Merge pull request #5330 from ZhengXinwei-F/task-manager

 Implement task manager to complete cloud edge task execution

5 files changed, 119 insertions, 91 deletions

diff --git a/manifests/charts/cloudcore/crds/operations_v1alpha1_imageprepulljob.yaml b/manifests/charts/cloudcore/crds/operations_v1alpha1_imageprepulljob.yaml
index 903516d32..a044a1532 100644
--- a/manifests/charts/cloudcore/crds/operations_v1alpha1_imageprepulljob.yaml
+++ b/manifests/charts/cloudcore/crds/operations_v1alpha1_imageprepulljob.yaml
@@ -48,6 +48,16 @@ spec:
                     items:
                       type: string
                     type: array
+                  concurrency:
+                    description: Concurrency specifies the maximum number of edge
+                      nodes that can pull images at the same time. The default Concurrency
+                      value is 1.
+                    format: int32
+                    type: integer
+                  failureTolerate:
+                    description: FailureTolerate specifies the task tolerance failure
+                      ratio. The default FailureTolerate value is 0.1.
+                    type: string
                   imageSecrets:
                     description: ImageSecret specifies the secret for image pull if
                       private registry used. Use {namespace}/{secretName} in format.
@@ -119,10 +129,10 @@ spec:
                       failed on each edgenode. Default to 0
                     format: int32
                     type: integer
-                  timeoutSecondsOnEachNode:
-                    description: TimeoutSecondsOnEachNode limits the duration of the
-                      image prepull job on each edgenode. Default to 360. If set to
-                      0, we'll use the default value 360.
+                  timeoutSeconds:
+                    description: TimeoutSeconds limits the duration of the node prepull
+                      job on each edgenode. Default to 300. If set to 0, we'll use
+                      the default value 300.
                     format: int32
                     type: integer
                 type: object
@@ -130,14 +140,21 @@ spec:
           status:
             description: Status represents the status of ImagePrePullJob.
             properties:
+              action:
+                description: 'Action represents for the action of the ImagePrePullJob.
+                  There are two possible action values: Success, Failure.'
+                type: string
+              event:
+                description: 'Event represents for the event of the ImagePrePullJob.
+                  There are four possible event values: Init, Check, Pull, TimeOut.'
+                type: string
+              reason:
+                description: Reason represents for the reason of the ImagePrePullJob.
+                type: string
               state:
                 description: 'State represents for the state phase of the ImagePrePullJob.
-                  There are four possible state values: "", prechecking, prepulling,
-                  successful, failed.'
-                enum:
-                - prepulling
-                - successful
-                - failed
+                  There are five possible state values: "", checking, pulling, successful,
+                  failed.'
                 type: string
               status:
                 description: Status contains image prepull status for each edge node.
@@ -163,31 +180,42 @@ spec:
                             description: 'State represents for the state phase of
                               this image pull on the edge node There are two possible
                               state values: successful, failed.'
-                            enum:
-                            - prepulling
-                            - successful
-                            - failed
                             type: string
                         type: object
                       type: array
-                    nodeName:
-                      description: NodeName is the name of edge node.
-                      type: string
-                    reason:
-                      description: Reason represents the fail reason if images prepull
-                        failed on the edge node
-                      type: string
-                    state:
-                      description: 'State represents for the state phase of the ImagePrepullJob
-                        on the edge node. There are five possible state values: "",
-                        prepulling, successful, failed.'
-                      enum:
-                      - prepulling
-                      - successful
-                      - failed
-                      type: string
+                    nodeStatus:
+                      description: TaskStatus represents the status for each node
+                      properties:
+                        action:
+                          description: 'Action represents for the action of the ImagePrePullJob.
+                            There are three possible action values: Success, Failure,
+                            TimeOut.'
+                          type: string
+                        event:
+                          description: 'Event represents for the event of the ImagePrePullJob.
+                            There are three possible event values: Init, Check, Pull.'
+                          type: string
+                        nodeName:
+                          description: NodeName is the name of edge node.
+                          type: string
+                        reason:
+                          description: Reason represents for the reason of the ImagePrePullJob.
+                          type: string
+                        state:
+                          description: 'State represents for the upgrade state phase
+                            of the edge node. There are several possible state values:
+                            "", Upgrading, BackingUp, RollingBack and Checking.'
+                          type: string
+                        time:
+                          description: Time represents for the running time of the
+                            ImagePrePullJob.
+                          type: string
+                      type: object
                   type: object
                 type: array
+              time:
+                description: Time represents for the running time of the ImagePrePullJob.
+                type: string
             type: object
         required:
         - spec
diff --git a/manifests/charts/cloudcore/crds/operations_v1alpha1_nodeupgradejob.yaml b/manifests/charts/cloudcore/crds/operations_v1alpha1_nodeupgradejob.yaml
index 130df65ac..958272424 100644
--- a/manifests/charts/cloudcore/crds/operations_v1alpha1_nodeupgradejob.yaml
+++ b/manifests/charts/cloudcore/crds/operations_v1alpha1_nodeupgradejob.yaml
@@ -36,12 +36,22 @@ spec:
           spec:
             description: Specification of the desired behavior of NodeUpgradeJob.
             properties:
+              checkItems:
+                description: CheckItems specifies the items need to be checked before
+                  the task is executed. The default CheckItems value is nil.
+                items:
+                  type: string
+                type: array
               concurrency:
                 description: Concurrency specifies the max number of edge nodes that
                   can be upgraded at the same time. The default Concurrency value
                   is 1.
                 format: int32
                 type: integer
+              failureTolerate:
+                description: FailureTolerate specifies the task tolerance failure
+                  ratio. The default FailureTolerate value is 0.1.
+                type: string
               image:
                 description: 'Image specifies a container image name, the image contains:
                   keadm and edgecore. keadm is used as upgradetool, to install the
@@ -111,75 +121,71 @@ spec:
                   job. Default to 300. If set to 0, we'll use the default value 300.
                 format: int32
                 type: integer
-              upgradeTool:
-                description: UpgradeTool is a request to decide use which upgrade
-                  tool. If it is empty, the upgrade job simply use default upgrade
-                  tool keadm to do upgrade operation.
-                type: string
               version:
                 type: string
             type: object
           status:
             description: Most recently observed status of the NodeUpgradeJob.
             properties:
-              state:
-                description: 'State represents for the state phase of the NodeUpgradeJob.
-                  There are three possible state values: "", upgrading and completed.'
-                enum:
-                - upgrading
-                - completed
+              action:
+                description: 'Action represents for the action of the ImagePrePullJob.
+                  There are two possible action values: Success, Failure.'
                 type: string
-              status:
+              currentVersion:
+                description: CurrentVersion represents for the current status of the
+                  EdgeCore.
+                type: string
+              event:
+                description: 'Event represents for the event of the ImagePrePullJob.
+                  There are six possible event values: Init, Check, BackUp, Upgrade,
+                  TimeOut, Rollback.'
+                type: string
+              historicVersion:
+                description: HistoricVersion represents for the historic status of
+                  the EdgeCore.
+                type: string
+              nodeStatus:
                 description: Status contains upgrade Status for each edge node.
                 items:
-                  description: UpgradeStatus stores the status of Upgrade for each
-                    edge node.
+                  description: TaskStatus stores the status of Upgrade for each edge
+                    node.
                   properties:
-                    history:
-                      description: History is the last upgrade result of the edge
-                        node.
-                      properties:
-                        fromVersion:
-                          description: FromVersion is the version which the edge node
-                            is upgraded from.
-                          type: string
-                        historyID:
-                          description: HistoryID is to uniquely identify an Upgrade
-                            Operation.
-                          type: string
-                        reason:
-                          description: Reason is the error reason of Upgrade failure.
-                            If the upgrade is successful, this reason is an empty
-                            string.
-                          type: string
-                        result:
-                          description: Result represents the result of upgrade.
-                          enum:
-                          - upgrade_success
-                          - upgrade_failed_rollback_success
-                          - upgrade_failed_rollback_failed
-                          type: string
-                        toVersion:
-                          description: ToVersion is the version which the edge node
-                            is upgraded to.
-                          type: string
-                        upgradeTime:
-                          description: UpgradeTime is the time of this Upgrade.
-                          type: string
-                      type: object
+                    action:
+                      description: 'Action represents for the action of the ImagePrePullJob.
+                        There are three possible action values: Success, Failure,
+                        TimeOut.'
+                      type: string
+                    event:
+                      description: 'Event represents for the event of the ImagePrePullJob.
+                        There are three possible event values: Init, Check, Pull.'
+                      type: string
                     nodeName:
                       description: NodeName is the name of edge node.
                       type: string
+                    reason:
+                      description: Reason represents for the reason of the ImagePrePullJob.
+                      type: string
                     state:
                       description: 'State represents for the upgrade state phase of
-                        the edge node. There are three possible state values: "",
-                        upgrading and completed.'
-                      enum:
-                      - upgrading
-                      - completed
+                        the edge node. There are several possible state values: "",
+                        Upgrading, BackingUp, RollingBack and Checking.'
+                      type: string
+                    time:
+                      description: Time represents for the running time of the ImagePrePullJob.
                       type: string
                   type: object
                 type: array
+              reason:
+                description: Reason represents for the reason of the ImagePrePullJob.
+                type: string
+              state:
+                description: 'State represents for the state phase of the NodeUpgradeJob.
+                  There are several possible state values: "", Upgrading, BackingUp,
+                  RollingBack and Checking.'
+                type: string
+              time:
+                description: Time represents for the running time of the ImagePrePullJob.
+                type: string
             type: object
         type: object
     served: true
diff --git a/manifests/charts/cloudcore/templates/configmap_cloudcore.yaml b/manifests/charts/cloudcore/templates/configmap_cloudcore.yaml
index 2986cfd91..23ca97d38 100644
--- a/manifests/charts/cloudcore/templates/configmap_cloudcore.yaml
+++ b/manifests/charts/cloudcore/templates/configmap_cloudcore.yaml
@@ -53,7 +53,5 @@ data:
       iptablesManager:
         enable: {{ .Values.iptablesManager.enable }}
         mode: {{ .Values.iptablesManager.mode }}
-      nodeUpgradeJobController:
-        enable: {{ .Values.cloudCore.modules.nodeUpgradeJobController.enable }}
-      imagePrePullController:
-        enable: {{ .Values.cloudCore.modules.imagePrePullController.enable }}
+      taskManager:
+        enable: {{ .Values.cloudCore.modules.taskManager.enable }}
diff --git a/manifests/charts/cloudcore/values.yaml b/manifests/charts/cloudcore/values.yaml
index a09482c03..db61c8e14 100644
--- a/manifests/charts/cloudcore/values.yaml
+++ b/manifests/charts/cloudcore/values.yaml
@@ -57,9 +57,7 @@ cloudCore:
       enable: false
     router:
       enable: false
-    nodeUpgradeJobController:
-      enable: false
-    imagePrePullController:
+    taskManager:
       enable: false
   service:
     enable: true
diff --git a/manifests/profiles/version.yaml b/manifests/profiles/version.yaml
index 5b3deb7e1..d5a1bf5c2 100644
--- a/manifests/profiles/version.yaml
+++ b/manifests/profiles/version.yaml
@@ -51,9 +51,7 @@ cloudCore:
       enable: false
     router:
       enable: false
-    nodeUpgradeJobController:
-      enable: false
-    imagePrePullController:
+    taskManager:
       enable: false
   service:
     enable: false