsert checks each rule id and
falls back to the wildcard for creates and sync rows.
- Managed roles keep the same access as before: admin writes, editor and
viewer read. A migration backfills the tuples for existing orgs.
- Integration tests cover managed roles plus wildcard and rule-scoped
custom roles.
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
Closes https://github.com/SigNoz/nerve-pod/issues/463
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- using semantic conventions described in `signoz-semanatic-conventions`
repository for audit
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
contributes to https://github.com/SigNoz/keystone-pod/issues/188
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- add `kafka` auditor provider that produces audit logs to a
kafka/redpanda topic as `otlp_proto`
- uses franz-go v1.21.7.
- no delivery timeout or retry cap: on franz-go < v1.22.2 these can ack
a record that was never written (twmb/franz-go#1484).
- added unit tests
<!--Reference issues using `Closes #issue-number` to enable automatic
closure on merge. -->
#### Issues closed by this PR
closes https://github.com/SigNoz/keystone-pod/issues/2
<!--Please delete paragraphs that you did not use before submitting.-->
#### Description
Extends the metrics heatmap to logs, traces and audit logs, so a heatmap
can be built ad hoc from raw events with any filter instead of from a
pre-generated metric.
- The aggregation is the bare field to bucket, e.g. `duration_nano`.
Each cell counts events. Any other aggregation, such as `count()`, is
rejected, and a query takes one aggregation.
- Rows without the field are excluded by default: `<field> EXISTS` is
ANDed onto the filter, so it also applies to the resource filter and the
group limit. A present but non-numeric value counts as 0.
- `bucketOptions` is optional and defaults to the log axis, as for
metrics. Distribution requests work the same way.
- The heatmap is the time series query with the value bucket joining the
group keys, so a row is one cell and the response shape is unchanged.
Results are cached, and a cached axis is trimmed to the requested
window.
- The bucket expressions move to `pkg/querybuilder/heatmap.go`, shared
with metrics. The value is read once through a `WITH` alias instead of
being repeated in every bound. The metrics SQL is unchanged.
- A disabled query in a heatmap request feeds a formula, so it builds as
a time series, as metrics already did.
- A field nothing carries draws an empty chart with the key-not-found
warning.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/338
#### Additional Information
- No API schema change.
- Integration tests (`querierlogs/17_heatmap.py`,
`queriertraces/16_heatmap.py`, `auditquerier/02_heatmap.py`) cover
linear, log and default axes, group by, rows without the field, and the
rejections.
- Backend only. Dashboards first, explorers later.
#### Description
- A logs or traces time series query with a group by and a limit ranks
its groups over the statement window. On a partial cache hit the gap
statement ranked over the gap, so the response was a union of the
winners of each piece, each with points for part of the window. A
narrower window served from a wider cached entry got the wider window's
winners.
- No piece of a window can stand in for the window, so such a query is
no longer served from the cache. Its fingerprint is empty, the same way
an AI builder query opts out today. Metrics are not affected, their
limit is applied after the full series set is read.
- Integration tests: a top-2 over a window whose first half is cached
must return the window's two winners with points for the whole window,
and a top-1 over the second half of a cached window must return that
half's winner. Both match the request with noCache.
#### Issues closed by this PR
ClosesSigNoz/nerve-pod#627
#### Additional Information
- The first commit adds the tests alone. CI ran them against main and
both failed. The second commit is the fix.
- This is the stop gap. Caching these queries exactly needs the ranking
summary of each piece, top-K totals with a threshold, so the merge can
rank with bounds and fall back when it cannot prove the top-N. That is a
design item, not a cache patch.
#### Description
- A timeShift query's window is moved into the shifted clock when the
query is built. On a partial cache hit the missing range comes from that
window, and the ranged gap query shifted it again, so the gap was read
one shift too far back, and the wrong points were served and written
back to the cache.
- The ranged query now takes the range as given. The spec copy already
carries the shift, so the gap statement reads the right data and the
postprocess shift moves it forward once.
- Integration test: a timeShift(3600) logs query with its first 30
minutes cached must return the same points for the next 20 minutes as
the request with noCache.
#### Issues closed by this PR
ClosesSigNoz/nerve-pod#625
#### Additional Information
- The first commit adds the test alone. CI ran it against main and it
failed with only that test failing in the suite. The second commit is
the fix.
#### Description
The querier computes a formula in Go after its queries return, so a
formula having cannot run as SQL HAVING. It was accepted and ignored.
The having now filters the formula values in Go. It uses the having
grammar, validation, and messages of a builder query, with the formula
name and __result as the only references. A time series keeps the
points that match and drops a series with no points left. A scalar
result keeps the rows that match. The having runs before functions,
order, and limit, as HAVING does in SQL. A bad having fails the
request before any query runs.
#### Issues closed by this PR
ClosesSigNoz/nerve-pod#612
#### Additional Information
Assisted-by: Claude Opus 5.5
#### Description
- A negative formula `limit` now returns 400, as for builder queries.
Before, validation accepted it, and the series limit treated it as no
limit.
- An empty order key on a `raw`, `raw_stream`, or `trace` request now
returns 400. Before, validation accepted it. Aggregation requests keep
their own order key check, which lists the valid keys.
- A missing `requestType` returned `invalid request type: {}` and
suggested `timeseries`, which the decoder rejects. It now returns
`requestType is required`, with the valid request types as suggestions.
The decoder error for an unknown value lists the same set.
- Validation now accepts exactly what `RequestType.Enum` lists. The
spec, the decoder error, and validation read one list, so they cannot
drift apart.
#### Issues closed by this PR
ClosesSigNoz/nerve-pod#609
#### Additional Information
- Existing alerts should not be affected. Stored rules load without
validation, and the querier validates only in the `query_range` handler.
The UI does not send a negative formula limit or an empty order key.
- Part of https://github.com/SigNoz/nerve-pod/issues/428
#### Description
- The query_range examples cover signals and request types but none of
the functions a builder query can carry, so a client has no place to
learn the argument shapes.
- Adds one example per function family: timeShift, fillZero, the four
threshold functions, runningDiff, ewma, median, cumulativeSum, log,
anomaly. Each description says what the function does to the series,
which arguments it takes, their units and their defaults.
- A test unmarshals every example into a `QueryRangeRequest` and
validates it, so an example cannot drift from the request type.
- The OpenAPI spec is regenerated in its own commit.
#### Description
- Saving a rule, and cloning or renaming one from its details page, now
refreshes the v3 rules list instead of v2. The list page (#13322) and
`useListAllRules` (#13440) both read v3, so nothing reads the v2 list
anymore.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/718
#### Description
- The AI assistant's Alerts context picker now loads rules with one
`/api/v3/rules` request (`limit: 200`, the most the API returns per
call) instead of the deprecated `GET /api/v2/rules`. The Dashboards tab
next to it already loads one page (`limit: 1000`) the same way.
- Alert context chips read their names from the same cache entry.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/717
#### Description
- The planned downtime rule picker now reads `/api/v3/rules` through a
new `useListAllRules` hook instead of the deprecated `GET
/api/v2/rules`.
- The picker lists every rule, and v3 returns at most 200 per call, so
the hook fetches the remaining pages in parallel, like
`useChannelOptions` does for channels. Orgs with 200 rules or fewer make
one request.
- The planned downtime story now mocks `/api/v3/rules`.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/713
#### Description
- `GET /api/v2/rules` is deprecated in favor of the paginated
`/api/v3/rules`, and #13158 removes it. This moves the home page alerts
widget to v3 ahead of that.
- The widget still shows firing rules first, then the most recently
updated others, up to 5. It asks v3 for the firing rules first and
fetches the latest rules only when fewer than 5 are firing, instead of
loading every rule.
- Clicking a rule now opens the overview with only `ruleId`, like the
alert list, triggered alerts and Metrics Explorer. The overview loads
the rule and builds its query from it, so the `compositeQuery` the old
link carried is not needed.
- The home page story now mocks `/api/v3/rules`.
#### Issues closed by this PR
Closes https://github.com/SigNoz/pulse-pod/issues/712
#### Description
- `@SigNoz/pulse-frontend` now owns `components/ListFilter`,
`components/ListViews` and `hooks/listViews`. Only the dashboards list
and the alert rules list use them.
- It also owns `container/PlannedDowntime`, `hooks/alerts` and the home
page alert rules widget (`container/Home/AlertRules`).
- Until now these fell to the `/frontend/` catch-all, so changes to them
also needed `@SigNoz/frontend-maintainers`.
<!--A few plain bullets saying what changed and why, for a reviewer
skimming it - not a wall of text, not a restatement of the diff, not
generated boilerplate.-->
#### Description
- bump go version in ci to 1.25
---------
Co-authored-by: Ashwin Bhatkal <ashwin96@gmail.com>
2026-11-09 10:08:41 +00:00
6132 changed files with 132173 additions and 70444 deletions
@@ -11,7 +11,7 @@ Page stories are written with the `signoz-page-story` skill ([`.claude/skills/si
- **New page, new story. New state, new control or variant.** The skill decides which: a control on a response, a variant worth linking to, or a `play` story for a state only a click reaches.
- **A shared component is covered through its pages.** A change under `components/` or `container/` with no story of its own updates the stories of the pages that render it.
- **Plays live with the mocks.** A `play` is a named function exported from `stories/__story_mockdata__/<page>.play.ts`, which is collapsed for review like the rest of the AI-owned mocks. The story file only names it: `play: openEditDrawer`. A story file that still holds play bodies moves them there the first time you edit it.
- **A play passes for every control value.** Branch each check on the args it depends on (panel kind, data state, row count, hidden header). A value the page supports must never fail the play.
- **A play passes for every control value.** Branch each check on the args it depends on (panel kind, data state, access, check state and revoked permissions, license, held-open tooltips, a count at 0, a view or tab that drops the target, a write state). A value the page supports must never fail the play. The `signoz-page-story` skill's **Plays** section lists the cases plays have missed.
- **Mocks answer every request a reachable state makes.** Selecting a row, opening a view or switching a tab that fetches more hits a handler, never the 501 catch-all.
- **Already covered, say so.** When the existing story and its controls already reach the changed state, no story edit is needed; state that in the PR description.
- **Opt out only when there is no visual diff.** Skip stories only when the change renders nothing differently: types, api clients, hooks with no UI branch, tests. Deleting a page deletes its stories.
@@ -18,7 +18,9 @@ tag <diff> <tag> <story-id>... # what the reviewer should look at
export <diff> --out <dir> # a static site, hostable
```
`<storybook>` is a running dev server (`http://localhost:6006`), a build directory holding `index.json`, or a project directory with `.storybook/`. The frontend has **no `build-storybook` script**, only `storybook dev -p 6006`, so point `capture` at a dev server.
`<storybook>` is a running dev server (`http://localhost:6006`), a build directory holding `index.json`, or a project directory with `.storybook/`. **Capture from a build**: `pnpm storybook:build` in `frontend` (about 2 minutes) writes `storybook-static/`. A dev server serves the preview module by module, which adds 5 to 10 seconds to every story and, with several stories at once, ends in `page.goto: Timeout 30000ms exceeded` retries.
Pass `--globals 'a11y.manual:!true'` to every `capture`. The a11y addon otherwise runs axe after every story, about 2 seconds each, and nothing it does shows in a shot.
## Cover the change with stories first
@@ -30,7 +32,7 @@ Then confirm it worked: the two runs' story counts should differ by exactly the
## Rules that are not optional
**One freshly started Storybook per ref.** Never start one server and switch git branches under it. Vite's HMR reloads the page, but some modules do not re-initialise, so the shot differs silently from a clean boot and the difference looks like a real change. Settle time does not fix this and is not the variable. Use one worktree and one port per ref, and stop each server before the next.
**One freshly started Storybook per ref.** Never start one server and switch git branches under it. Vite's HMR reloads the page, but some modules do not re-initialise, so the shot differs silently from a clean boot and the difference looks like a real change. Settle time does not fix this and is not the variable. Use one worktree and one port per ref, and stop each server before the next. A build made in each ref's worktree has none of this.
**Capture both themes: `--theme dark,light`.** Without it sbshot shoots the story's own default and everything lands in a single `default/` folder, so every regression in the other theme is invisible. Themes double the shot count, so scope with `--stories` or `--title`, never by dropping a theme.
**Delete a story once its difference disappears.** Say a story checks that a button reads "View" rather than "Edit" when a permission is missing. If the label later becomes the same either way, that story passes no matter which permissions are held, including all of them, and it is testing nothing. When a behaviour stops depending on a permission, delete the story or rewrite it to lock in the new behaviour, and say which one you did in the PR.
## Plays
CI runs each play once, with its story's own args. Anyone can flip a control in the panel, and nothing runs the play against that value, so a play that breaks on it goes unnoticed. Write each play so it cannot.
Before writing the clicks, list what has to be on screen for the play's target to exist, and return early when the args take it away. These are the cases plays have missed:
- The controls every story has. `State` on loading or error renders no rows and no filter suggestions. `Access`, `Permissions` and `Revoked` remove rows, row actions and gated buttons, and the `editor` and `viewer` presets grant telemetry reads only, so a page gated on its own resource shows them nothing. `Check state` on loading disables every gated control, and on error an `AuthZButton` stays enabled while an `AuthZGuardContent` set to fail closed hides what it wraps. `License` and `Banner` decide the plan, so a cloud-only form is gone on the others. `Hold tooltips open` shows a label a second time, inside its tooltip.
- A count at 0. The page can swap its whole list for an empty state, and the header, filters and column picker go with it.
- A view, filter or invalid query that leaves no rows, even when the count is above 0.
- A tab without the widget. A filter box that one tab has is not on the next.
- A second list beside the main one, such as saved views. At 0, loading or failed, its actions are gone.
- Text that depends on an arg. A disabled rule offers Enable, not Disable. A write that fails raises a different toast from one that lands, and one left loading raises none. Pick the expected text from the args.
- Rows that move. A view, a sort or pinning changes which row is locked, pinned or legacy, so find the row by what it is, not by its index.
Make those checks with the page's own code instead of restating a condition. Build the rows with the builder the handler uses and the view's own filter, and check a permission with the page's own check through `accessFor(...).allows(...)`, or the role against the constant the page gates on, such as `componentPermission.action`. Export the checks from the mocks module, next to the handlers, so a handler change moves them too. The alert rules page has `listedRules`, `hasRowMenu`, `showsNoRulesPage` and `listsSavedViews`; the dashboards list has `listedDashboards`, `canList`, `canCreate` and `canUpdate`.
Two habits cover the rest:
- Use `findAllBy*` and take the first whenever the text can appear twice, such as in a tooltip or a popover title.
- Wrap `toBeVisible` in `waitFor` for anything that fades in. A tooltip or toast found the moment it mounts is not visible yet.
## Rules
- **Default is the loaded page.** `export const Default: Story = {}` with no args, every widget carrying data. Empty, loading and failed are variants or control values, never the default.
- **Endpoints the page owns go through `response.json`**, so the Data control covers loaded, loading and failed in one declaration. Endpoints the page cannot render without (ingestion detection, preferences, feature payloads) take a plain resolver so the shell survives the loading and error states.
- **Query-param state starts from `route`** (`/logs?tab=explorer`). In-page param navigation works inside a story; a different pathname is blocked and reported by the overlay. A control for a param is worth it only when the param is a page mode someone would want to flip.
- **File layout**: every story file for a page lives under `src/pages/<Page>/stories/`: `<Page>.stories.tsx`, `<Page>.stories.mocks.tsx`, payload builders in `stories/__story_mockdata__/<page>.ts`, play functions in `stories/__story_mockdata__/<page>.play.ts`. Nothing page-specific in `src/storybook/controls/`. A page that is a tab strip over several routes gets one story file per tab, in its own folder under the module page (`LogsModulePage/Pipelines/stories/Pipelines.stories.tsx`), each with its own mocks and `__story_mockdata__/`; the builders more than one tab needs stay in the module page's own `stories/__story_mockdata__/` (`AlertList/stories/__story_mockdata__/alerts.ts`), which a tab reaches as `../../stories/__story_mockdata__/alerts`. Every one of them renders the module page, so the tab strip is there, and the `route` its mocks return decides which tab is open. A page's permission stories go one level further down, in `stories/authz/<Page>.authz.stories.tsx`, on the page's own mocks: see **Permission stories**.
- **A state only a click reaches is a story with a `play` function**, not a control: a drawer, a modal, an edit mode the page holds in component state. Drive it with `userEvent` and the queries from `storybook/test`, take the first of a repeated row action, and wait on the state's own text. The page fetches before it renders a row, so the finder needs a timeout past the 1s default. The play is a named function exported from `__story_mockdata__/<page>.play.ts` and the story file only names it (`play: openEditDrawer`), so the reviewer reads states, not clicks; a story file that still holds play bodies moves them there the first time you edit it. It branches every check on the args it depends on: a control value the page supports never fails it. A state the app drops again on its own, such as one keyed on an array identity that a refetch replaces, does not get a story: it would not survive being looked at. A _sequence_ of such states, a wizard's steps or a questionnaire's pages, is still a control: declare the steps in the mocks module and walk them from a `play` on the meta that destructures `mount`, which is what makes Storybook replay it on an arg change. See [references/controls.md](references/controls.md).
- **A state only a click reaches is a story with a `play` function**, not a control: a drawer, a modal, an edit mode the page holds in component state. Drive it with `userEvent` and the queries from `storybook/test`, take the first of a repeated row action, and wait on the state's own text. The page fetches before it renders a row, so the finder needs a timeout past the 1s default. The play is a named function exported from `__story_mockdata__/<page>.play.ts` and the story file only names it (`play: openEditDrawer`), so the reviewer reads states, not clicks; a story file that still holds play bodies moves them there the first time you edit it. It branches every check on the args it depends on: a control value the page supports never fails it. See **Plays**. A state the app drops again on its own, such as one keyed on an array identity that a refetch replaces, does not get a story: it would not survive being looked at. A _sequence_ of such states, a wizard's steps or a questionnaire's pages, is still a control: declare the steps in the mocks module and walk them from a `play` on the meta that destructures `mount`, which is what makes Storybook replay it on an arg change. See [references/controls.md](references/controls.md).
- **The mocks are AI-owned and say so.** `<Page>.stories.mocks.tsx` and every file under a `__story_mockdata__/` open with this banner, above the imports:
- [ ] a page tagged `authz` has its `Authz` folder: one story per permission it reads, each reached by `revoked`, none of them a role preset, and each one checked by the `disabled` state of what the permission gates
- [ ] the mocks module and every `__story_mockdata__` file carry the AI-owned banner
- [ ] every control flipped once, its effect seen on screen
- [ ] every play passes with each control value its story exposes, and the story file holds no play body
- [ ] every play passes with each control value its story exposes, checked against the cases in **Plays**, and the story file holds no play body
- [ ] every request a reachable state makes (a selected row, an opened view, a switched tab) has a handler
- [ ] console clean: no `[storybook] no msw handler`, no 501, no msw unhandled request, no React warning
git diff --compact-summary --exit-code || (echo; echo "Unexpected difference in generated web settings types. Run pnpm generate:config:web-settings in frontend/ locally and commit."; exit 1)
test -z "$(git status --porcelain)" || (echo; echo "Unexpected difference in geo map outlines. Run pnpm build:geo-outlines in frontend/ locally and commit."; exit 1)
provider.settings.Logger().WarnContext(ctx,"failed to write active license to cache",errors.Attr(cacheErr),slog.String("org_id",organizationID.StringValue()))
provider.settings.Logger().WarnContext(ctx,"failed to read active license from cache",errors.Attr(err),slog.String("org_id",organizationID.StringValue()))
`Do not set "${property}" in CSS. Render the text with @signozhq/ui/typography and use its props, or disable this rule with a reason after "--" saying why Typography cannot.`,
`Do not set '${property}' in CSS. Render the text with @signozhq/ui/typography and use its props, or disable this rule with a reason after \`--\` saying why Typography cannot.`;
"import { useMemo } from 'react';\nexport const A = () => {\n\tconst s = useMemo(() => ({ textAlign: 'left' }), []);\n\treturn <div style={s} />;\n};",
errors:[{message:message('textAlign')}],
},
{
name:'object typed CSSProperties',
code:
"import type { CSSProperties } from 'react';\nexport const s: CSSProperties = { fontStyle: 'italic' };",
errors:[{message:message('fontStyle')}],
},
{
name:
'object typed React.CSSProperties, reported once when also used as style',
code:
"import React from 'react';\nconst s: React.CSSProperties = { fontSize: 1 };\nexport const A = () => <div style={s} />;",
errors:[{message:message('fontSize')}],
},
{
name:'Record of CSSProperties',
code:
"import type { CSSProperties } from 'react';\nexport const s: Record<string, CSSProperties> = { a: { lineHeight: 1 } };",
errors:[{message:message('lineHeight')}],
},
{
name:'as and satisfies CSSProperties',
code:
"import type { CSSProperties } from 'react';\nexport const a = { fontSize: 1 } as CSSProperties;\nexport const b = { fontWeight: 1 } satisfies CSSProperties;",
errors:[
{message:message('fontSize')},
{message:message('fontWeight')},
],
},
{
name:'styled-components template',
code:
"import styled from 'styled-components';\nexport const A = styled.div`\n\tdisplay: flex;\n\tfont-size: 12px;\n\tcolor: ${(p) => p.c};\n\ttext-transform: none;line-height: 1;\n`;",
'Disallow direct react-router / lib/history imports; import from src/lib/router instead',
'Disallow direct react-router imports; import from src/lib/router instead',
category:'React Router migration',
},
schema:[],
messages:{
v5Router:
'Do not import react-router-dom directly. Use the src/lib/router facade (useAppNavigate, useAppLocation, useAppParams, AppLink, Redirect, matchRoute) so a router upgrade stays contained.',
v6Compat:
'Do not import react-router-dom-v5-compat directly. Use the src/lib/router facade instead; the compat package is an implementation detail of the migration and disappears with it.',
historySingleton:
'Do not import the lib/history singleton. Use the helpers in src/lib/router/navigation.ts, or useAppNavigate() to follow the nearest router. history loses basename handling once the router owns it.',
router:
'Do not import react-router directly. Use the src/lib/router facade (useAppNavigate, useAppLocation, useAppParams, AppLink, Redirect, matchRoute) so a router upgrade stays contained.',
routerDom:
'Do not import react-router-dom. The app is on react-router; on v7 react-router-dom is only a re-export shim, and installing it puts two copies of the router in the tree. Use the src/lib/router facade.',
mock:
'Do not mock the router directly. Render at a route with tests/test-utils, or spy on the facade with the tests/router-mocks helpers.',
* ${(p) => p.small && 'font-size: 11px;'} strings in their interpolations
*
* Identifiers passed to a style prop are followed to a local object literal.
* SCSS is covered by the stylelint config.
*/
constMESSAGE=(property)=>
`Do not set '${property}' in CSS. Render the text with @signozhq/ui/typography and use its props, or disable this rule with a reason after \`--\` saying why Typography cannot.`;
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.