Merge pull request #5373 from rawagner/restart_reboot_mc

Bug 1834172: Detect remediation strategy for Machine

frontend/packages/console-app/src/components/nodes/node-dashboard/NodeAlerts.tsx

@@ -163,8 +163,11 @@ const NodeAlerts: React.FC = () => {
 
   return (
     <AlertsBody>
-      {!!healthCheck && (
-        <StatusItem Icon={YellowExclamationTriangleIcon} message={msg.CONDITIONS_WARNING}>
+      {!!healthCheck?.failingHealthCheck && (
+        <StatusItem
+          Icon={YellowExclamationTriangleIcon}
+          message={msg.CONDITIONS_WARNING(healthCheck.reboot)}
+        >
           <HealthChecksLink />
         </StatusItem>
       )}

frontend/packages/console-app/src/components/nodes/node-dashboard/NodeDashboard.tsx

@@ -5,7 +5,7 @@ import DashboardGrid from '@console/shared/src/components/dashboard/DashboardGri
 import { NodeKind } from '@console/internal/module/k8s';
 import { LimitRequested } from '@console/shared/src/components/dashboard/utilization-card/UtilizationItem';
 
-import { NodeDashboardContext } from './NodeDashboardContext';
+import { NodeDashboardContext, HealthCheck } from './NodeDashboardContext';
 import InventoryCard from './InventoryCard';
 import DetailsCard from './DetailsCard';
 import StatusCard from './StatusCard';
@@ -51,7 +51,7 @@ export const reducer = (state: NodeDashboardState, action: NodeDashboardAction)
       };
     }
     case ActionType.HEALTH_CHECK: {
-      if (action.payload === state.healthCheck) {
+      if (_.isEqual(action.payload, state.healthCheck)) {
         return state;
       }
       return {
@@ -89,7 +89,7 @@ const NodeDashboard: React.FC<NodeDashboardProps> = ({ obj }) => {
     [],
   );
   const setHealthCheck = React.useCallback(
-    (payload: boolean) => dispatch({ type: ActionType.HEALTH_CHECK, payload }),
+    (payload: HealthCheck) => dispatch({ type: ActionType.HEALTH_CHECK, payload }),
     [],
   );
 
@@ -122,11 +122,11 @@ type NodeDashboardState = {
   obj: NodeKind;
   cpuLimit: LimitRequested;
   memoryLimit: LimitRequested;
-  healthCheck: boolean;
+  healthCheck: HealthCheck;
 };
 
 type NodeDashboardAction =
   | { type: ActionType.CPU_LIMIT; payload: LimitRequested }
   | { type: ActionType.MEMORY_LIMIT; payload: LimitRequested }
-  | { type: ActionType.HEALTH_CHECK; payload: boolean }
+  | { type: ActionType.HEALTH_CHECK; payload: HealthCheck }
   | { type: ActionType.OBJ; payload: NodeKind };

frontend/packages/console-app/src/components/nodes/node-dashboard/NodeDashboardContext.tsx

@@ -8,12 +8,17 @@ export const NodeDashboardContext = React.createContext<NodeDashboardContext>({
   setHealthCheck: () => {},
 });
 
+export type HealthCheck = {
+  failingHealthCheck: boolean;
+  reboot: boolean;
+};
+
 type NodeDashboardContext = {
   obj?: NodeKind;
   cpuLimit?: LimitRequested;
   setCPULimit: (state: LimitRequested) => void;
   memoryLimit?: LimitRequested;
   setMemoryLimit: (state: LimitRequested) => void;
-  healthCheck?: boolean;
-  setHealthCheck: (state: boolean) => void;
+  healthCheck?: HealthCheck;
+  setHealthCheck: (state: HealthCheck) => void;
 };

frontend/packages/console-app/src/components/nodes/node-dashboard/NodeHealth.tsx

@@ -38,7 +38,8 @@ export const HealthChecksPopup: React.FC<HealthChecksPopupProps> = ({
   conditions = [],
   machineHealthChecks,
 }) => {
-  let showRestart: boolean = false;
+  let conditionFailing: boolean = false;
+  let reboot: boolean = false;
   const grouppedConditions = Object.values(
     _.groupBy(
       conditions.sort((a, b) => a.type.localeCompare(b.type)),
@@ -47,7 +48,11 @@ export const HealthChecksPopup: React.FC<HealthChecksPopupProps> = ({
   ).map((cds) => {
     const failing = cds.some((c) => c.failing);
     if (failing) {
-      showRestart = true;
+      conditionFailing = true;
+      reboot =
+        machineHealthChecks?.[0]?.metadata?.annotations?.[
+          'machine.openshift.io/remediation-strategy'
+        ] === 'external-baremetal';
     }
     return {
       title: cds[0].type,
@@ -90,14 +95,14 @@ export const HealthChecksPopup: React.FC<HealthChecksPopupProps> = ({
           ))}
         </StatusPopupSection>
       )}
-      {showRestart && (
+      {conditionFailing && (
         <Alert
           variant="warning"
           isInline
-          title="Restart pending"
+          title={`${reboot ? 'Reboot' : 'Reprovision'} pending`}
           className="co-node-health__popup-alert"
         >
-          {CONDITIONS_WARNING}
+          {CONDITIONS_WARNING(reboot)}
         </Alert>
       )}
       {machineHealthChecks?.length > 1 && (
@@ -204,9 +209,24 @@ export const HealthChecksItem: React.FC = () => {
   const healthChecks = useK8sWatchResource<MachineHealthCheckKind[]>(machineHealthChecksResource);
   const healthState = getMachineHealth(obj, machine, healthChecks);
 
-  setHealthCheck(
-    healthState.state === HealthState.WARNING && healthState.conditions.some((c) => c.failing),
-  );
+  let failingHealthCheck = false;
+  let reboot = false;
+  _.forEach(healthState.conditions, (c) => {
+    if (c.failing) {
+      failingHealthCheck = true;
+      reboot =
+        healthState.matchingHC?.[0]?.metadata?.annotations?.[
+          'machine.openshift.io/remediation-strategy'
+        ] === 'external-baremetal';
+      return false;
+    }
+    return true;
+  });
+
+  setHealthCheck({
+    failingHealthCheck,
+    reboot,
+  });
 
   return (
     <HealthItem title="Health Checks" popupTitle="Health Checks" {...healthState}>

frontend/packages/console-app/src/components/nodes/node-dashboard/messages.ts

@@ -1,5 +1,7 @@
-export const CONDITIONS_WARNING =
-  'One or more health check remediation conditions have been met. The node will restart automatically.';
+export const CONDITIONS_WARNING = (reboot = false) =>
+  `One or more health check remediation conditions have been met. The node will ${
+    reboot ? 'reboot' : 'reprovision'
+  } automatically.`;
 
 export const CPU_LIMIT_REQ_ERROR =
   'This node’s CPU resources are overcommitted. The total CPU resource limit of all pods exceeds the node’s total capacity. The total CPU requested is also approaching the node’s capacity. Pod performance will be throttled under high load, and new pods may not be schedulable on this node.';

frontend/packages/metal3-plugin/src/components/baremetal-nodes/dashboard/BareMetalNodeDashboard.tsx

@@ -4,7 +4,10 @@ import DashboardGrid from '@console/shared/src/components/dashboard/DashboardGri
 import { NodeKind, K8sResourceKind } from '@console/internal/module/k8s';
 import UtilizationCard from '@console/app/src/components/nodes/node-dashboard/UtilizationCard';
 import ActivityCard from '@console/app/src/components/nodes/node-dashboard/ActivityCard';
-import { NodeDashboardContext } from '@console/app/src/components/nodes/node-dashboard/NodeDashboardContext';
+import {
+  NodeDashboardContext,
+  HealthCheck,
+} from '@console/app/src/components/nodes/node-dashboard/NodeDashboardContext';
 import {
   reducer,
   initialState,
@@ -49,7 +52,7 @@ const BareMetalNodeDashboard: React.FC<BareMetalNodeDashboardProps> = ({
     [],
   );
   const setHealthCheck = React.useCallback(
-    (payload: boolean) => dispatch({ type: ActionType.HEALTH_CHECK, payload }),
+    (payload: HealthCheck) => dispatch({ type: ActionType.HEALTH_CHECK, payload }),
     [],
   );