[8.9][Fleet] added agent logs top errors from 100 hits (#162135) (#163278)

Backport https://github.com/elastic/kibana/pull/162135

x-pack/plugins/fleet/server/collectors/agent_logs_top_errors.test.ts

@@ -0,0 +1,84 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License
+ * 2.0; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+import type { ElasticsearchClient } from '@kbn/core-elasticsearch-server';
+
+import { getAgentLogsTopErrors } from './agent_logs_top_errors';
+
+describe('getAgentLogsTopErrors', () => {
+  it('should return top 3 errors from 100 hits', async () => {
+    const esClientMock = {
+      search: jest.fn().mockImplementation((params) => {
+        if (params.index === 'logs-elastic_agent-*')
+          return {
+            hits: {
+              hits: [
+                {
+                  _source: {
+                    message: 'error 2',
+                  },
+                },
+                {
+                  _source: {
+                    message: 'error 2',
+                  },
+                },
+                {
+                  _source: {
+                    message: 'error 3',
+                  },
+                },
+                {
+                  _source: {
+                    message: 'error 3',
+                  },
+                },
+                {
+                  _source: {
+                    message: 'error 3',
+                  },
+                },
+                {
+                  _source: {
+                    message: 'error 1',
+                  },
+                },
+              ],
+            },
+          };
+        else
+          return {
+            hits: {
+              hits: [
+                {
+                  _source: {
+                    message: 'fleet server error 2',
+                  },
+                },
+                {
+                  _source: {
+                    message: 'fleet server error 2',
+                  },
+                },
+                {
+                  _source: {
+                    message: 'fleet server error 1',
+                  },
+                },
+              ],
+            },
+          };
+      }),
+    } as unknown as ElasticsearchClient;
+
+    const topErrors = await getAgentLogsTopErrors(esClientMock);
+    expect(topErrors).toEqual({
+      agent_logs_top_errors: ['error 3', 'error 2', 'error 1'],
+      fleet_server_logs_top_errors: ['fleet server error 2', 'fleet server error 1'],
+    });
+  });
+});

x-pack/plugins/fleet/server/collectors/agent_logs_top_errors.ts

@@ -7,6 +7,10 @@
 
 import type { ElasticsearchClient } from '@kbn/core-elasticsearch-server';
 
+import { sortBy } from 'lodash';
+
+import { DATA_TIERS } from '../../common/constants';
+
 import { appContextService } from '../services';
 
 export interface AgentLogsData {
@@ -29,10 +33,16 @@ export async function getAgentLogsTopErrors(
     const queryTopMessages = (index: string) =>
       esClient.search({
         index,
-        size: 0,
+        size: 100,
+        _source: ['message'],
         query: {
           bool: {
             filter: [
+              {
+                terms: {
+                  _tier: DATA_TIERS,
+                },
+              },
               {
                 term: {
                   'log.level': 'error',
@@ -48,35 +58,32 @@ export async function getAgentLogsTopErrors(
             ],
           },
         },
-        aggs: {
-          message_sample: {
-            sampler: {
-              shard_size: 200,
-            },
-            aggs: {
-              categories: {
-                categorize_text: {
-                  field: 'message',
-                  size: 10,
-                },
-              },
-            },
-          },
-        },
       });
 
-    const transformBuckets = (resp: any) =>
-      ((resp?.aggregations?.message_sample as any)?.categories?.buckets ?? [])
+    const getTopErrors = (resp: any) => {
+      const counts = (resp?.hits.hits ?? []).reduce((acc: any, curr: any) => {
+        if (!acc[curr._source.message]) {
+          acc[curr._source.message] = 0;
+        }
+        acc[curr._source.message]++;
+        return acc;
+      }, {});
+      const top3 = sortBy(
+        Object.entries(counts).map(([key, value]) => ({ key, value })),
+        'value'
+      )
         .slice(0, 3)
-        .map((bucket: any) => bucket.key);
+        .reverse();
+      return top3.map(({ key, value }) => key);
+    };
 
     const agentResponse = await queryTopMessages('logs-elastic_agent-*');
 
     const fleetServerResponse = await queryTopMessages('logs-elastic_agent.fleet_server-*');
 
     return {
-      agent_logs_top_errors: transformBuckets(agentResponse),
-      fleet_server_logs_top_errors: transformBuckets(fleetServerResponse),
+      agent_logs_top_errors: getTopErrors(agentResponse),
+      fleet_server_logs_top_errors: getTopErrors(fleetServerResponse),
     };
   } catch (error) {
     if (error.statusCode === 404) {

x-pack/plugins/fleet/server/collectors/register.ts

@@ -21,6 +21,7 @@ import type { FleetServerUsage } from './fleet_server_collector';
 import { getAgentPoliciesUsage } from './agent_policies';
 import type { AgentPanicLogsData } from './agent_logs_panics';
 import { getPanicLogsLastHour } from './agent_logs_panics';
+import { getAgentLogsTopErrors } from './agent_logs_top_errors';
 
 export interface Usage {
   agents_enabled: boolean;
@@ -64,8 +65,7 @@ export const fetchFleetUsage = async (
     fleet_server_config: await getFleetServerConfig(soClient),
     agent_policies: await getAgentPoliciesUsage(soClient),
     ...(await getPanicLogsLastHour(esClient)),
-    // TODO removed top errors telemetry as it causes this issue: https://github.com/elastic/kibana/issues/148976
-    // ...(await getAgentLogsTopErrors(esClient)),
+    ...(await getAgentLogsTopErrors(esClient)),
   };
   return usage;
 };

x-pack/plugins/fleet/server/integration_tests/fleet_usage_telemetry.test.ts

@@ -382,8 +382,12 @@ describe('fleet usage telemetry', () => {
             message: 'stderr panic some other panic',
           },
         ],
-        // agent_logs_top_errors: ['stderr panic close of closed channel'],
-        // fleet_server_logs_top_errors: ['failed to unenroll offline agents'],
+        agent_logs_top_errors: [
+          'stderr panic some other panic',
+          'stderr panic close of closed channel',
+          'this should not be included in metrics',
+        ],
+        fleet_server_logs_top_errors: ['failed to unenroll offline agents'],
       })
     );
   });

x-pack/plugins/fleet/server/services/telemetry/fleet_usages_schema.ts

@@ -211,4 +211,18 @@ export const fleetUsagesSchema: RootSchema<any> = {
       },
     },
   },
+  agent_logs_top_errors: {
+    type: 'array',
+    items: {
+      type: 'text',
+      _meta: { description: 'Top messages from agent error logs' },
+    },
+  },
+  fleet_server_logs_top_errors: {
+    type: 'array',
+    items: {
+      type: 'text',
+      _meta: { description: 'Top messages from fleet server error logs' },
+    },
+  },
 };