Logs Volume Timeouts: Add a custom message with options (retry, close) for request timeouts (#65434)

* Logs Volume: identify timeouts and provide remediation action

* Supplementary result error: refactor updated component API

* Create helper to identify timeout errors

* Update timeout identifying function

* Add unit test

* Update panel unit test

* Update public/app/features/explore/utils/logsVolumeResponse.ts

Co-authored-by: Giordano Ricci <me@giordanoricci.com>

* Use some instead of reduce

* Change alert type to info

* Add comment

* Remove unnecessary optional chaining

* Remove unnecessary condition

* Remove unnecessary wrapping arrow function

---------

Co-authored-by: Giordano Ricci <me@giordanoricci.com>

public/app/features/explore/Logs.tsx

@@ -392,6 +392,7 @@ class UnthemedLogs extends PureComponent<Props, State> {
               onLoadLogsVolume={loadLogsVolumeData}
               onHiddenSeriesChanged={this.onToggleLogLevel}
               eventBus={this.logsVolumeEventBus}
+              onClose={() => this.onToggleLogsVolumeCollapse(false)}
             />
           )}
         </Collapse>

public/app/features/explore/LogsVolumePanelList.test.tsx

@@ -1,4 +1,5 @@
-import { fireEvent, render, screen } from '@testing-library/react';
+import { render, screen } from '@testing-library/react';
+import userEvent from '@testing-library/user-event';
 import React from 'react';
 
 import { DataQueryResponse, LoadingState, EventBusSrv } from '@grafana/data';
@@ -12,7 +13,7 @@ jest.mock('./Graph/ExploreGraph', () => {
   };
 });
 
-function renderPanel(logsVolumeData?: DataQueryResponse) {
+function renderPanel(logsVolumeData?: DataQueryResponse, onLoadLogsVolume = () => {}) {
   render(
     <LogsVolumePanelList
       absoluteRange={{ from: 0, to: 1 }}
@@ -21,7 +22,7 @@ function renderPanel(logsVolumeData?: DataQueryResponse) {
       width={100}
       onUpdateTimeRange={() => {}}
       logsVolumeData={logsVolumeData}
-      onLoadLogsVolume={() => {}}
+      onLoadLogsVolume={onLoadLogsVolume}
       onHiddenSeriesChanged={() => null}
       eventBus={new EventBusSrv()}
     />
@@ -40,7 +41,7 @@ describe('LogsVolumePanelList', () => {
     expect(screen.getByText('Test error message')).toBeInTheDocument();
   });
 
-  it('shows long warning message', () => {
+  it('shows long warning message', async () => {
     // we make a long message
     const messagePart = 'One two three four five six seven eight nine ten.';
     const message = messagePart + ' ' + messagePart + ' ' + messagePart;
@@ -48,7 +49,22 @@ describe('LogsVolumePanelList', () => {
     renderPanel({ state: LoadingState.Error, error: { data: { message } }, data: [] });
     expect(screen.getByText('Failed to load log volume for this query')).toBeInTheDocument();
     expect(screen.queryByText(message)).not.toBeInTheDocument();
-    fireEvent.click(screen.getByRole('button', { name: 'Show details' }));
+    await userEvent.click(screen.getByRole('button', { name: 'Show details' }));
     expect(screen.getByText(message)).toBeInTheDocument();
   });
+
+  it('a custom message for timeout errors', async () => {
+    const onLoadCallback = jest.fn();
+    renderPanel(
+      {
+        state: LoadingState.Error,
+        error: { data: { message: '{"status":"error","errorType":"timeout","error":"context deadline exceeded"}' } },
+        data: [],
+      },
+      onLoadCallback
+    );
+    expect(screen.getByText('The logs volume query is taking too long and has timed out')).toBeInTheDocument();
+    await userEvent.click(screen.getByRole('button', { name: 'Retry' }));
+    expect(onLoadCallback).toHaveBeenCalled();
+  });
 });

public/app/features/explore/LogsVolumePanelList.tsx

@@ -19,6 +19,7 @@ import { mergeLogsVolumeDataFrames } from '../logs/utils';
 
 import { LogsVolumePanel } from './LogsVolumePanel';
 import { SupplementaryResultError } from './SupplementaryResultError';
+import { isTimeoutErrorResponse } from './utils/logsVolumeResponse';
 
 type Props = {
   logsVolumeData: DataQueryResponse | undefined;
@@ -30,6 +31,7 @@ type Props = {
   onLoadLogsVolume: () => void;
   onHiddenSeriesChanged: (hiddenSeries: string[]) => void;
   eventBus: EventBus;
+  onClose?(): void;
 };
 
 export const LogsVolumePanelList = ({
@@ -42,6 +44,7 @@ export const LogsVolumePanelList = ({
   eventBus,
   splitOpen,
   timeZone,
+  onClose,
 }: Props) => {
   const logVolumes: Record<string, DataFrame[]> = useMemo(() => {
     const grouped = groupBy(logsVolumeData?.data || [], 'meta.custom.datasourceName');
@@ -59,10 +62,22 @@ export const LogsVolumePanelList = ({
     return !isLogsVolumeLimited(data) && zoomRatio && zoomRatio < 1;
   });
 
+  const timeoutError = isTimeoutErrorResponse(logsVolumeData);
+
   if (logsVolumeData?.state === LoadingState.Loading) {
     return <span>Loading...</span>;
-  }
+  } else if (timeoutError) {
-  if (logsVolumeData?.error !== undefined) {
+    return (
+      <SupplementaryResultError
+        title="The logs volume query is taking too long and has timed out"
+        // Using info to avoid users thinking that the actual query has failed.
+        severity="info"
+        suggestedAction="Retry"
+        onSuggestedAction={onLoadLogsVolume}
+        onRemove={onClose}
+      />
+    );
+  } else if (logsVolumeData?.error !== undefined) {
     return <SupplementaryResultError error={logsVolumeData.error} title="Failed to load log volume for this query" />;
   }
   return (

public/app/features/explore/SupplementaryResultError.tsx

@@ -1,23 +1,27 @@
 import React, { useState } from 'react';
 
 import { DataQueryError } from '@grafana/data';
-import { Alert, Button } from '@grafana/ui';
+import { Alert, AlertVariant, Button } from '@grafana/ui';
 
 type Props = {
-  error: DataQueryError;
+  error?: DataQueryError;
   title: string;
+  severity?: AlertVariant;
+  suggestedAction?: string;
+  onSuggestedAction?(): void;
+  onRemove?(): void;
 };
 export function SupplementaryResultError(props: Props) {
   const [isOpen, setIsOpen] = useState(false);
   const SHORT_ERROR_MESSAGE_LIMIT = 100;
-  const { error, title } = props;
+  const { error, title, suggestedAction, onSuggestedAction, onRemove, severity = 'warning' } = props;
   // generic get-error-message-logic, taken from
   // /public/app/features/explore/ErrorContainer.tsx
-  const message = error.message || error.data?.message || '';
+  const message = error?.message || error?.data?.message || '';
   const showButton = !isOpen && message.length > SHORT_ERROR_MESSAGE_LIMIT;
 
   return (
-    <Alert title={title} severity="warning">
+    <Alert title={title} severity={severity} onRemove={onRemove}>
       {showButton ? (
         <Button
           variant="secondary"
@@ -31,6 +35,11 @@ export function SupplementaryResultError(props: Props) {
       ) : (
         message
       )}
+      {suggestedAction && onSuggestedAction && (
+        <Button variant="primary" size="xs" onClick={onSuggestedAction}>
+          {suggestedAction}
+        </Button>
+      )}
     </Alert>
   );
 }

public/app/features/explore/utils/logsVolumeResponse.test.ts

@@ -0,0 +1,66 @@
+import { DataQueryResponse } from '@grafana/data';
+
+import { isTimeoutErrorResponse } from './logsVolumeResponse';
+
+const errorA =
+  'Get "http://localhost:3100/loki/api/v1/query_range?direction=backward&end=1680001200000000000&limit=1000&query=sum+by+%28level%29+%28count_over_time%28%7Bcontainer_name%3D%22docker-compose-app-1%22%7D%5B1h%5D%29%29&start=1679914800000000000&step=3600000ms": net/http: request canceled (Client.Timeout exceeded while awaiting headers)';
+const errorB = '{"status":"error","errorType":"timeout","error":"context deadline exceeded"}';
+
+describe('isTimeoutErrorResponse', () => {
+  test.each([errorA, errorB])(
+    'identifies timeout errors in the error.message attribute when the message is `%s`',
+    (timeoutError: string) => {
+      const response: DataQueryResponse = {
+        data: [],
+        error: {
+          message: timeoutError,
+        },
+      };
+      expect(isTimeoutErrorResponse(response)).toBe(true);
+    }
+  );
+  test.each([errorA, errorB])(
+    'identifies timeout errors in the errors.message attribute when the message is `%s`',
+    (timeoutError: string) => {
+      const response: DataQueryResponse = {
+        data: [],
+        errors: [
+          {
+            message: 'Something else',
+          },
+          {
+            message: timeoutError,
+          },
+        ],
+      };
+      expect(isTimeoutErrorResponse(response)).toBe(true);
+    }
+  );
+  test.each([errorA, errorB])(
+    'identifies timeout errors in the errors.data.message attribute when the message is `%s`',
+    (timeoutError: string) => {
+      const response: DataQueryResponse = {
+        data: [],
+        errors: [
+          {
+            data: {
+              message: 'Something else',
+            },
+          },
+          {
+            data: {
+              message: timeoutError,
+            },
+          },
+        ],
+      };
+      expect(isTimeoutErrorResponse(response)).toBe(true);
+    }
+  );
+  test('does not report false positives', () => {
+    const response: DataQueryResponse = {
+      data: [],
+    };
+    expect(isTimeoutErrorResponse(response)).toBe(false);
+  });
+});

public/app/features/explore/utils/logsVolumeResponse.ts

@@ -0,0 +1,18 @@
+import { DataQueryError, DataQueryResponse } from '@grafana/data';
+
+// Currently we can only infer if an error response is a timeout or not.
+export function isTimeoutErrorResponse(response: DataQueryResponse | undefined): boolean {
+  if (!response) {
+    return false;
+  }
+  if (!response.error && !response.errors) {
+    return false;
+  }
+
+  const errors = response.error ? [response.error] : response.errors || [];
+
+  return errors.some((error: DataQueryError) => {
+    const message = `${error.message || error.data?.message}`?.toLowerCase();
+    return message.includes('timeout');
+  });
+}