[Fleet] adding upgrade details UPG_FAILED check to stuck in updating (#173628)

## Summary

Closes https://github.com/elastic/kibana/issues/171419

Added upgrade details check to restart upgrade callout when agent is
stuck in updating.
Changed the copy slightly when the agent has failed upgrade state.

<img width="653" alt="image"
src="50061732-aa92-495b-b683-4d96114a1d5a">

Existing copy when there is no upgrade details (stuck in updating for
more than 2 hours):
<img width="609" alt="image"
src="14413c4a-ce51-49fa-a217-68993c4378fd">

Adjusted Agent list tooltip as well if upgrade details state is failed.

<img width="513" alt="image"
src="6836e02e-2c30-4c0a-91ea-8baddb0ccf03">



### Checklist

- [x] [Unit or functional
tests](https://www.elastic.co/guide/en/kibana/master/development-tests.html)
were updated or added to match the most common scenarios

---------

Co-authored-by: Kibana Machine <42973632+kibanamachine@users.noreply.github.com>

x-pack/plugins/fleet/common/services/agent_status.ts

@@ -61,10 +61,15 @@ export const AGENT_UPDATING_TIMEOUT_HOURS = 2;
 
 export function isStuckInUpdating(agent: Agent): boolean {
   return (
-    agent.status === 'updating' &&
-    !!agent.upgrade_started_at &&
-    !agent.upgraded_at &&
-    Date.now() - Date.parse(agent.upgrade_started_at) >
-      AGENT_UPDATING_TIMEOUT_HOURS * 60 * 60 * 1000
+    (agent.status !== 'offline' && agent.active && isAgentInFailedUpgradeState(agent)) ||
+    (agent.status === 'updating' &&
+      !!agent.upgrade_started_at &&
+      !agent.upgraded_at &&
+      Date.now() - Date.parse(agent.upgrade_started_at) >
+        AGENT_UPDATING_TIMEOUT_HOURS * 60 * 60 * 1000)
   );
 }
+
+export function isAgentInFailedUpgradeState(agent: Agent): boolean {
+  return agent.upgrade_details?.state === 'UPG_FAILED';
+}

x-pack/plugins/fleet/public/applications/fleet/sections/agents/components/agent_health.test.tsx

@@ -40,6 +40,8 @@ describe('AgentHealth', () => {
       true
     );
 
+    utils.getByText('Agent may be stuck updating.');
+
     act(() => {
       fireEvent.click(utils.getByTestId('restartUpgradeBtn'));
     });
@@ -47,6 +49,56 @@ describe('AgentHealth', () => {
     utils.findByText('Upgrade Modal');
   });
 
+  it('should render agent health with callout when agent has upgrade state failed', () => {
+    const { utils } = renderAgentHealth(
+      {
+        active: true,
+        status: 'online',
+        upgrade_started_at: '2022-11-21T12:27:24Z',
+        upgrade_details: {
+          state: 'UPG_FAILED',
+        },
+      } as any,
+      true
+    );
+
+    utils.getByText('Agent upgrade is stuck in failed state.');
+
+    utils.getByTestId('restartUpgradeBtn');
+  });
+
+  it('should not render agent health with callout when agent has upgrade state failed but offline', () => {
+    const { utils } = renderAgentHealth(
+      {
+        active: true,
+        status: 'offline',
+        upgrade_started_at: '2022-11-21T12:27:24Z',
+        upgrade_details: {
+          state: 'UPG_FAILED',
+        },
+      } as any,
+      true
+    );
+
+    expect(utils.queryByTestId('restartUpgradeBtn')).not.toBeInTheDocument();
+  });
+
+  it('should not render agent health with callout when agent has upgrade state failed but inactive', () => {
+    const { utils } = renderAgentHealth(
+      {
+        active: false,
+        status: 'unenrolled',
+        upgrade_started_at: '2022-11-21T12:27:24Z',
+        upgrade_details: {
+          state: 'UPG_FAILED',
+        },
+      } as any,
+      true
+    );
+
+    expect(utils.queryByTestId('restartUpgradeBtn')).not.toBeInTheDocument();
+  });
+
   it('should not render agent health with callout when agent not stuck updating', () => {
     const { utils } = renderAgentHealth(
       {
@@ -58,6 +110,7 @@ describe('AgentHealth', () => {
     );
 
     expect(utils.queryByTestId('restartUpgradeBtn')).not.toBeInTheDocument();
+    expect(utils.queryByText('Agent may be stuck updating.')).not.toBeInTheDocument();
   });
 
   it('should not render agent health with callout when not from details', () => {

x-pack/plugins/fleet/public/applications/fleet/sections/agents/components/agent_health.tsx

@@ -23,6 +23,7 @@ import { euiLightVars as euiVars } from '@kbn/ui-theme';
 
 import {
   getPreviousAgentStatusForOfflineAgents,
+  isAgentInFailedUpgradeState,
   isStuckInUpdating,
 } from '../../../../../../common/services/agent_status';
 
@@ -152,12 +153,19 @@ export const AgentHealth: React.FunctionComponent<Props> = ({ agent, fromDetails
             <p>{lastCheckinText}</p>
             <p>{lastCheckInMessageText}</p>
             {isStuckInUpdating(agent) ? (
-              <p>
+              isAgentInFailedUpgradeState(agent) ? (
                 <FormattedMessage
-                  id="xpack.fleet.agentHealth.restartUpgradeTooltipText"
-                  defaultMessage="Agent may be stuck updating. Consider restarting the upgrade."
+                  id="xpack.fleet.agentHealth.failedUpgradeTooltipText"
+                  defaultMessage="Agent upgrade failed. Consider restarting the upgrade."
                 />
-              </p>
+              ) : (
+                <p>
+                  <FormattedMessage
+                    id="xpack.fleet.agentHealth.restartUpgradeTooltipText"
+                    defaultMessage="Agent may be stuck updating. Consider restarting the upgrade."
+                  />
+                </p>
+              )
             ) : null}
           </>
         }
@@ -183,17 +191,27 @@ export const AgentHealth: React.FunctionComponent<Props> = ({ agent, fromDetails
             size="m"
             color="warning"
             title={
-              <FormattedMessage
-                id="xpack.fleet.agentHealth.stuckUpdatingTitle"
-                defaultMessage="Agent may be stuck updating."
-              />
+              isAgentInFailedUpgradeState(agent) ? (
+                <FormattedMessage
+                  id="xpack.fleet.agentHealth.failedUpgradeTitle"
+                  defaultMessage="Agent upgrade is stuck in failed state."
+                />
+              ) : (
+                <FormattedMessage
+                  id="xpack.fleet.agentHealth.stuckUpdatingTitle"
+                  defaultMessage="Agent may be stuck updating."
+                />
+              )
             }
           >
             <p>
               <FormattedMessage
                 id="xpack.fleet.agentHealth.stuckUpdatingText"
-                defaultMessage="Agent has been updating for a while, and may be stuck. Consider restarting the upgrade. {learnMore}"
+                defaultMessage="{stuckMessage} Consider restarting the upgrade. {learnMore}"
                 values={{
+                  stuckMessage: isAgentInFailedUpgradeState(agent)
+                    ? 'Agent upgrade failed.'
+                    : 'Agent has been updating for a while, and may be stuck.',
                   learnMore: (
                     <div>
                       <EuiLink href={docLinks.links.fleet.upgradeElasticAgent} target="_blank">

x-pack/plugins/translations/translations/fr-FR.json

@@ -16579,7 +16579,6 @@
     "xpack.fleet.agentFlyout.standaloneRadioOption": "{standaloneMessage} – Exécutez un agent Elastic Agent de façon autonome pour le configurer et le mettre à jour manuellement sur l'hôte sur lequel il est installé.",
     "xpack.fleet.agentHealth.checkinMessageText": "Dernier message de vérification : {lastCheckinMessage}",
     "xpack.fleet.agentHealth.checkInTooltipText": "Dernier archivage le {lastCheckIn}",
-    "xpack.fleet.agentHealth.stuckUpdatingText": "L'agent effectue la mise à jour depuis un certain temps ; il est peut-être bloqué. Envisagez de redémarrer la mise à niveau. {learnMore}",
     "xpack.fleet.agentList.noFilteredAgentsPrompt": "Aucun agent trouvé. {clearFiltersLink}",
     "xpack.fleet.agentLogs.logDisabledCallOutDescription": "Mettez à jour la politique de l'agent {settingsLink} pour activer la collecte de logs.",
     "xpack.fleet.agentLogs.oldAgentWarningTitle": "La vue Logs requiert Elastic Agent 7.11 ou une version ultérieure. Pour mettre à niveau un agent, accédez au menu Actions ou {downloadLink} une version plus récente.",

x-pack/plugins/translations/translations/ja-JP.json

@@ -16592,7 +16592,6 @@
     "xpack.fleet.agentFlyout.standaloneRadioOption": "{standaloneMessage} – Elasticエージェントをスタンドアロンで実行して、エージェントがインストールされているホストで、手動でエージェントを構成および更新します。",
     "xpack.fleet.agentHealth.checkinMessageText": "前回のチェックインメッセージ：{lastCheckinMessage}",
     "xpack.fleet.agentHealth.checkInTooltipText": "前回確認日時：{lastCheckIn}",
-    "xpack.fleet.agentHealth.stuckUpdatingText": "エージェントはしばらく更新が止まっている可能性があります。アップグレードの再開を検討してください。{learnMore}",
     "xpack.fleet.agentList.noFilteredAgentsPrompt": "エージェントが見つかりません。{clearFiltersLink}",
     "xpack.fleet.agentLogs.logDisabledCallOutDescription": "エージェントのポリシー{settingsLink}を更新して、ログ収集を有効にします。",
     "xpack.fleet.agentLogs.oldAgentWarningTitle": "ログの表示には、Elastic Agent 7.11以降が必要です。エージェントをアップグレードするには、［アクション］メニューに移動するか、新しいバージョンを{downloadLink}。",

x-pack/plugins/translations/translations/zh-CN.json

@@ -16592,7 +16592,6 @@
     "xpack.fleet.agentFlyout.standaloneRadioOption": "{standaloneMessage} – 独立运行 Elastic 代理，以在安装代理的主机上手动配置和更新代理。",
     "xpack.fleet.agentHealth.checkinMessageText": "上次签入消息：{lastCheckinMessage}",
     "xpack.fleet.agentHealth.checkInTooltipText": "上次签入时间 {lastCheckIn}",
-    "xpack.fleet.agentHealth.stuckUpdatingText": "代理已更新一段时间，并可能陷入停滞。请考虑重新开始升级。{learnMore}",
     "xpack.fleet.agentList.noFilteredAgentsPrompt": "找不到代理。{clearFiltersLink}",
     "xpack.fleet.agentLogs.logDisabledCallOutDescription": "更新代理的策略 {settingsLink} 以启用日志收集。",
     "xpack.fleet.agentLogs.oldAgentWarningTitle": "“日志”视图需要 Elastic Agent 7.11 或更高版本。要升级代理，请前往“操作”菜单或{downloadLink}更新的版本。",