Added Openkruise workload integration health check scripts (argoproj#…

…16238)

Signed-off-by: Mahesh <maheshkasbe010@gmail.com>
Co-authored-by: Ishita Sequeira <46771830+ishitasequeira@users.noreply.github.com>
Signed-off-by: Kevin Lyda <kevin@lyda.ie>

resource_customizations/apps.kruise.io/AdvancedCronJob/health.lua

@@ -0,0 +1,36 @@
+hs = { status = "Progressing", message = "AdvancedCronJobs has active jobs" }
+-- Extract lastScheduleTime and convert to time objects
+lastScheduleTime = nil
+
+if obj.status.lastScheduleTime ~= nil then
+  local year, month, day, hour, min, sec = string.match(obj.status.lastScheduleTime, "(%d+)-(%d+)-(%d+)T(%d+):(%d+):(%d+)Z")
+  lastScheduleTime = os.time({year=year, month=month, day=day, hour=hour, min=min, sec=sec})
+end
+
+
+if lastScheduleTime == nil and obj.spec.paused == true then 
+    hs.status = "Suspended"
+    hs.message = "AdvancedCronJob is Paused"
+    return hs
+end
+
+-- AdvancedCronJobs are progressing if they have any object in the "active" state
+if obj.status.active ~= nil and #obj.status.active > 0 then
+    hs.status = "Progressing"
+    hs.message = "AdvancedCronJobs has active jobs"
+    return hs
+end
+-- AdvancedCronJobs are Degraded if they don't have lastScheduleTime
+if lastScheduleTime == nil then
+    hs.status = "Degraded"
+    hs.message = "AdvancedCronJobs has not run successfully"
+    return hs
+end
+-- AdvancedCronJobs are healthy if they have lastScheduleTime
+if lastScheduleTime ~= nil then
+    hs.status = "Healthy"
+    hs.message = "AdvancedCronJobs has run successfully"
+    return hs
+end
+
+return hs

resource_customizations/apps.kruise.io/AdvancedCronJob/health_test.yaml

@@ -0,0 +1,17 @@
+tests:
+  - healthStatus:
+      status: Healthy
+      message: AdvancedCronJobs has run successfully
+    inputPath: testdata/lastScheduleTime.yaml
+  - healthStatus:
+      status: Degraded
+      message: AdvancedCronJobs has not run successfully
+    inputPath: testdata/notScheduled.yaml
+  - healthStatus:
+      status: Progressing
+      message: AdvancedCronJobs has active jobs
+    inputPath: testdata/activeJobs.yaml
+  - healthStatus:
+      status: Suspended
+      message: AdvancedCronJob is Paused
+    inputPath: testdata/suspended.yaml

resource_customizations/apps.kruise.io/AdvancedCronJob/testdata/activeJobs.yaml

@@ -0,0 +1,30 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: AdvancedCronJob
+metadata:
+  name: acj-test
+spec:
+  schedule: "*/1 * * * *"
+  template:
+    broadcastJobTemplate:
+      spec:
+        template:
+          spec:
+            containers:
+              - name: pi
+                image: perl
+                command: ["perl",  "-Mbignum=bpi", "-wle", "print bpi(2000)"]
+            restartPolicy: Never
+        completionPolicy:
+          type: Always
+          ttlSecondsAfterFinished: 30
+
+status:
+  active:
+    - apiVersion: apps.kruise.io/v1alpha1
+      kind: BroadcastJob
+      name: acj-test-1694882400
+      namespace: default
+      resourceVersion: '4012'
+      uid: 2b08a429-a43b-4382-8e5d-3db0c72b5b13
+  lastScheduleTime: '2023-09-16T16:40:00Z'
+  type: BroadcastJob

resource_customizations/apps.kruise.io/AdvancedCronJob/testdata/lastScheduleTime.yaml

@@ -0,0 +1,23 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: AdvancedCronJob
+metadata:
+  name: acj-test
+spec:
+  schedule: "*/1 * * * *"
+  template:
+    broadcastJobTemplate:
+      spec:
+        template:
+          spec:
+            containers:
+              - name: pi
+                image: perl
+                command: ["perl",  "-Mbignum=bpi", "-wle", "print bpi(2000)"]
+            restartPolicy: Never
+        completionPolicy:
+          type: Always
+          ttlSecondsAfterFinished: 30
+
+status:
+  lastScheduleTime: "2023-09-16T16:29:00Z"
+  type: BroadcastJob

resource_customizations/apps.kruise.io/AdvancedCronJob/testdata/notScheduled.yaml

@@ -0,0 +1,22 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: AdvancedCronJob
+metadata:
+  name: acj-test
+spec:
+  schedule: "*/1 * * * *"
+  template:
+    broadcastJobTemplate:
+      spec:
+        template:
+          spec:
+            containers:
+              - name: pi
+                image: perl
+                command: ["perl",  "-Mbignum=bpi", "-wle", "print bpi(2000)"]
+            restartPolicy: Never
+        completionPolicy:
+          type: Always
+          ttlSecondsAfterFinished: 30
+
+status:
+  lastScheduleTime: null

resource_customizations/apps.kruise.io/AdvancedCronJob/testdata/suspended.yaml

@@ -0,0 +1,23 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: AdvancedCronJob
+metadata:
+  name: acj-test
+spec:
+  schedule: "*/1 * * * *"
+  template:
+    broadcastJobTemplate:
+      spec:
+        template:
+          spec:
+            containers:
+              - name: pi
+                image: perl
+                command: ["perl",  "-Mbignum=bpi", "-wle", "print bpi(2000)"]
+            restartPolicy: Never
+        completionPolicy:
+          type: Always
+          ttlSecondsAfterFinished: 30
+  paused: true
+
+status:
+  type: BroadcastJob

resource_customizations/apps.kruise.io/BroadcastJob/health.lua

@@ -0,0 +1,32 @@
+hs={ status= "Progressing", message= "BroadcastJob is still running" }
+
+if obj.status ~= nil then 
+
+-- BroadcastJob are healthy if desired number and succeeded number is equal
+    if obj.status.desired == obj.status.succeeded and obj.status.phase == "completed" then 
+        hs.status = "Healthy"
+        hs.message = "BroadcastJob is completed successfully"
+        return hs
+    end
+-- BroadcastJob are progressing if active is not equal to 0
+    if obj.status.active ~= 0 and obj.status.phase == "running" then
+        hs.status = "Progressing"
+        hs.message = "BroadcastJob is still running"
+        return hs
+    end
+-- BroadcastJob are progressing if failed is not equal to 0
+    if obj.status.failed ~= 0  and obj.status.phase == "failed" then
+        hs.status = "Degraded"
+        hs.message = "BroadcastJob failed"
+        return hs
+    end
+
+    if obj.status.phase == "paused" and obj.spec.paused == true then 
+        hs.status = "Suspended"
+        hs.message = "BroadcastJob is Paused"
+        return hs
+    end
+
+end
+
+return hs

resource_customizations/apps.kruise.io/BroadcastJob/health_test.yaml

@@ -0,0 +1,17 @@
+tests:
+  - healthStatus:
+      status: Healthy
+      message: "BroadcastJob is completed successfully"
+    inputPath: testdata/succeeded.yaml
+  - healthStatus:
+      status: Degraded
+      message: "BroadcastJob failed"
+    inputPath: testdata/failed.yaml
+  - healthStatus:
+      status: Progressing
+      message: "BroadcastJob is still running"
+    inputPath: testdata/running.yaml
+  - healthStatus:
+      status: Suspended
+      message: "BroadcastJob is Paused"
+    inputPath: testdata/suspended.yaml

resource_customizations/apps.kruise.io/BroadcastJob/testdata/failed.yaml

@@ -0,0 +1,31 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: BroadcastJob
+metadata:
+  name: failed-job
+spec:
+  template:
+    spec:
+      containers:
+        - name: guestbook
+          image: openkruise/guestbook:v3
+          command: ["exit",  "1"] # a dummy command to fail
+      restartPolicy: Never
+  completionPolicy:
+    type: Always
+    ttlSecondsAfterFinished: 60 # the job will be deleted after 60 seconds
+
+status:
+  active: 0
+  completionTime: '2023-09-17T14:31:38Z'
+  conditions:
+    - lastProbeTime: '2023-09-17T14:31:38Z'
+      lastTransitionTime: '2023-09-17T14:31:38Z'
+      message: failure policy is FailurePolicyTypeFailFast and failed pod is found
+      reason: Failed
+      status: 'True'
+      type: Failed
+  desired: 1
+  failed: 1
+  phase: failed
+  startTime: '2023-09-17T14:31:32Z'
+  succeeded: 0

resource_customizations/apps.kruise.io/BroadcastJob/testdata/running.yaml

@@ -0,0 +1,22 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: BroadcastJob
+metadata:
+  name: download-image
+spec:
+  template:
+    spec:
+      containers:
+        - name: guestbook
+          image: openkruise/guestbook:v3
+          command: ["echo",  "started"] # a dummy command to do nothing
+      restartPolicy: Never
+  completionPolicy:
+    type: Always
+    ttlSecondsAfterFinished: 60 # the job will be deleted after 60 seconds
+status:
+  active: 1
+  desired: 1
+  failed: 0
+  phase: running
+  startTime: '2023-09-17T14:43:30Z'
+  succeeded: 0

resource_customizations/apps.kruise.io/BroadcastJob/testdata/succeeded.yaml

@@ -0,0 +1,31 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: BroadcastJob
+metadata:
+  name: download-image
+spec:
+  template:
+    spec:
+      containers:
+        - name: guestbook
+          image: openkruise/guestbook:v3
+          command: ["echo",  "started"] # a dummy command to do nothing
+      restartPolicy: Never
+  completionPolicy:
+    type: Always
+    ttlSecondsAfterFinished: 60 # the job will be deleted after 60 seconds
+status:
+  active: 0
+  completionTime: '2023-09-17T14:35:14Z'
+  conditions:
+    - lastProbeTime: '2023-09-17T14:35:14Z'
+      lastTransitionTime: '2023-09-17T14:35:14Z'
+      message: Job completed, 1 pods succeeded, 0 pods failed
+      reason: Complete
+      status: 'True'
+      type: Complete
+  desired: 1
+  failed: 0
+  phase: completed
+  startTime: '2023-09-17T14:35:07Z'
+  succeeded: 1
+

resource_customizations/apps.kruise.io/BroadcastJob/testdata/suspended.yaml

@@ -0,0 +1,31 @@
+apiVersion: apps.kruise.io/v1alpha1
+kind: BroadcastJob
+metadata:
+  name: download-image
+spec:
+  template:
+    spec:
+      containers:
+        - name: guestbook
+          image: openkruise/guestbook:v3
+          command: ["echo",  "started"] # a dummy command to do nothing
+      restartPolicy: Never
+  paused: true
+  completionPolicy:
+    type: Always
+    ttlSecondsAfterFinished: 60 # the job will be deleted after 60 seconds
+status:
+  active: 0
+  completionTime: '2023-09-17T14:35:14Z'
+  conditions:
+    - lastProbeTime: '2023-09-17T14:35:14Z'
+      lastTransitionTime: '2023-09-17T14:35:14Z'
+      message: Job completed, 1 pods succeeded, 0 pods failed
+      reason: Complete
+      status: 'True'
+      type: Complete
+  desired: 1
+  failed: 0
+  phase: paused
+  startTime: '2023-09-17T14:35:07Z'
+  succeeded: 0

resource_customizations/apps.kruise.io/CloneSet/health.lua

@@ -0,0 +1,33 @@
+hs={ status = "Progressing", message = "Waiting for initialization" }
+
+if obj.status ~= nil then
+
+    if obj.metadata.generation == obj.status.observedGeneration then
+
+        if obj.spec.updateStrategy.paused == true or not obj.status.updatedAvailableReplicas then
+            hs.status = "Suspended"
+            hs.message = "Cloneset is paused"
+            return hs
+        elseif obj.spec.updateStrategy.partition ~= 0 and obj.metadata.generation > 1 then
+            if obj.status.updatedReplicas >= obj.status.expectedUpdatedReplicas then
+                hs.status = "Suspended"
+                hs.message = "Cloneset needs manual intervention"
+                return hs
+            end
+
+        elseif obj.status.updatedAvailableReplicas == obj.status.replicas then
+            hs.status = "Healthy"
+            hs.message = "All Cloneset workloads are ready and updated"    
+            return hs
+
+        else
+            if obj.status.updatedAvailableReplicas ~= obj.status.replicas then
+                hs.status = "Degraded"
+                hs.message = "Some replicas are not ready or available"
+                return hs
+            end
+        end
+    end
+end
+
+return hs

resource_customizations/apps.kruise.io/CloneSet/health_test.yaml

@@ -0,0 +1,21 @@
+tests:
+  - healthStatus:
+      status: Healthy
+      message: "All Cloneset workloads are ready and updated"
+    inputPath: testdata/healthy.yaml
+  - healthStatus:
+      status: Degraded
+      message: "Some replicas are not ready or available"
+    inputPath: testdata/degraded.yaml
+  - healthStatus:
+      status: Progressing
+      message: "Waiting for initialization"
+    inputPath: testdata/unknown.yaml
+  - healthStatus: 
+      status: Suspended
+      message: "Cloneset is paused"
+    inputpath: testdata/suspended.yaml
+  - healthStatus: 
+      status: Suspended
+      message: "Cloneset needs manual intervention"
+    inputpath: testdata/partition_suspended.yaml