mirror of
https://github.com/SigNoz/signoz.git
synced 2026-04-21 11:20:28 +01:00
Compare commits
4 Commits
platform-p
...
add-exampl
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
a72ab8421f | ||
|
|
72bc4d143f | ||
|
|
ed17003329 | ||
|
|
fd8b886617 |
3140
docs/api/openapi.yml
3140
docs/api/openapi.yml
File diff suppressed because it is too large
Load Diff
@@ -4774,7 +4774,7 @@ export interface RuletypesPostableRuleDTO {
|
||||
* @type string
|
||||
*/
|
||||
alert: string;
|
||||
alertType?: RuletypesAlertTypeDTO;
|
||||
alertType: RuletypesAlertTypeDTO;
|
||||
/**
|
||||
* @type object
|
||||
*/
|
||||
@@ -4899,7 +4899,7 @@ export interface RuletypesRuleDTO {
|
||||
* @type string
|
||||
*/
|
||||
alert: string;
|
||||
alertType?: RuletypesAlertTypeDTO;
|
||||
alertType: RuletypesAlertTypeDTO;
|
||||
/**
|
||||
* @type object
|
||||
*/
|
||||
@@ -4984,8 +4984,8 @@ export interface RuletypesRuleConditionDTO {
|
||||
*/
|
||||
algorithm?: string;
|
||||
compositeQuery: RuletypesAlertCompositeQueryDTO;
|
||||
matchType: RuletypesMatchTypeDTO;
|
||||
op: RuletypesCompareOperatorDTO;
|
||||
matchType?: RuletypesMatchTypeDTO;
|
||||
op?: RuletypesCompareOperatorDTO;
|
||||
/**
|
||||
* @type boolean
|
||||
*/
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
import { useCallback, useRef } from 'react';
|
||||
import { useCallback, useMemo, useRef } from 'react';
|
||||
import ChartLayout from 'container/DashboardContainer/visualization/layout/ChartLayout/ChartLayout';
|
||||
import Legend from 'lib/uPlotV2/components/Legend/Legend';
|
||||
import {
|
||||
@@ -30,6 +30,7 @@ export default function ChartWrapper({
|
||||
onDestroy = noop,
|
||||
children,
|
||||
layoutChildren,
|
||||
yAxisUnit,
|
||||
customTooltip,
|
||||
pinnedTooltipElement,
|
||||
'data-testid': testId,
|
||||
@@ -62,6 +63,13 @@ export default function ChartWrapper({
|
||||
[customTooltip],
|
||||
);
|
||||
|
||||
const syncMetadata = useMemo(
|
||||
() => ({
|
||||
yAxisUnit,
|
||||
}),
|
||||
[yAxisUnit],
|
||||
);
|
||||
|
||||
return (
|
||||
<PlotContextProvider>
|
||||
<ChartLayout
|
||||
@@ -99,6 +107,7 @@ export default function ChartWrapper({
|
||||
averageLegendWidth + TOOLTIP_WIDTH_PADDING,
|
||||
)}
|
||||
syncKey={syncKey}
|
||||
syncMetadata={syncMetadata}
|
||||
render={renderTooltipCallback}
|
||||
pinnedTooltipElement={pinnedTooltipElement}
|
||||
/>
|
||||
|
||||
@@ -24,13 +24,12 @@ export default function Histogram(props: HistogramChartProps): JSX.Element {
|
||||
}
|
||||
const tooltipProps: HistogramTooltipProps = {
|
||||
...props,
|
||||
timezone: rest.timezone,
|
||||
yAxisUnit: rest.yAxisUnit,
|
||||
decimalPrecision: rest.decimalPrecision,
|
||||
};
|
||||
return <HistogramTooltip {...tooltipProps} />;
|
||||
},
|
||||
[customTooltip, rest.timezone, rest.yAxisUnit, rest.decimalPrecision],
|
||||
[customTooltip, rest.yAxisUnit, rest.decimalPrecision],
|
||||
);
|
||||
|
||||
return (
|
||||
|
||||
@@ -12,10 +12,7 @@ interface BaseChartProps {
|
||||
height: number;
|
||||
showTooltip?: boolean;
|
||||
showLegend?: boolean;
|
||||
timezone?: Timezone;
|
||||
canPinTooltip?: boolean;
|
||||
yAxisUnit?: string;
|
||||
decimalPrecision?: PrecisionOption;
|
||||
pinnedTooltipElement?: (clickData: TooltipClickData) => React.ReactNode;
|
||||
customTooltip?: (props: TooltipRenderArgs) => React.ReactNode;
|
||||
'data-testid'?: string;
|
||||
@@ -32,18 +29,31 @@ interface UPlotBasedChartProps {
|
||||
layoutChildren?: React.ReactNode;
|
||||
}
|
||||
|
||||
interface UPlotChartDataProps {
|
||||
yAxisUnit?: string;
|
||||
decimalPrecision?: PrecisionOption;
|
||||
}
|
||||
|
||||
export interface TimeSeriesChartProps
|
||||
extends BaseChartProps,
|
||||
UPlotBasedChartProps {}
|
||||
UPlotBasedChartProps,
|
||||
UPlotChartDataProps {
|
||||
timezone?: Timezone;
|
||||
}
|
||||
|
||||
export interface HistogramChartProps
|
||||
extends BaseChartProps,
|
||||
UPlotBasedChartProps {
|
||||
UPlotBasedChartProps,
|
||||
UPlotChartDataProps {
|
||||
isQueriesMerged?: boolean;
|
||||
}
|
||||
|
||||
export interface BarChartProps extends BaseChartProps, UPlotBasedChartProps {
|
||||
export interface BarChartProps
|
||||
extends BaseChartProps,
|
||||
UPlotBasedChartProps,
|
||||
UPlotChartDataProps {
|
||||
isStackedBarChart?: boolean;
|
||||
timezone?: Timezone;
|
||||
}
|
||||
|
||||
export type ChartProps =
|
||||
|
||||
@@ -123,13 +123,13 @@ function BarPanel(props: PanelWrapperProps): JSX.Element {
|
||||
}}
|
||||
plotRef={onPlotRef}
|
||||
onDestroy={onPlotDestroy}
|
||||
yAxisUnit={widget.yAxisUnit}
|
||||
decimalPrecision={widget.decimalPrecision}
|
||||
data={chartData as uPlot.AlignedData}
|
||||
width={containerDimensions.width}
|
||||
height={containerDimensions.height}
|
||||
layoutChildren={layoutChildren}
|
||||
isStackedBarChart={widget.stackedBarChart ?? false}
|
||||
yAxisUnit={widget.yAxisUnit}
|
||||
decimalPrecision={widget.decimalPrecision}
|
||||
timezone={timezone}
|
||||
>
|
||||
<ContextMenu
|
||||
|
||||
@@ -3,8 +3,6 @@ import { PanelWrapperProps } from 'container/PanelWrapper/panelWrapper.types';
|
||||
import { useIsDarkMode } from 'hooks/useDarkMode';
|
||||
import { useResizeObserver } from 'hooks/useDimensions';
|
||||
import { LegendPosition } from 'lib/uPlotV2/components/types';
|
||||
import { DashboardCursorSync } from 'lib/uPlotV2/plugins/TooltipPlugin/types';
|
||||
import { useTimezone } from 'providers/Timezone';
|
||||
import uPlot from 'uplot';
|
||||
|
||||
import Histogram from '../../charts/Histogram/Histogram';
|
||||
@@ -29,7 +27,6 @@ function HistogramPanel(props: PanelWrapperProps): JSX.Element {
|
||||
const containerDimensions = useResizeObserver(graphRef);
|
||||
|
||||
const isDarkMode = useIsDarkMode();
|
||||
const { timezone } = useTimezone();
|
||||
|
||||
const config = useMemo(() => {
|
||||
return prepareHistogramPanelConfig({
|
||||
@@ -92,11 +89,9 @@ function HistogramPanel(props: PanelWrapperProps): JSX.Element {
|
||||
onDestroy={(): void => {
|
||||
uPlotRef.current = null;
|
||||
}}
|
||||
isQueriesMerged={widget.mergeAllActiveQueries}
|
||||
yAxisUnit={widget.yAxisUnit}
|
||||
decimalPrecision={widget.decimalPrecision}
|
||||
syncMode={DashboardCursorSync.Crosshair}
|
||||
timezone={timezone}
|
||||
isQueriesMerged={widget.mergeAllActiveQueries}
|
||||
data={chartData as uPlot.AlignedData}
|
||||
width={containerDimensions.width}
|
||||
height={containerDimensions.height}
|
||||
|
||||
@@ -48,8 +48,8 @@ jest.mock(
|
||||
{JSON.stringify({
|
||||
legendPosition: props.legendConfig?.position,
|
||||
isQueriesMerged: props.isQueriesMerged,
|
||||
yAxisUnit: props.yAxisUnit,
|
||||
decimalPrecision: props.decimalPrecision,
|
||||
yAxisUnit: props?.yAxisUnit,
|
||||
decimalPrecision: props?.decimalPrecision,
|
||||
})}
|
||||
</div>
|
||||
{props.layoutChildren}
|
||||
|
||||
@@ -112,9 +112,9 @@ function TimeSeriesPanel(props: PanelWrapperProps): JSX.Element {
|
||||
legendConfig={{
|
||||
position: widget?.legendPosition ?? LegendPosition.BOTTOM,
|
||||
}}
|
||||
timezone={timezone}
|
||||
yAxisUnit={widget.yAxisUnit}
|
||||
decimalPrecision={widget.decimalPrecision}
|
||||
timezone={timezone}
|
||||
data={chartData as uPlot.AlignedData}
|
||||
width={containerDimensions.width}
|
||||
height={containerDimensions.height}
|
||||
|
||||
@@ -62,10 +62,10 @@ export interface TooltipRenderArgs {
|
||||
|
||||
export interface BaseTooltipProps {
|
||||
showTooltipHeader?: boolean;
|
||||
timezone?: Timezone;
|
||||
yAxisUnit?: string;
|
||||
decimalPrecision?: PrecisionOption;
|
||||
content?: TooltipContentItem[];
|
||||
timezone?: Timezone;
|
||||
}
|
||||
|
||||
export interface TimeSeriesTooltipProps
|
||||
|
||||
@@ -4,6 +4,7 @@ import cx from 'classnames';
|
||||
import { getFocusedSeriesAtPosition } from 'lib/uPlotLib/plugins/onClickPlugin';
|
||||
import uPlot from 'uplot';
|
||||
|
||||
import { syncCursorRegistry } from './syncCursorRegistry';
|
||||
import {
|
||||
createInitialControllerState,
|
||||
createSetCursorHandler,
|
||||
@@ -40,6 +41,7 @@ export default function TooltipPlugin({
|
||||
maxHeight = 600,
|
||||
syncMode = DashboardCursorSync.None,
|
||||
syncKey = '_tooltip_sync_global_',
|
||||
syncMetadata,
|
||||
pinnedTooltipElement,
|
||||
canPinTooltip = false,
|
||||
}: TooltipPluginProps): JSX.Element | null {
|
||||
@@ -100,7 +102,29 @@ export default function TooltipPlugin({
|
||||
// crosshair / tooltip can follow the dashboard-wide cursor.
|
||||
if (syncMode !== DashboardCursorSync.None && config.scales[0]?.props.time) {
|
||||
config.setCursor({
|
||||
sync: { key: syncKey, scales: ['x', null] },
|
||||
sync: { key: syncKey, scales: ['x', 'y'] },
|
||||
});
|
||||
|
||||
// Show the horizontal crosshair only when the receiving panel shares
|
||||
// the same y-axis unit as the source panel. When this panel is the
|
||||
// source (cursor.event != null) the line is always shown and this
|
||||
// panel's metadata is written to the registry so receivers can read it.
|
||||
config.addHook('setCursor', (u: uPlot): void => {
|
||||
const yCursorEl = u.root.querySelector<HTMLElement>('.u-cursor-y');
|
||||
if (!yCursorEl) {
|
||||
return;
|
||||
}
|
||||
|
||||
if (u.cursor.event != null) {
|
||||
// This panel is the source — publish metadata and always show line.
|
||||
syncCursorRegistry.setMetadata(syncKey, syncMetadata);
|
||||
yCursorEl.style.display = '';
|
||||
} else {
|
||||
// This panel is receiving sync — show only if units match.
|
||||
const sourceMeta = syncCursorRegistry.getMetadata(syncKey);
|
||||
yCursorEl.style.display =
|
||||
sourceMeta?.yAxisUnit === syncMetadata?.yAxisUnit ? '' : 'none';
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
|
||||
@@ -0,0 +1,24 @@
|
||||
import type { TooltipSyncMetadata } from './types';
|
||||
|
||||
/**
|
||||
* Module-level registry that tracks the metadata of the panel currently
|
||||
* acting as the cursor source (the one being hovered) per sync group.
|
||||
*
|
||||
* uPlot fires the source panel's setCursor hook before broadcasting to
|
||||
* receivers, so the registry is always populated before receivers read it.
|
||||
*
|
||||
* Receivers use this to make decisions such as:
|
||||
* - Whether to show the horizontal crosshair line (matching yAxisUnit)
|
||||
* - Future: what to render inside the tooltip (matching groupBy, etc.)
|
||||
*/
|
||||
const metadataBySyncKey = new Map<string, TooltipSyncMetadata | undefined>();
|
||||
|
||||
export const syncCursorRegistry = {
|
||||
setMetadata(syncKey: string, metadata: TooltipSyncMetadata | undefined): void {
|
||||
metadataBySyncKey.set(syncKey, metadata);
|
||||
},
|
||||
|
||||
getMetadata(syncKey: string): TooltipSyncMetadata | undefined {
|
||||
return metadataBySyncKey.get(syncKey);
|
||||
},
|
||||
};
|
||||
@@ -34,11 +34,16 @@ export interface TooltipLayoutInfo {
|
||||
height: number;
|
||||
}
|
||||
|
||||
export interface TooltipSyncMetadata {
|
||||
yAxisUnit?: string;
|
||||
}
|
||||
|
||||
export interface TooltipPluginProps {
|
||||
config: UPlotConfigBuilder;
|
||||
canPinTooltip?: boolean;
|
||||
syncMode?: DashboardCursorSync;
|
||||
syncKey?: string;
|
||||
syncMetadata?: TooltipSyncMetadata;
|
||||
render: (args: TooltipRenderArgs) => ReactNode;
|
||||
pinnedTooltipElement?: (clickData: TooltipClickData) => ReactNode;
|
||||
maxWidth?: number;
|
||||
|
||||
@@ -516,7 +516,7 @@ describe('TooltipPlugin', () => {
|
||||
);
|
||||
|
||||
expect(setCursorSpy).toHaveBeenCalledWith({
|
||||
sync: { key: 'dashboard-sync', scales: ['x', null] },
|
||||
sync: { key: 'dashboard-sync', scales: ['x', 'y'] },
|
||||
});
|
||||
});
|
||||
|
||||
|
||||
@@ -44,6 +44,7 @@ func (provider *provider) addRulerRoutes(router *mux.Router) error {
|
||||
Description: "This endpoint creates a new alert rule",
|
||||
Request: new(ruletypes.PostableRule),
|
||||
RequestContentType: "application/json",
|
||||
RequestExamples: postableRuleExamples(),
|
||||
Response: new(ruletypes.Rule),
|
||||
ResponseContentType: "application/json",
|
||||
SuccessStatusCode: http.StatusCreated,
|
||||
@@ -54,27 +55,28 @@ func (provider *provider) addRulerRoutes(router *mux.Router) error {
|
||||
}
|
||||
|
||||
if err := router.Handle("/api/v2/rules/{id}", handler.New(provider.authZ.EditAccess(provider.rulerHandler.UpdateRuleByID), handler.OpenAPIDef{
|
||||
ID: "UpdateRuleByID",
|
||||
Tags: []string{"rules"},
|
||||
Summary: "Update alert rule",
|
||||
Description: "This endpoint updates an alert rule by ID",
|
||||
Request: new(ruletypes.PostableRule),
|
||||
RequestContentType: "application/json",
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusBadRequest, http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
ID: "UpdateRuleByID",
|
||||
Tags: []string{"rules"},
|
||||
Summary: "Update alert rule",
|
||||
Description: "This endpoint updates an alert rule by ID",
|
||||
Request: new(ruletypes.PostableRule),
|
||||
RequestContentType: "application/json",
|
||||
RequestExamples: postableRuleExamples(),
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusBadRequest, http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
})).Methods(http.MethodPut).GetError(); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
if err := router.Handle("/api/v2/rules/{id}", handler.New(provider.authZ.EditAccess(provider.rulerHandler.DeleteRuleByID), handler.OpenAPIDef{
|
||||
ID: "DeleteRuleByID",
|
||||
Tags: []string{"rules"},
|
||||
Summary: "Delete alert rule",
|
||||
Description: "This endpoint deletes an alert rule by ID",
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
ID: "DeleteRuleByID",
|
||||
Tags: []string{"rules"},
|
||||
Summary: "Delete alert rule",
|
||||
Description: "This endpoint deletes an alert rule by ID",
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
})).Methods(http.MethodDelete).GetError(); err != nil {
|
||||
return err
|
||||
}
|
||||
@@ -86,6 +88,7 @@ func (provider *provider) addRulerRoutes(router *mux.Router) error {
|
||||
Description: "This endpoint applies a partial update to an alert rule by ID",
|
||||
Request: new(ruletypes.PostableRule),
|
||||
RequestContentType: "application/json",
|
||||
RequestExamples: postableRuleExamples(),
|
||||
Response: new(ruletypes.Rule),
|
||||
ResponseContentType: "application/json",
|
||||
SuccessStatusCode: http.StatusOK,
|
||||
@@ -102,6 +105,7 @@ func (provider *provider) addRulerRoutes(router *mux.Router) error {
|
||||
Description: "This endpoint fires a test notification for the given rule definition",
|
||||
Request: new(ruletypes.PostableRule),
|
||||
RequestContentType: "application/json",
|
||||
RequestExamples: postableRuleExamples(),
|
||||
Response: new(ruletypes.GettableTestRule),
|
||||
ResponseContentType: "application/json",
|
||||
SuccessStatusCode: http.StatusOK,
|
||||
@@ -156,27 +160,27 @@ func (provider *provider) addRulerRoutes(router *mux.Router) error {
|
||||
}
|
||||
|
||||
if err := router.Handle("/api/v1/downtime_schedules/{id}", handler.New(provider.authZ.EditAccess(provider.rulerHandler.UpdateDowntimeScheduleByID), handler.OpenAPIDef{
|
||||
ID: "UpdateDowntimeScheduleByID",
|
||||
Tags: []string{"downtimeschedules"},
|
||||
Summary: "Update downtime schedule",
|
||||
Description: "This endpoint updates a downtime schedule by ID",
|
||||
Request: new(ruletypes.PostablePlannedMaintenance),
|
||||
RequestContentType: "application/json",
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusBadRequest, http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
ID: "UpdateDowntimeScheduleByID",
|
||||
Tags: []string{"downtimeschedules"},
|
||||
Summary: "Update downtime schedule",
|
||||
Description: "This endpoint updates a downtime schedule by ID",
|
||||
Request: new(ruletypes.PostablePlannedMaintenance),
|
||||
RequestContentType: "application/json",
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusBadRequest, http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
})).Methods(http.MethodPut).GetError(); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
if err := router.Handle("/api/v1/downtime_schedules/{id}", handler.New(provider.authZ.EditAccess(provider.rulerHandler.DeleteDowntimeScheduleByID), handler.OpenAPIDef{
|
||||
ID: "DeleteDowntimeScheduleByID",
|
||||
Tags: []string{"downtimeschedules"},
|
||||
Summary: "Delete downtime schedule",
|
||||
Description: "This endpoint deletes a downtime schedule by ID",
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
ID: "DeleteDowntimeScheduleByID",
|
||||
Tags: []string{"downtimeschedules"},
|
||||
Summary: "Delete downtime schedule",
|
||||
Description: "This endpoint deletes a downtime schedule by ID",
|
||||
SuccessStatusCode: http.StatusNoContent,
|
||||
ErrorStatusCodes: []int{http.StatusNotFound},
|
||||
SecuritySchemes: newSecuritySchemes(types.RoleEditor),
|
||||
})).Methods(http.MethodDelete).GetError(); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
733
pkg/apiserver/signozapiserver/ruler_examples.go
Normal file
733
pkg/apiserver/signozapiserver/ruler_examples.go
Normal file
@@ -0,0 +1,733 @@
|
||||
package signozapiserver
|
||||
|
||||
import "github.com/SigNoz/signoz/pkg/http/handler"
|
||||
|
||||
// postableRuleExamples returns example payloads attached to every rule-write
|
||||
// endpoint. They cover each alert type, rule type, and composite-query shape.
|
||||
func postableRuleExamples() []handler.OpenAPIExample {
|
||||
rolling := func(evalWindow, frequency string) map[string]any {
|
||||
return map[string]any{
|
||||
"kind": "rolling",
|
||||
"spec": map[string]any{"evalWindow": evalWindow, "frequency": frequency},
|
||||
}
|
||||
}
|
||||
renotify := func(interval string, states ...string) map[string]any {
|
||||
s := make([]any, 0, len(states))
|
||||
for _, v := range states {
|
||||
s = append(s, v)
|
||||
}
|
||||
return map[string]any{
|
||||
"enabled": true,
|
||||
"interval": interval,
|
||||
"alertStates": s,
|
||||
}
|
||||
}
|
||||
|
||||
return []handler.OpenAPIExample{
|
||||
{
|
||||
Name: "metric_threshold_single",
|
||||
Summary: "Metric threshold single builder query",
|
||||
Description: "Fires when a pod consumes more than 80% of its requested CPU for the whole evaluation window. Uses `k8s.pod.cpu_request_utilization`.",
|
||||
Value: map[string]any{
|
||||
"alert": "Pod CPU above 80% of request",
|
||||
"alertType": "METRIC_BASED_ALERT",
|
||||
"description": "CPU usage for api-service pods exceeds 80% of the requested CPU",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"unit": "percentunit",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "metrics",
|
||||
"stepInterval": 60,
|
||||
"aggregations": []any{map[string]any{"metricName": "k8s.pod.cpu_request_utilization", "timeAggregation": "avg", "spaceAggregation": "max"}},
|
||||
"filter": map[string]any{"expression": "k8s.deployment.name = 'api-service'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "k8s.pod.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "deployment.environment", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
},
|
||||
"legend": "{{k8s.pod.name}} ({{deployment.environment}})",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "A",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "all_the_times",
|
||||
"target": 0.8,
|
||||
"channels": []any{"slack-platform", "pagerduty-oncall"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("15m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"k8s.pod.name", "deployment.environment"},
|
||||
"renotify": renotify("4h", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"severity": "critical", "team": "platform"},
|
||||
"annotations": map[string]any{
|
||||
"description": "Pod {{$k8s.pod.name}} CPU is at {{$value}} of request in {{$deployment.environment}}.",
|
||||
"summary": "Pod CPU above {{$threshold}} of request",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "metric_threshold_formula",
|
||||
Summary: "Metric threshold multi-query formula",
|
||||
Description: "Computes disk utilization as (1 - available/capacity) * 100 by combining two disabled base queries with a builder_formula. The formula emits 0–100, so compositeQuery.unit is set to \"percent\" and the target is a bare number.",
|
||||
Value: map[string]any{
|
||||
"alert": "PersistentVolume above 80% utilization",
|
||||
"alertType": "METRIC_BASED_ALERT",
|
||||
"description": "Disk utilization for a persistent volume is above 80%",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"unit": "percent",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "metrics",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"metricName": "k8s.volume.available", "timeAggregation": "max", "spaceAggregation": "max"}},
|
||||
"filter": map[string]any{"expression": "k8s.volume.type = 'persistentVolumeClaim'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "k8s.persistentvolumeclaim.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "k8s.namespace.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "B",
|
||||
"signal": "metrics",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"metricName": "k8s.volume.capacity", "timeAggregation": "max", "spaceAggregation": "max"}},
|
||||
"filter": map[string]any{"expression": "k8s.volume.type = 'persistentVolumeClaim'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "k8s.persistentvolumeclaim.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "k8s.namespace.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_formula",
|
||||
"spec": map[string]any{
|
||||
"name": "F1",
|
||||
"expression": "(1 - A/B) * 100",
|
||||
"legend": "{{k8s.persistentvolumeclaim.name}} ({{k8s.namespace.name}})",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "F1",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "at_least_once",
|
||||
"target": 80,
|
||||
"channels": []any{"slack-storage"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("30m", "5m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"k8s.namespace.name", "k8s.persistentvolumeclaim.name"},
|
||||
"renotify": renotify("2h", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"severity": "critical"},
|
||||
"annotations": map[string]any{
|
||||
"description": "Volume {{$k8s.persistentvolumeclaim.name}} in {{$k8s.namespace.name}} is {{$value}}% full.",
|
||||
"summary": "Disk utilization above {{$threshold}}%",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "metric_promql",
|
||||
Summary: "Metric threshold PromQL rule",
|
||||
Description: "PromQL expression instead of the builder. Dotted OTEL resource attributes are quoted (\"deployment.environment\"). Useful for queries that combine series with group_right or other Prom operators.",
|
||||
Value: map[string]any{
|
||||
"alert": "Kafka consumer group lag above 1000",
|
||||
"alertType": "METRIC_BASED_ALERT",
|
||||
"description": "Consumer group lag computed via PromQL",
|
||||
"ruleType": "promql_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "promql",
|
||||
"panelType": "graph",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "promql",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"query": "(max by(topic, partition, \"deployment.environment\")(kafka_log_end_offset) - on(topic, partition, \"deployment.environment\") group_right max by(group, topic, partition, \"deployment.environment\")(kafka_consumer_committed_offset)) > 0",
|
||||
"legend": "{{topic}}/{{partition}} ({{group}})",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "A",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "all_the_times",
|
||||
"target": 1000,
|
||||
"channels": []any{"slack-data-platform", "pagerduty-data"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("10m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"group", "topic"},
|
||||
"renotify": renotify("1h", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"severity": "critical"},
|
||||
"annotations": map[string]any{
|
||||
"description": "Consumer group {{$group}} is {{$value}} messages behind on {{$topic}}/{{$partition}}.",
|
||||
"summary": "Kafka consumer lag high",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "metric_anomaly",
|
||||
Summary: "Metric anomaly rule (v1 only)",
|
||||
Description: "Anomaly rules are not yet supported under schemaVersion v2alpha1, so this example uses the v1 shape. Wraps a builder query in the `anomaly` function with daily seasonality SigNoz compares each point against the forecast for that time of day. Fires when the anomaly score stays below the threshold for the entire window; `requireMinPoints` guards against noisy intervals.",
|
||||
Value: map[string]any{
|
||||
"alert": "Anomalous drop in ingested spans",
|
||||
"alertType": "METRIC_BASED_ALERT",
|
||||
"description": "Detect an abrupt drop in span ingestion using a z-score anomaly function",
|
||||
"ruleType": "anomaly_rule",
|
||||
"version": "v5",
|
||||
"evalWindow": "24h",
|
||||
"frequency": "3h",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "metrics",
|
||||
"stepInterval": 21600,
|
||||
"aggregations": []any{map[string]any{"metricName": "otelcol_receiver_accepted_spans", "timeAggregation": "rate", "spaceAggregation": "sum"}},
|
||||
"filter": map[string]any{"expression": "tenant_tier = 'premium'"},
|
||||
"groupBy": []any{map[string]any{"name": "tenant_id", "fieldContext": "attribute", "fieldDataType": "string"}},
|
||||
"functions": []any{
|
||||
map[string]any{
|
||||
"name": "anomaly",
|
||||
"args": []any{map[string]any{"name": "z_score_threshold", "value": 2}},
|
||||
},
|
||||
},
|
||||
"legend": "{{tenant_id}}",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"op": "below",
|
||||
"matchType": "all_the_times",
|
||||
"target": 2,
|
||||
"algorithm": "standard",
|
||||
"seasonality": "daily",
|
||||
"selectedQueryName": "A",
|
||||
"requireMinPoints": true,
|
||||
"requiredNumPoints": 3,
|
||||
},
|
||||
"labels": map[string]any{"severity": "warning"},
|
||||
"preferredChannels": []any{"slack-ingestion"},
|
||||
"annotations": map[string]any{
|
||||
"description": "Ingestion rate for tenant {{$tenant_id}} is anomalously low (z-score {{$value}}).",
|
||||
"summary": "Span ingestion anomaly",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "logs_threshold",
|
||||
Summary: "Logs threshold count() over filter",
|
||||
Description: "Counts matching log records (ERROR severity + body contains) over a rolling window. Fires at least once per evaluation when the count exceeds zero.",
|
||||
Value: map[string]any{
|
||||
"alert": "Payments service panic logs",
|
||||
"alertType": "LOGS_BASED_ALERT",
|
||||
"description": "Any panic log line emitted by the payments service",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "logs",
|
||||
"stepInterval": 60,
|
||||
"aggregations": []any{map[string]any{"expression": "count()"}},
|
||||
"filter": map[string]any{"expression": "service.name = 'payments-api' AND severity_text = 'ERROR' AND body CONTAINS 'panic'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "k8s.pod.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "deployment.environment", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
},
|
||||
"legend": "{{k8s.pod.name}} ({{deployment.environment}})",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "A",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "at_least_once",
|
||||
"target": 0,
|
||||
"channels": []any{"slack-payments", "pagerduty-payments"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("5m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"k8s.pod.name", "deployment.environment"},
|
||||
"renotify": renotify("15m", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"severity": "critical", "team": "payments"},
|
||||
"annotations": map[string]any{
|
||||
"description": "{{$k8s.pod.name}} emitted {{$value}} panic log(s) in {{$deployment.environment}}.",
|
||||
"summary": "Payments service panic",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "logs_error_rate_formula",
|
||||
Summary: "Logs error rate error count / total count × 100",
|
||||
Description: "Two disabled log count queries (A = errors, B = total) combined via a builder_formula into a percentage. Classic service-level error-rate alert pattern for log-based signals.",
|
||||
Value: map[string]any{
|
||||
"alert": "Payments-api error log rate above 1%",
|
||||
"alertType": "LOGS_BASED_ALERT",
|
||||
"description": "Error log ratio as a percentage of total logs for payments-api",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"unit": "percent",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "logs",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"expression": "count()"}},
|
||||
"filter": map[string]any{"expression": "service.name = 'payments-api' AND severity_text IN ['ERROR', 'FATAL']"},
|
||||
"groupBy": []any{map[string]any{"name": "deployment.environment", "fieldContext": "resource", "fieldDataType": "string"}},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "B",
|
||||
"signal": "logs",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"expression": "count()"}},
|
||||
"filter": map[string]any{"expression": "service.name = 'payments-api'"},
|
||||
"groupBy": []any{map[string]any{"name": "deployment.environment", "fieldContext": "resource", "fieldDataType": "string"}},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_formula",
|
||||
"spec": map[string]any{
|
||||
"name": "F1",
|
||||
"expression": "(A / B) * 100",
|
||||
"legend": "{{deployment.environment}}",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "F1",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "at_least_once",
|
||||
"target": 1,
|
||||
"channels": []any{"slack-payments"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("5m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"deployment.environment"},
|
||||
"renotify": renotify("30m", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"severity": "critical", "team": "payments"},
|
||||
"annotations": map[string]any{
|
||||
"description": "Error log rate in {{$deployment.environment}} is {{$value}}%",
|
||||
"summary": "Payments-api error rate above {{$threshold}}%",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "traces_threshold_latency",
|
||||
Summary: "Traces threshold p99 latency (ns → s conversion)",
|
||||
Description: "Builder query against the traces signal with p99(duration_nano). The series unit is ns (compositeQuery.unit), the target is in seconds (threshold.targetUnit) SigNoz converts before comparing. Canonical shape when series and target live in different units.",
|
||||
Value: map[string]any{
|
||||
"alert": "Search API p99 latency above 5s",
|
||||
"alertType": "TRACES_BASED_ALERT",
|
||||
"description": "p99 duration of the search endpoint exceeds 5s",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"unit": "ns",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "traces",
|
||||
"stepInterval": 60,
|
||||
"aggregations": []any{map[string]any{"expression": "p99(duration_nano)"}},
|
||||
"filter": map[string]any{"expression": "service.name = 'search-api' AND name = 'GET /api/v1/search'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "service.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "http.route", "fieldContext": "attribute", "fieldDataType": "string"},
|
||||
},
|
||||
"legend": "{{service.name}} {{http.route}}",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "A",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "warning",
|
||||
"op": "above",
|
||||
"matchType": "at_least_once",
|
||||
"target": 5,
|
||||
"targetUnit": "s",
|
||||
"channels": []any{"slack-search"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("5m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"service.name", "http.route"},
|
||||
"renotify": renotify("30m", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"severity": "warning", "team": "search"},
|
||||
"annotations": map[string]any{
|
||||
"description": "p99 latency for {{$service.name}} on {{$http.route}} crossed {{$threshold}}s.",
|
||||
"summary": "Search-api latency degraded",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "traces_error_rate_formula",
|
||||
Summary: "Traces error rate error spans / total spans × 100",
|
||||
Description: "Two disabled trace count queries (A = error spans where hasError=true, B = total spans) combined via a builder_formula into a percentage. Mirrors the common request-error-rate dashboard shape.",
|
||||
Value: map[string]any{
|
||||
"alert": "Search-api error rate above 5%",
|
||||
"alertType": "TRACES_BASED_ALERT",
|
||||
"description": "Request error rate for search-api, grouped by route",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"unit": "percent",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "traces",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"expression": "count()"}},
|
||||
"filter": map[string]any{"expression": "service.name = 'search-api' AND hasError = true"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "service.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "http.route", "fieldContext": "attribute", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "B",
|
||||
"signal": "traces",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"expression": "count()"}},
|
||||
"filter": map[string]any{"expression": "service.name = 'search-api'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "service.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "http.route", "fieldContext": "attribute", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_formula",
|
||||
"spec": map[string]any{
|
||||
"name": "F1",
|
||||
"expression": "(A / B) * 100",
|
||||
"legend": "{{service.name}} {{http.route}}",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "F1",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "at_least_once",
|
||||
"target": 5,
|
||||
"channels": []any{"slack-search", "pagerduty-search"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("5m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"service.name", "http.route"},
|
||||
"renotify": renotify("15m", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"severity": "critical", "team": "search"},
|
||||
"annotations": map[string]any{
|
||||
"description": "Error rate on {{$service.name}} {{$http.route}} is {{$value}}%",
|
||||
"summary": "Search-api error rate above {{$threshold}}%",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "tiered_thresholds",
|
||||
Summary: "Tiered thresholds with per-tier channels",
|
||||
Description: "Two tiers (warning and critical) in a single rule, each with its own target, op, matchType, and channels so warnings and pages route to different receivers. `alertOnAbsent` + `absentFor` fires a no-data alert when the query returns no series for 15 consecutive evaluations.",
|
||||
Value: map[string]any{
|
||||
"alert": "Kafka consumer lag warn / critical",
|
||||
"alertType": "METRIC_BASED_ALERT",
|
||||
"description": "Warn at lag ≥ 50 and page at ≥ 200, tiered via thresholds.spec.",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "metrics",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"metricName": "kafka_log_end_offset", "timeAggregation": "max", "spaceAggregation": "max"}},
|
||||
"filter": map[string]any{"expression": "topic != '__consumer_offsets'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "topic", "fieldContext": "attribute", "fieldDataType": "string"},
|
||||
map[string]any{"name": "partition", "fieldContext": "attribute", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "B",
|
||||
"signal": "metrics",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"metricName": "kafka_consumer_committed_offset", "timeAggregation": "max", "spaceAggregation": "max"}},
|
||||
"filter": map[string]any{"expression": "topic != '__consumer_offsets'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "topic", "fieldContext": "attribute", "fieldDataType": "string"},
|
||||
map[string]any{"name": "partition", "fieldContext": "attribute", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_formula",
|
||||
"spec": map[string]any{
|
||||
"name": "F1",
|
||||
"expression": "A - B",
|
||||
"legend": "{{topic}}/{{partition}}",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"alertOnAbsent": true,
|
||||
"absentFor": 15,
|
||||
"selectedQueryName": "F1",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "warning",
|
||||
"op": "above",
|
||||
"matchType": "all_the_times",
|
||||
"target": 50,
|
||||
"channels": []any{"slack-kafka-info"},
|
||||
},
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "all_the_times",
|
||||
"target": 200,
|
||||
"channels": []any{"slack-kafka-alerts", "pagerduty-kafka"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("5m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"topic"},
|
||||
"renotify": renotify("15m", "firing"),
|
||||
},
|
||||
"labels": map[string]any{"team": "data-platform"},
|
||||
"annotations": map[string]any{
|
||||
"description": "Consumer lag for {{$topic}} partition {{$partition}} is {{$value}}.",
|
||||
"summary": "Kafka consumer lag",
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "notification_settings",
|
||||
Summary: "Full notification settings (grouping, nodata renotify, grace period)",
|
||||
Description: "Demonstrates the full notificationSettings surface: `groupBy` merges alerts across labels to cut noise, `newGroupEvalDelay` gives newly-appearing series a grace period before firing, `renotify` re-alerts every 30m while firing OR while the alert is in nodata (missing data is treated as actionable), and `usePolicy: false` means channels come from the threshold entries rather than global routing policies. Set `usePolicy: true` to skip per-threshold channels and route via the org-level notification policy instead.",
|
||||
Value: map[string]any{
|
||||
"alert": "API 5xx error rate above 1%",
|
||||
"alertType": "TRACES_BASED_ALERT",
|
||||
"description": "Noise-controlled 5xx error rate alert with renotify on gaps",
|
||||
"ruleType": "threshold_rule",
|
||||
"version": "v5",
|
||||
"schemaVersion": "v2alpha1",
|
||||
"condition": map[string]any{
|
||||
"compositeQuery": map[string]any{
|
||||
"queryType": "builder",
|
||||
"panelType": "graph",
|
||||
"unit": "percent",
|
||||
"queries": []any{
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "A",
|
||||
"signal": "traces",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"expression": "count()"}},
|
||||
"filter": map[string]any{"expression": "service.name CONTAINS 'api' AND http.status_code >= 500"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "service.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "deployment.environment", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_query",
|
||||
"spec": map[string]any{
|
||||
"name": "B",
|
||||
"signal": "traces",
|
||||
"stepInterval": 60,
|
||||
"disabled": true,
|
||||
"aggregations": []any{map[string]any{"expression": "count()"}},
|
||||
"filter": map[string]any{"expression": "service.name CONTAINS 'api'"},
|
||||
"groupBy": []any{
|
||||
map[string]any{"name": "service.name", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
map[string]any{"name": "deployment.environment", "fieldContext": "resource", "fieldDataType": "string"},
|
||||
},
|
||||
},
|
||||
},
|
||||
map[string]any{
|
||||
"type": "builder_formula",
|
||||
"spec": map[string]any{
|
||||
"name": "F1",
|
||||
"expression": "(A / B) * 100",
|
||||
"legend": "{{service.name}} ({{deployment.environment}})",
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"selectedQueryName": "F1",
|
||||
"thresholds": map[string]any{
|
||||
"kind": "basic",
|
||||
"spec": []any{
|
||||
map[string]any{
|
||||
"name": "critical",
|
||||
"op": "above",
|
||||
"matchType": "at_least_once",
|
||||
"target": 1,
|
||||
"channels": []any{"slack-api-alerts", "pagerduty-oncall"},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
"evaluation": rolling("5m", "1m"),
|
||||
"notificationSettings": map[string]any{
|
||||
"groupBy": []any{"service.name", "deployment.environment"},
|
||||
"newGroupEvalDelay": "2m",
|
||||
"usePolicy": false,
|
||||
"renotify": renotify("30m", "firing", "nodata"),
|
||||
},
|
||||
"labels": map[string]any{"team": "platform"},
|
||||
"annotations": map[string]any{
|
||||
"description": "{{$service.name}} 5xx rate in {{$deployment.environment}} is {{$value}}%.",
|
||||
"summary": "API service error rate elevated",
|
||||
},
|
||||
},
|
||||
},
|
||||
}
|
||||
}
|
||||
32
pkg/apiserver/signozapiserver/ruler_examples_test.go
Normal file
32
pkg/apiserver/signozapiserver/ruler_examples_test.go
Normal file
@@ -0,0 +1,32 @@
|
||||
package signozapiserver
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"testing"
|
||||
|
||||
"github.com/SigNoz/signoz/pkg/types/ruletypes"
|
||||
)
|
||||
|
||||
// TestPostableRuleExamplesValidate verifies every example payload returned by
|
||||
// postableRuleExamples() round-trips through PostableRule.UnmarshalJSON and
|
||||
// passes Validate(). If an example drifts from the runtime contract this
|
||||
// breaks loudly so the spec doesn't ship invalid payloads to users.
|
||||
func TestPostableRuleExamplesValidate(t *testing.T) {
|
||||
for _, example := range postableRuleExamples() {
|
||||
t.Run(example.Name, func(t *testing.T) {
|
||||
raw, err := json.Marshal(example.Value)
|
||||
if err != nil {
|
||||
t.Fatalf("marshal example: %v", err)
|
||||
}
|
||||
|
||||
var rule ruletypes.PostableRule
|
||||
if err := json.Unmarshal(raw, &rule); err != nil {
|
||||
t.Fatalf("unmarshal: %v\npayload: %s", err, raw)
|
||||
}
|
||||
|
||||
if err := rule.Validate(); err != nil {
|
||||
t.Fatalf("Validate: %v\npayload: %s", err, raw)
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
@@ -114,11 +114,11 @@ type AlertCompositeQuery struct {
|
||||
|
||||
type RuleCondition struct {
|
||||
CompositeQuery *AlertCompositeQuery `json:"compositeQuery" required:"true"`
|
||||
CompareOperator CompareOperator `json:"op" required:"true"`
|
||||
CompareOperator CompareOperator `json:"op,omitzero"`
|
||||
Target *float64 `json:"target,omitempty"`
|
||||
AlertOnAbsent bool `json:"alertOnAbsent,omitempty"`
|
||||
AbsentFor uint64 `json:"absentFor,omitempty"`
|
||||
MatchType MatchType `json:"matchType" required:"true"`
|
||||
MatchType MatchType `json:"matchType,omitzero"`
|
||||
TargetUnit string `json:"targetUnit,omitempty"`
|
||||
Algorithm string `json:"algorithm,omitempty"`
|
||||
Seasonality Seasonality `json:"seasonality,omitzero"`
|
||||
|
||||
@@ -50,13 +50,13 @@ const (
|
||||
// PostableRule is used to create alerting rule from HTTP api.
|
||||
type PostableRule struct {
|
||||
AlertName string `json:"alert" required:"true"`
|
||||
AlertType AlertType `json:"alertType,omitempty"`
|
||||
AlertType AlertType `json:"alertType" required:"true"`
|
||||
Description string `json:"description,omitempty"`
|
||||
RuleType RuleType `json:"ruleType,omitzero" required:"true"`
|
||||
RuleType RuleType `json:"ruleType" required:"true"`
|
||||
EvalWindow valuer.TextDuration `json:"evalWindow,omitzero"`
|
||||
Frequency valuer.TextDuration `json:"frequency,omitzero"`
|
||||
|
||||
RuleCondition *RuleCondition `json:"condition,omitempty" required:"true"`
|
||||
RuleCondition *RuleCondition `json:"condition" required:"true"`
|
||||
Labels map[string]string `json:"labels,omitempty"`
|
||||
Annotations map[string]string `json:"annotations,omitempty"`
|
||||
|
||||
@@ -67,9 +67,9 @@ type PostableRule struct {
|
||||
|
||||
PreferredChannels []string `json:"preferredChannels,omitempty"`
|
||||
|
||||
Version string `json:"version,omitempty"`
|
||||
Version string `json:"version"`
|
||||
|
||||
Evaluation *EvaluationEnvelope `yaml:"evaluation,omitempty" json:"evaluation,omitempty"`
|
||||
Evaluation *EvaluationEnvelope `json:"evaluation,omitempty"`
|
||||
SchemaVersion string `json:"schemaVersion,omitempty"`
|
||||
|
||||
NotificationSettings *NotificationSettings `json:"notificationSettings,omitempty"`
|
||||
|
||||
Reference in New Issue
Block a user