Edge Node Clusters: nodes, volumes, replicas, upgrades

Node Info: role, age, ip
Pod Info: status, restarts, ip
Volume and Replica Info: rebuild progress, health
Upgrade Status: cluster wide and node components

Signed-off-by: Andrew Durbin <[email protected]>

proto/info/edge_node_cluster.proto

@@ -0,0 +1,236 @@
+// Copyright(c) 2024 Zededa, Inc.
+// All rights reserved.
+
+syntax = "proto3";
+
+package org.lfedge.eve.info;
+option go_package = "github.com/lf-edge/eve-api/go/info";
+option java_package = "org.lfedge.eve.info";
+
+import "info/info.proto";
+
+enum KubeNodeStatus {
+  STATUS_UNKNOWN = 0;
+  STATUS_READY = 1;
+  STATUS_NOT_READY = 2;
+  STATUS_SCHEDULING_DISABLED = 3;
+  STATUS_OUT_OF_DISK = 4;
+  STATUS_MEMORY_PRESSURE = 5;
+}
+
+message KubeNodeInfo {
+  // Name of the node
+  string name = 1;
+
+  // Status of the node
+  KubeNodeStatus status = 2;
+
+  // Role of the node is server or not
+  bool roleServer = 3;
+
+  // Age of the node, in seconds ago
+  uint64 age = 4;
+
+  // Version of the API Server running on the node
+  string apiServerVersion = 5;
+
+  // Internal IP address of the node
+  string internalIP = 6;
+}
+
+enum StorageHealthStatus {
+  HEALTH_UNKNOWN = 0;
+  HEALTH_HEALTHY = 1;
+  HEALTH_DEGRAEDED_2_REPLICA_AVAILABLE_REPLICATING = 2;
+  HEALTH_DEGRADED_2_REPLICA_AVAILABLE_NOT_REPLICATING = 3;
+  HEALTH_DEGRADED_1_REPLICA_AVAILABLE_REPLICATING = 4;
+  HEALTH_DEGRADED_1_REPLICA_AVAILABLE_NOT_REPLICATING = 5;
+  HEALTH_FAILED = 6;
+}
+
+enum StorageVolumeState {
+  VOLUME_STATUS_UNKNOWN = 0;
+  VOLUME_STATUS_CREATING = 1;
+  VOLUME_STATUS_ATTACHED = 2;
+  VOLUME_STATUS_DETACHED = 3;
+  VOLUME_STATUS_ATTACHING = 4;
+  VOLUME_STATUS_DETACHING = 5;
+  VOLUME_STATUS_DELETING = 6;
+}
+
+enum StorageVolumeRobustness {
+  VOLUME_ROBUSTNESS_UNKNOWN = 0;
+  VOLUME_ROBUSTNESS_HEALTHY = 1;
+  VOLUME_ROBUSTNESS_DEGRADED = 2;
+  VOLUME_ROBUSTNESS_FAULTED = 3;
+}
+
+enum StorageVolumePVCStatus {
+  VOLUME_PVC_STATUS_UNKNOWN = 0;
+  VOLUME_PVC_STATUS_BOUND = 1;
+  VOLUME_PVC_STATUS_PENDING = 2;
+  VOLUME_PVC_STATUS_AVAILABLE = 3;
+  VOLUME_PVC_STATUS_RELEASED = 4;
+  VOLUME_PVC_STATUS_FAILED = 5;
+}
+
+enum StorageVolumeReplicaStatus {
+  VOLUME_REPLICA_STATUS_UNKNOWN = 0;
+  VOLUME_REPLICA_STATUS_REBUILDING = 1;
+  VOLUME_REPLICA_STATUS_ONLINE = 2;
+  VOLUME_REPLICA_STATUS_FAILED = 3;
+}
+
+message KubePodNameSpaceInfo {
+  // Name of the namespace
+  string name = 1;
+
+  // Number of pods in the namespace
+  uint32 podCount = 2;
+
+  // Number of pods in the namespace that are running
+  uint32 podRunningCount = 3;
+
+  // Number of pods in the namespace that are pending
+  uint32 podPendingCount = 4;
+
+  // Number of pods in the namespace that are failed
+  uint32 podFailedCount = 5;
+
+  // Number of pods in the namespace that are succeeded
+  uint32 podSucceededCount = 6;
+}
+
+enum KubePodStatus {
+  POD_STATUS_UNKNOWN = 0;
+  POD_STATUS_PENDING = 1;
+  POD_STATUS_RUNNING = 2;
+  POD_STATUS_SUCCEEDED = 3;
+  POD_STATUS_CONTAINER_CREATING = 4;
+  POD_STATUS_CRASHLOOP_BACKOFF = 5;
+  POD_STATUS_ERROR = 6;
+  POD_STATUS_EVICTED = 7;
+  POD_STATUS_FAILED = 8;
+}
+
+message KubeEVEAppPodInfo {
+  // Name of the EVE application
+  string name = 1;
+
+  // Application Status
+  KubePodStatus status = 2;
+
+  // Restart count of the application
+  uint32 restartCount = 3;
+
+  // Restart time of the application, seconds ago
+  uint64 restartTime = 4;
+
+  // Age of the application, seconds ago
+  uint64 age = 5;
+
+  // IP address of the application, on cni0 interface
+  string ipAddress = 6;
+
+  // Node name on which the application is running
+  string nodeName = 7;
+}
+
+message KubeVolumeReplicaInfo {
+  // Name of the volume replica
+  string name = 1;
+
+  // Rebuild progress of the volume replica
+  uint32 rebuildProgressPercentage = 2;
+
+  // Replica status
+  StorageVolumeReplicaStatus status = 3;
+}
+
+message KubeVolumeInfo {
+  // Name of the volume
+  string name = 1;
+
+  // Status of the volume
+  StorageVolumeState state = 2;
+
+  // Robustness of the volume
+  StorageVolumeRobustness robustness = 3;
+
+  // Age of the volume, seconds ago
+  uint64 age = 4;
+
+  // Provisioned size of the volume in bytes
+  uint64 provisionedBytes = 5;
+
+  // Allocated size of the volume in bytes
+  uint64 allocatedBytes = 6;
+
+  // PV/PVC status of the volume
+  StorageVolumePVCStatus pvcStatus = 7;
+
+  // Replicas of the volume
+  repeated KubeVolumeReplicaInfo replica = 8;
+}
+
+message KubeStorageInfo {
+  // Overall status of Longhorn
+  StorageHealthStatus health = 1;
+
+  // time of the health status transition, seconds ago
+  uint64 transitionTime = 2;
+
+  // Status of all the volumes in Longhorn
+  repeated KubeVolumeInfo volumes = 3;
+}
+
+enum KubeCompUpgradeStatus {
+  KUBE_COMP_UPGRADE_STATUS_UNKNOWN = 0;
+  KUBE_COMP_UPGRADE_STATUS_DOWNLOAD = 1;
+  KUBE_COMP_UPGRADE_STATUS_DOWNLOAD_FAILED = 2;
+  KUBE_COMP_UPGRADE_STATUS_IN_PROGRESS = 3; 
+  KUBE_COMP_UPGRADE_STATUS_FAILED = 4;
+  KUBE_COMP_UPGRADE_STATUS_COMPLETED = 5;
+}
+
+enum KubeComp {
+  COMP_UNKNOWN = 0;
+  COMP_CONTAINERD = 1; //every node will publish
+  COMP_K3S = 2; // every node will publish
+  COMP_MULTUS = 3; // only the seed node will publish the remaining here and below
+  COMP_KUBEVIRT = 4;
+  COMP_CDI = 5;
+  COMP_LONGHORN = 6;
+} 
+
+message KubeClusterUpgradeStatus {
+  // currentNode will have an empty value when no node is in an upgrade
+  string currentNode = 1;	
+
+  // component currently under upgrade, COMP_UNKNOWN when no upgrades in progress
+  KubeComp component = 2;
+
+  // status of the current component under upgrade, KUBE_COMP_UPGRADE_STATUS_UNKNOWN when 
+  //   no upgrades in progress
+  KubeCompUpgradeStatus status = 3;
+
+  // error info in case of failure
+  ErrorInfo error = 4;
+}
+
+message KubeClusterInfo {
+  // Info message on a list of cluster nodes
+  repeated KubeNodeInfo nodes = 1;
+
+  // Info message on a list of namespaces's pods summary
+  repeated KubePodNameSpaceInfo podNameSpaces = 2;
+
+  // Info message on a list of EVE applications
+  repeated KubeEVEAppPodInfo eveApps = 3;
+
+  // Info message on cluster storage
+  KubeStorageInfo storage = 4;
+
+  // The current status of the cluster upgrade
+  KubeClusterUpgradeStatus upgrade = 5;
+}