PWX-35939 Exposing cluster domain of a node in the SDK

SDK_CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## Releases
 
+### v0.176.0 - (02/15/2024)
+
+* Add cluster domain field to StorageNode
+
 ### v0.175.0 - (12/15/2023)
 
 * Add defrag status data structure

api/api.go

@@ -1036,6 +1036,7 @@ func (s *Node) ToStorageNode() *StorageNode {
  SecurityStatus: s.SecurityStatus,
  SchedulerTopology: s.SchedulerTopology,
  NonQuorumMember: s.NonQuorumMember,
+ ClusterDomain: s.DomainID,
  }
 
  node.Disks = make(map[string]*StorageResource)

api/api.proto

@@ -1633,6 +1633,8 @@ message StorageNode {
  // when initializing until it reaches a point where we can definitely determine whether
  // it is a quorum member or not.
  bool non_quorum_member = 22;
+ // Name of the cluster domain where the node belongs.
+ string cluster_domain = 23;
 }
 
 // StorageCluster represents the state and information about the cluster
@@ -4111,7 +4113,7 @@ message CollectDiagsJob {
 
 // DefragJob describes a job to run defragmentation on cluster nodes
 message DefragJob {
- // MaxDurationHours defines the time limit in hours 
+ // MaxDurationHours defines the time limit in hours
  double max_duration_hours = 1;
  // MaxNodesInParallel defines the maximum number of nodes running the defrag job in parallel
  uint32 max_nodes_in_parallel = 2;
@@ -4123,7 +4125,7 @@ message DefragJob {
  // if not provided, will run on all nodes
  // cannot coexist with IncludeNodes
  repeated string exclude_nodes = 4;
- // NodeSelector is a list of node label `key=value` pairs separated by comma, 
+ // NodeSelector is a list of node label `key=value` pairs separated by comma,
  // which selects the nodes to be run on for the job
  // can coexist with ExcludeNodes but cannot coexist with IncludeNodes
  repeated string node_selector = 5;
@@ -4141,7 +4143,7 @@ message DefragNodeStatus {
  string running_schedule = 2;
 }
 
-// DefragNodeStatus describes the defragmentation status of a pool 
+// DefragNodeStatus describes the defragmentation status of a pool
 message DefragPoolStatus {
  // NumIterations counts the number of times the pool gets defraged
  uint32 num_iterations = 1;
@@ -5659,7 +5661,7 @@ message SdkVersion {
  // SDK version major value of this specification
  Major = 0;
  // SDK version minor value of this specification
- Minor = 175;
+ Minor = 176;
  // SDK version patch value of this specification
  Patch = 0;
  }
@@ -6340,7 +6342,7 @@ message SdkVerifyChecksumStartRequest {
 // SdkVerifyChecksumStartResponse defines the response for a
 // SdkVerifyChecksumStartRequest.
 message SdkVerifyChecksumStartResponse {
- // Status code representing the state of the verify checksum operation 
+ // Status code representing the state of the verify checksum operation
  VerifyChecksum.VerifyChecksumStatus status = 1;
  // Text blob containing ASCII text providing details of the operation
  string message = 2;
@@ -6375,6 +6377,3 @@ message SdkVerifyChecksumStopResponse {
  // Text blob containing ASCII text providing details of the operation
  string message = 1;
 }
-
-
-

api/server/sdk/node_test.go

@@ -22,8 +22,8 @@ import (
  "testing"
  "time"
 
+ "github.com/golang/mock/gomock"
  "github.com/stretchr/testify/assert"
- "github.com/golang/mock/gomock"
  "google.golang.org/grpc/codes"
  "google.golang.org/grpc/status"
 
@@ -156,6 +156,7 @@ func TestSdkNodeEnumerateWithFilters(t *testing.T) {
  },
  },
  NonQuorumMember: true,
+ DomainID: "blue",
  },
  },
  }
@@ -174,6 +175,7 @@ func TestSdkNodeEnumerateWithFilters(t *testing.T) {
  },
  },
  NonQuorumMember: true,
+ ClusterDomain: "blue",
  }
 
  s.MockCluster().EXPECT().Enumerate().Return(cluster, nil).Times(1)
@@ -246,6 +248,7 @@ func TestSdkNodeInspect(t *testing.T) {
  },
  HWType: api.HardwareType_VirtualMachine,
  NonQuorumMember: true,
+ DomainID: "blue",
  }
  s.MockCluster().EXPECT().Inspect(nodeid).Return(node, nil).Times(1)
 
@@ -271,6 +274,7 @@ func TestSdkNodeInspect(t *testing.T) {
  assert.Equal(t, rn.GetStatus(), node.Status)
  assert.Equal(t, rn.GetHWType(), node.HWType)
  assert.Equal(t, node.NonQuorumMember, rn.NonQuorumMember)
+ assert.Equal(t, node.DomainID, rn.ClusterDomain)
 
  // Check Disk
  assert.Len(t, rn.GetDisks(), 2)
@@ -361,6 +365,7 @@ func TestSdkNodeInspectCurrent(t *testing.T) {
  },
  },
  NonQuorumMember: true,
+ DomainID: "blue",
  }
 
  cluster := api.Cluster{
@@ -393,6 +398,7 @@ func TestSdkNodeInspectCurrent(t *testing.T) {
  assert.Equal(t, rn.GetStatus(), node.Status)
  assert.Equal(t, rn.GetHWType(), node.HWType)
  assert.Equal(t, node.NonQuorumMember, rn.NonQuorumMember)
+ assert.Equal(t, node.DomainID, rn.ClusterDomain)
 
  // Check Disk
  assert.Len(t, rn.GetDisks(), 1)

cluster/manager/manager.go

@@ -289,6 +289,7 @@ func (c *ClusterManager) getNodeEntry(nodeID string, clustDBRef *cluster.Cluster
  n.HWType = v.HWType
  n.SecurityStatus = v.SecurityStatus
  n.NonQuorumMember = v.NonQuorumMember
+ n.DomainID = v.ClusterDomain
  } else {
  logrus.Warnf("Could not query NodeID %v", nodeID)
  // Node entry won't be refreshed form DB, will use the "offline" original
@@ -1157,7 +1158,7 @@ func (c *ClusterManager) waitForQuorum(exist bool) error {
  return nil
 }
 
-func (c *ClusterManager) initializeCluster(db kvdb.Kvdb, selfClusterDomain string) (
+func (c *ClusterManager) initializeCluster(db kvdb.Kvdb) (
  *cluster.ClusterInfo,
  error,
 ) {
@@ -1220,10 +1221,9 @@ func (c *ClusterManager) initListeners(
  db kvdb.Kvdb,
  nodeExists *bool,
  nodeInitialized bool,
- selfClusterDomain string,
 ) (uint64, *cluster.ClusterInfo, error) {
  // Initialize the cluster if required
- clusterInfo, err := c.initializeCluster(db, selfClusterDomain)
+ clusterInfo, err := c.initializeCluster(db)
  if err != nil {
  return 0, nil, err
  }
@@ -1328,13 +1328,11 @@ func (c *ClusterManager) initializeAndStartHeartbeat(
  kvdb kvdb.Kvdb,
  exist *bool,
  nodeInitialized bool,
- selfClusterDomain string,
 ) (uint64, *cluster.ClusterInfo, error) {
  lastIndex, clusterInfo, err := c.initListeners(
  kvdb,
  exist,
  nodeInitialized,
- selfClusterDomain,
  )
  if err != nil {
  return 0, nil, err
@@ -1475,6 +1473,7 @@ func (c *ClusterManager) StartWithConfiguration(
  c.gossipPort = gossipPort
  c.selfNode.GossipPort = gossipPort
  c.selfClusterDomain = selfClusterDomain
+ c.selfNode.DomainID = selfClusterDomain
  if err != nil {
  logrus.Errorf("Failed to get external IP address for mgt/data interfaces: %s.",
  err)
@@ -1527,7 +1526,6 @@ func (c *ClusterManager) StartWithConfiguration(
  kv,
  &exist,
  nodeInitialized,
- selfClusterDomain,
  )
  if err != nil {
  return err
@@ -1643,6 +1641,7 @@ func (c *ClusterManager) nodes(clusterDB *cluster.ClusterInfo) []*api.Node {
  node.NodeLabels = n.NodeLabels
  node.SecurityStatus = n.SecurityStatus
  node.NonQuorumMember = n.NonQuorumMember
+ node.DomainID = n.ClusterDomain
  }
  nodes = append(nodes, &node)
  }