Skip to content

Commit

Permalink
[Telemetry][Security Solution] Fix flaky tests (#190044)
Browse files Browse the repository at this point in the history 
## Summary

Fixes: #188234
#187719 and
#178918

The flakiness was while calculating the Detection Rules task
invocations. It could have two different RCs: 1) The code didn't retry
in case the task wasn't executed yet, which makes sense in a CI
environment, which is slower than a dev environment; 2) The timestamp to
filter out requests was calculated after the task was triggered, and if
the task is executed fast enough, it could lead to empty responses
because of that.
  • Loading branch information
@szaffarano
szaffarano authored Aug 7, 2024
1 parent a763907 commit 358e104
Showing 1 changed file with 45 additions and 58 deletions.
103 changes: 45 additions & 58 deletions x-pack/plugins/security_solution/server/integration_tests/telemetry.test.ts
View file
Original file line number Diff line number Diff line change
Expand Up @@ -148,8 +148,7 @@ describe('telemetry tasks', () => {
});
});

// FLAKY: https://github.com/elastic/kibana/issues/187719
describe.skip('detection-rules', () => {
describe('detection-rules', () => {
it('should execute when scheduled', async () => {
await mockAndScheduleDetectionRulesTask();

Expand All @@ -169,8 +168,7 @@ describe('telemetry tasks', () => {
});

it('should send task metrics', async () => {
const task = await mockAndScheduleDetectionRulesTask();
const started = performance.now();
const [task, started] = await mockAndScheduleDetectionRulesTask();

const requests = await getTaskMetricsRequests(task, started);

Expand All @@ -181,13 +179,10 @@ describe('telemetry tasks', () => {
});
});

// FLAKY: https://github.com/elastic/kibana/issues/178918
// FLAKY: https://github.com/elastic/kibana/issues/187720
describe.skip('sender configuration', () => {
describe('sender configuration', () => {
it('should use legacy sender by default', async () => {
// launch a random task and verify it uses the new configuration
const task = await mockAndScheduleDetectionRulesTask();
const started = performance.now();
const [task, started] = await mockAndScheduleDetectionRulesTask();

const requests = await getTaskMetricsRequests(task, started);
expect(requests.length).toBeGreaterThan(0);
Expand Down Expand Up @@ -216,8 +211,7 @@ describe('telemetry tasks', () => {
expect(found).toBeFalsy();
});

const task = await mockAndScheduleDetectionRulesTask();
const started = performance.now();
const [task, started] = await mockAndScheduleDetectionRulesTask();

const requests = await getTaskMetricsRequests(task, started);
expect(requests.length).toBeGreaterThan(0);
Expand Down Expand Up @@ -258,8 +252,7 @@ describe('telemetry tasks', () => {
});
});

// FLAKY: https://github.com/elastic/kibana/issues/189192
describe.skip('endpoint-diagnostics', () => {
describe('endpoint-diagnostics', () => {
it('should execute when scheduled', async () => {
await mockAndScheduleEndpointDiagnosticsTask();

Expand Down Expand Up @@ -298,8 +291,7 @@ describe('telemetry tasks', () => {
});
});

// FLAKY: https://github.com/elastic/kibana/issues/189330
describe.skip('endpoint-meta-telemetry', () => {
describe('endpoint-meta-telemetry', () => {
beforeEach(async () => {
await initEndpointIndices(esClient);
});
Expand Down Expand Up @@ -335,8 +327,7 @@ describe('telemetry tasks', () => {
Promise.reject(Error(errorMessage))
);

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const requests = await getTaskMetricsRequests(task, started);

Expand All @@ -361,8 +352,7 @@ describe('telemetry tasks', () => {

agentClient.listAgents = jest.fn((_) => Promise.reject(Error(errorMessage)));

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -401,8 +391,7 @@ describe('telemetry tasks', () => {
})
);

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -434,8 +423,7 @@ describe('telemetry tasks', () => {

telemetryReceiver.fetchPolicyConfigs = jest.fn((_) => Promise.reject(Error(errorMessage)));

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -478,8 +466,7 @@ describe('telemetry tasks', () => {
} as unknown as AgentPolicy);
});

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -512,8 +499,7 @@ describe('telemetry tasks', () => {
return Promise.reject(Error(errorMessage));
});

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -545,8 +531,7 @@ describe('telemetry tasks', () => {
return Promise.resolve(new Map());
});

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -579,8 +564,7 @@ describe('telemetry tasks', () => {
return Promise.reject(Error(errorMessage));
});

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -615,8 +599,7 @@ describe('telemetry tasks', () => {
return Promise.resolve(new Map());
});

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand Down Expand Up @@ -663,8 +646,7 @@ describe('telemetry tasks', () => {
return esClient.search(query);
});

const task = await mockAndScheduleEndpointTask();
const started = performance.now();
const [task, started] = await mockAndScheduleEndpointTask();

const endpointMetaRequests = await getEndpointMetaRequests();

Expand All @@ -688,8 +670,7 @@ describe('telemetry tasks', () => {
});
});

// FLAKY: https://github.com/elastic/kibana/issues/188234
describe.skip('telemetry-prebuilt-rule-alerts', () => {
describe('telemetry-prebuilt-rule-alerts', () => {
it('should execute when scheduled', async () => {
await mockAndSchedulePrebuiltRulesTask();

Expand Down Expand Up @@ -722,8 +703,7 @@ describe('telemetry tasks', () => {

telemetryReceiver.fetchPrebuiltRuleAlertsBatch = mockedGenerator;

const task = await mockAndSchedulePrebuiltRulesTask();
const started = performance.now();
const [task, started] = await mockAndSchedulePrebuiltRulesTask();

const requests = await getTaskMetricsRequests(task, started);

Expand Down Expand Up @@ -781,7 +761,7 @@ describe('telemetry tasks', () => {
});
}

async function mockAndScheduleDetectionRulesTask(): Promise<SecurityTelemetryTask> {
async function mockAndScheduleDetectionRulesTask(): Promise<[SecurityTelemetryTask, number]> {
const task = getTelemetryTask(tasks, 'security:telemetry-detection-rules');

// create some data
Expand All @@ -797,50 +777,52 @@ describe('telemetry tasks', () => {
exceptionsListItem.push(exceptionListItem);

// schedule task to run ASAP
await eventually(async () => {
return eventually(async () => {
const started = performance.now();
await taskManagerPlugin.runSoon(task.getTaskId());
return [task, started];
});

return task;
}

async function mockAndScheduleEndpointTask(): Promise<SecurityTelemetryTask> {
async function mockAndScheduleEndpointTask(): Promise<[SecurityTelemetryTask, number]> {
const task = getTelemetryTask(tasks, 'security:endpoint-meta-telemetry');

await mockEndpointData(esClient, kibanaServer.coreStart.savedObjects);

// schedule task to run ASAP
await eventually(async () => {
return eventually(async () => {
const started = performance.now();
await taskManagerPlugin.runSoon(task.getTaskId());
return [task, started];
});

return task;
}

async function mockAndSchedulePrebuiltRulesTask(): Promise<SecurityTelemetryTask> {
async function mockAndSchedulePrebuiltRulesTask(): Promise<[SecurityTelemetryTask, number]> {
const task = getTelemetryTask(tasks, 'security:telemetry-prebuilt-rule-alerts');

await mockPrebuiltRulesData(esClient);

// schedule task to run ASAP
await eventually(async () => {
return eventually(async () => {
const started = performance.now();
await taskManagerPlugin.runSoon(task.getTaskId());
return [task, started];
});

return task;
}

async function mockAndScheduleEndpointDiagnosticsTask(): Promise<SecurityTelemetryTask> {
async function mockAndScheduleEndpointDiagnosticsTask(): Promise<
[SecurityTelemetryTask, number]
> {
const task = getTelemetryTask(tasks, 'security:endpoint-diagnostics');

await createMockedEndpointAlert(kibanaServer.coreStart.elasticsearch.client.asInternalUser);

// schedule task to run ASAP
await eventually(async () => {
return eventually(async () => {
const started = performance.now();
await taskManagerPlugin.runSoon(task.getTaskId());
return [task, started];
});

return task;
}

function mockAxiosGet(bufferConfig: unknown = fakeBufferAndSizesConfigAsyncDisabled) {
Expand Down Expand Up @@ -877,6 +859,7 @@ describe('telemetry tasks', () => {
requestConfig: AxiosRequestConfig<unknown> | undefined;
}>
> {
const taskType = getTelemetryTaskType(task);
return eventually(async () => {
const calls = mockedAxiosPost.mock.calls.flatMap(([url, data, config]) => {
return (data as string).split('\n').map((body) => {
Expand All @@ -886,20 +869,24 @@ describe('telemetry tasks', () => {

const requests = calls.filter(({ url, body }) => {
return (
body.indexOf(getTelemetryTaskType(task)) !== -1 &&
body.indexOf(taskType) !== -1 &&
url.startsWith(ENDPOINT_STAGING) &&
url.endsWith('task-metrics')
);
});
expect(requests.length).toBeGreaterThan(0);
return requests
const filtered = requests
.map((r) => {
return {
taskMetric: JSON.parse(r.body) as TaskMetric,
requestConfig: r.config,
};
})
.filter((t) => t.taskMetric.start_time >= olderThan);
.filter((t) => {
return t.taskMetric.start_time >= olderThan;
});
expect(filtered.length).toBeGreaterThan(0);
return filtered;
});
}
});
Expand Down

0 comments on commit 358e104

Please sign in to comment.