[Fleet] Filter non automatic upgrade failed from automatic upgrade status (#240328)

nchaulet · web-flow · commit 9a932c0621e3 · 2025-10-24T19:25:55.000-04:00
diff --git a/oas_docs/bundle.json b/oas_docs/bundle.json
@@ -16266,6 +16266,12 @@
                           "agents": {
                             "type": "number"
                           },
+                          "failedUpgradeActionIds": {
+                            "items": {
+                              "type": "string"
+                            },
+                            "type": "array"
+                          },
                           "failedUpgradeAgents": {
                             "type": "number"
                           },
diff --git a/oas_docs/bundle.serverless.json b/oas_docs/bundle.serverless.json
@@ -16266,6 +16266,12 @@
                           "agents": {
                             "type": "number"
                           },
+                          "failedUpgradeActionIds": {
+                            "items": {
+                              "type": "string"
+                            },
+                            "type": "array"
+                          },
                           "failedUpgradeAgents": {
                             "type": "number"
                           },
diff --git a/oas_docs/output/kibana.serverless.yaml b/oas_docs/output/kibana.serverless.yaml
@@ -23291,6 +23291,10 @@ paths:
                       properties:
                         agents:
                           type: number
+                        failedUpgradeActionIds:
+                          items:
+                            type: string
+                          type: array
                         failedUpgradeAgents:
                           type: number
                         version:
diff --git a/oas_docs/output/kibana.yaml b/oas_docs/output/kibana.yaml
@@ -27035,6 +27035,10 @@ paths:
                       properties:
                         agents:
                           type: number
+                        failedUpgradeActionIds:
+                          items:
+                            type: string
+                          type: array
                         failedUpgradeAgents:
                           type: number
                         version:
diff --git a/x-pack/platform/plugins/shared/fleet/common/types/rest_spec/agent_policy.ts b/x-pack/platform/plugins/shared/fleet/common/types/rest_spec/agent_policy.ts
@@ -41,6 +41,7 @@ export interface CurrentVersionCount {
   version: string;
   agents: number;
   failedUpgradeAgents: number;
+  failedUpgradeActionIds?: string[];
 }
 
 export interface GetAutoUpgradeAgentsStatusResponse {
diff --git a/x-pack/platform/plugins/shared/fleet/public/applications/fleet/sections/agents/components/manage_auto_upgrade_agents_modal/status_column.tsx b/x-pack/platform/plugins/shared/fleet/public/applications/fleet/sections/agents/components/manage_auto_upgrade_agents_modal/status_column.tsx
@@ -27,14 +27,23 @@ export const StatusColumn: React.FunctionComponent<{
   const { data: autoUpgradeAgentsStatus } = useGetAutoUpgradeAgentsStatusQuery(agentPolicyId);
   const getAgentsHref = useCallback(
     (failed?: boolean): string => {
+      const currentVersionAgentStatus = autoUpgradeAgentsStatus?.currentVersions.find(
+        (value) => value.version === version
+      );
       const kuery = failed
-        ? `policy_id:"${agentPolicyId}" AND upgrade_details.state:"UPG_FAILED" AND upgrade_details.target_version:"${version}"`
+        ? `policy_id:"${agentPolicyId}" AND upgrade_details.state:"UPG_FAILED" AND upgrade_details.target_version:"${version}"${
+            currentVersionAgentStatus && currentVersionAgentStatus.failedUpgradeActionIds?.length
+              ? ` AND (${currentVersionAgentStatus.failedUpgradeActionIds
+                  .map((id) => `upgrade_details.action_id:"${id}"`)
+                  .join(' OR ')})`
+              : ''
+          }`
         : `policy_id:"${agentPolicyId}" AND agent.version:"${version}"`;
       return getHref('agent_list', {
         kuery: encodeURIComponent(kuery),
       });
     },
-    [getHref, agentPolicyId, version]
+    [getHref, agentPolicyId, version, autoUpgradeAgentsStatus]
   );
 
   const calcPercentage = useCallback(
diff --git a/x-pack/platform/plugins/shared/fleet/server/services/agents/auto_upgrade_agents_status.test.ts b/x-pack/platform/plugins/shared/fleet/server/services/agents/auto_upgrade_agents_status.test.ts
@@ -5,14 +5,31 @@
  * 2.0.
  */
 
+import { createAppContextStartContractMock } from '../../mocks';
+import { appContextService } from '../app_context';
+
+import { getAgentActions } from './actions';
+
 import { getAutoUpgradeAgentsStatus } from './auto_upgrade_agents_status';
 
+jest.mock('./actions');
+
 describe('getAutoUpgradeAgentsStatus', () => {
+  beforeEach(() => {
+    appContextService.start(createAppContextStartContractMock());
+    jest.mocked(getAgentActions).mockImplementation(async (_: any, actionId: string) => {
+      if (actionId === 'action-1') {
+        return [{ id: 'action-1', is_automatic: true }];
+      } else {
+        return [{ id: 'action-2', is_automatic: false }];
+      }
+    });
+  });
   it('should kuery with active agents filter when listing agents', async () => {
     const agentClient = {
       listAgents: jest.fn().mockResolvedValue({
         aggregations: {
-          versions: {
+          action_id_versions: {
             buckets: [],
           },
         },
@@ -35,4 +52,51 @@ describe('getAutoUpgradeAgentsStatus', () => {
       })
     );
   });
+
+  it('should return only failed agents from automatic upgrade actions', async () => {
+    const agentClient = {
+      listAgents: jest
+        .fn()
+        .mockResolvedValueOnce({
+          total: 5,
+        })
+        .mockResolvedValueOnce({
+          aggregations: {
+            action_id_versions: {
+              buckets: [
+                {
+                  key: ['1.0.0', 'action-1'],
+                  doc_count: 10,
+                },
+                {
+                  key: ['1.0.0', 'action-2'],
+                  doc_count: 5,
+                },
+              ],
+            },
+          },
+          total: 15,
+        }),
+    } as any;
+
+    const agentPolicyId = 'test-policy-id';
+
+    const res = await getAutoUpgradeAgentsStatus(agentClient, agentPolicyId);
+
+    expect(res).toMatchInlineSnapshot(`
+      Object {
+        "currentVersions": Array [
+          Object {
+            "agents": 0,
+            "failedUpgradeActionIds": Array [
+              "action-1",
+            ],
+            "failedUpgradeAgents": 10,
+            "version": "1.0.0",
+          },
+        ],
+        "totalAgents": 5,
+      }
+    `);
+  });
 });
diff --git a/x-pack/platform/plugins/shared/fleet/server/services/agents/auto_upgrade_agents_status.ts b/x-pack/platform/plugins/shared/fleet/server/services/agents/auto_upgrade_agents_status.ts
@@ -5,13 +5,18 @@
  * 2.0.
  */
 
+import pMap from 'p-map';
+
 import { AGENTS_PREFIX } from '../../../common';
 import { AgentStatusKueryHelper } from '../../../common/services';
 import type {
   CurrentVersionCount,
   GetAutoUpgradeAgentsStatusResponse,
 } from '../../../common/types/rest_spec/agent_policy';
+import { appContextService } from '../app_context';
+import { MAX_CONCURRENT_AGENT_POLICIES_OPERATIONS_20 } from '../../constants';
 
+import { getAgentActions } from './actions';
 import type { AgentClient } from './agent_service';
 
 export async function getAutoUpgradeAgentsStatus(
@@ -55,22 +60,61 @@ export async function getAutoUpgradeAgentsStatus(
       perPage: 0,
       kuery: `${AgentStatusKueryHelper.buildKueryForActiveAgents()} AND ${AGENTS_PREFIX}.policy_id:"${agentPolicyId}" AND ${AGENTS_PREFIX}.upgrade_details.state:"UPG_FAILED"`,
       aggregations: {
-        versions: {
-          terms: {
-            field: 'upgrade_details.target_version.keyword',
+        action_id_versions: {
+          multi_terms: {
+            terms: [
+              {
+                field: 'upgrade_details.target_version.keyword',
+              },
+              {
+                field: 'upgrade_details.action_id',
+              },
+            ],
             size: 1000,
           },
         },
       },
     })
-    .then((result) => {
-      (result.aggregations?.versions as any)?.buckets.forEach(
-        (bucket: { key: string; doc_count: number }) =>
-          (currentVersionsMap[bucket.key] = {
-            version: bucket.key,
-            agents: currentVersionsMap[bucket.key]?.agents ?? 0,
-            failedUpgradeAgents: bucket.doc_count,
-          })
+    .then(async (result) => {
+      if (!result.aggregations?.action_id_versions) {
+        return;
+      }
+
+      const actionCacheIsAutomatic = new Map<string, boolean>();
+
+      await pMap(
+        (result.aggregations.action_id_versions as any)?.buckets ?? [],
+        async (bucket: { key: string[]; doc_count: number }) => {
+          const version = bucket.key[0];
+          const actionId = bucket.key[1];
+
+          let isAutomatic = actionCacheIsAutomatic.get(actionId);
+          if (isAutomatic === undefined) {
+            const actions = await getAgentActions(
+              appContextService.getInternalUserESClient(),
+              actionId
+            );
+            isAutomatic = actions?.some((action) => action.is_automatic) ?? false;
+            actionCacheIsAutomatic.set(actionId, isAutomatic);
+          }
+
+          if (isAutomatic) {
+            if (!currentVersionsMap[version]) {
+              currentVersionsMap[version] = {
+                version,
+                agents: 0,
+                failedUpgradeAgents: 0,
+              };
+            }
+
+            currentVersionsMap[version].failedUpgradeAgents += bucket.doc_count;
+            if (!currentVersionsMap[version].failedUpgradeActionIds) {
+              currentVersionsMap[version].failedUpgradeActionIds = [];
+            }
+            currentVersionsMap[version].failedUpgradeActionIds!.push(actionId);
+          }
+        },
+        { concurrency: MAX_CONCURRENT_AGENT_POLICIES_OPERATIONS_20 }
       );
     });
 
diff --git a/x-pack/platform/plugins/shared/fleet/server/types/models/agent_policy.ts b/x-pack/platform/plugins/shared/fleet/server/types/models/agent_policy.ts
@@ -337,6 +337,7 @@ export const GetAutoUpgradeAgentsStatusResponseSchema = schema.object({
       version: schema.string(),
       agents: schema.number(),
       failedUpgradeAgents: schema.number(),
+      failedUpgradeActionIds: schema.maybe(schema.arrayOf(schema.string())),
     })
   ),
   totalAgents: schema.number(),
diff --git a/x-pack/platform/test/fleet_api_integration/apis/agent_policy/agent_policy.ts b/x-pack/platform/test/fleet_api_integration/apis/agent_policy/agent_policy.ts
@@ -2143,6 +2143,7 @@ export default function (providerContext: FtrProviderContext) {
         await generateAgent(providerContext, 'healhty', 'agent-2', policyWithAgents.id, '8.16.1', {
           state: 'UPG_FAILED',
           target_version: '8.16.3',
+          action_id: 'test-action-automatic',
         });
         await generateAgent(
           providerContext,
@@ -2151,6 +2152,20 @@ export default function (providerContext: FtrProviderContext) {
           policyWithAgents.id,
           '8.16.1'
         );
+        await es.index({
+          index: '.fleet-actions',
+          refresh: 'wait_for',
+          body: {
+            '@timestamp': new Date().toISOString(),
+            expiration: new Date().toISOString(),
+            agents: ['agent-2'],
+            action_id: 'test-action-automatic',
+            data: {},
+            is_automatic: true,
+            type: 'UPGRADE',
+          },
+        });
+
         const { body } = await supertest
           .get(`/api/fleet/agent_policies/${policyWithAgents.id}/auto_upgrade_agents_status`)
           .set('kbn-xsrf', 'xxx')
@@ -2167,6 +2182,7 @@ export default function (providerContext: FtrProviderContext) {
               agents: 0,
               failedUpgradeAgents: 1,
               version: '8.16.3',
+              failedUpgradeActionIds: ['test-action-automatic'],
             },
           ],
           totalAgents: 2,

Original file line number	Diff line number	Diff line change
`@@ -41,6 +41,7 @@ export interface CurrentVersionCount {`
`41`	`41`	`version: string;`
`42`	`42`	`agents: number;`
`43`	`43`	`failedUpgradeAgents: number;`
	`44`	`+ failedUpgradeActionIds?: string[];`
`44`	`45`	`}`
`45`	`46`
`46`	`47`	`export interface GetAutoUpgradeAgentsStatusResponse {`