feat: max-static-node-id

ydb-platform · Jun 27, 2024 · 06eba36 · 06eba36
1 parent a41cd39
commit 06eba36
Show file tree

Hide file tree

Showing 10 changed files with 23 additions and 9 deletions.
diff --git a/cmd/run.go b/cmd/run.go
@@ -13,8 +13,8 @@ import (
 )
 
 func NewRunCmd() *cobra.Command {
-	restartOpts := options.RestartOptionsInstance
 	rootOpts := options.RootOptionsInstance
+	restartOpts := options.RestartOptionsInstance
 	restarter := restarters.NewRunRestarter(options.Logger)
 
 	cmd := cli.SetDefaultsOn(&cobra.Command{

diff --git a/pkg/options/restart.go b/pkg/options/restart.go
@@ -22,6 +22,7 @@ const (
 	DefaultRetryCount              = 3
 	DefaultRestartDurationSeconds  = 60
 	DefaultCMSQueryIntervalSeconds = 10
+	DefaultMaxStaticNodeId         = 50000
 )
 
 var AvailabilityModes = []string{"strong", "weak", "force"}
@@ -62,6 +63,8 @@ type RestartOptions struct {
 
 	KubeconfigPath string
 	K8sNamespace   string
+
+	MaxStaticNodeId int
 }
 
 var (
@@ -81,6 +84,10 @@ func (o *RestartOptions) Validate() error {
 		return fmt.Errorf("specified --kubeconfig, but not --k8s-namespace")
 	}
 
+	if o.MaxStaticNodeId < 0 {
+		return fmt.Errorf("specified invalid max-static-node-id: %d. Must be positive", o.MaxStaticNodeId)
+	}
+
 	if o.RestartDuration < 0 {
 		return fmt.Errorf("specified invalid restart duration seconds: %d. Must be positive", o.RestartDuration)
 	}
@@ -204,6 +211,10 @@ after that would be considered a regular cluster failure`)
 for this invocation must be the same as for the previous invocation, and this can not be verified at runtime since 
 the ydbops utility is stateless. Use at your own risk.`)
 
+	fs.IntVar(&o.MaxStaticNodeId, "max-static-node-id", DefaultMaxStaticNodeId,
+		`This argument is used to help ydbops distinguish storage and dynamic nodes. 
+Nodes with this nodeId or less will be considered storage.`)
+
 	profile.PopulateFromProfileLater(
 		fs.StringVar, &o.KubeconfigPath, "kubeconfig",
 		"",

diff --git a/pkg/rolling/restarters/interface.go b/pkg/rolling/restarters/interface.go
@@ -26,4 +26,5 @@ type FilterNodeParams struct {
 	SelectedTenants []string
 	SelectedNodeIds []uint32
 	SelectedHosts   []string
+	MaxStaticNodeId uint32
 }
diff --git a/pkg/rolling/restarters/primitives.go b/pkg/rolling/restarters/primitives.go
@@ -14,10 +14,10 @@ import (
 	"github.com/ydb-platform/ydbops/pkg/options"
 )
 
-func FilterStorageNodes(nodes []*Ydb_Maintenance.Node) []*Ydb_Maintenance.Node {
+func FilterStorageNodes(nodes []*Ydb_Maintenance.Node, maxStaticNodeId uint32) []*Ydb_Maintenance.Node {
 	return collections.FilterBy(nodes,
 		func(node *Ydb_Maintenance.Node) bool {
-			return node.GetStorage() != nil
+			return node.GetStorage() != nil && node.NodeId < maxStaticNodeId
 		},
 	)
 }

diff --git a/pkg/rolling/restarters/run.go b/pkg/rolling/restarters/run.go
@@ -63,7 +63,7 @@ func (r *RunRestarter) Filter(spec FilterNodeParams, cluster ClusterNodesInfo) [
 	var runScopeNodes []*Ydb_Maintenance.Node
 
 	if r.storageOnly {
-		storageNodes := FilterStorageNodes(cluster.AllNodes)
+		storageNodes := FilterStorageNodes(cluster.AllNodes, spec.MaxStaticNodeId)
 		runScopeNodes = PopulateByCommonFields(storageNodes, spec)
 	} else if r.dynnodeOnly {
 		tenantNodes := FilterTenantNodes(cluster.AllNodes)

diff --git a/pkg/rolling/restarters/storage_k8s.go b/pkg/rolling/restarters/storage_k8s.go
@@ -69,7 +69,7 @@ func (r *StorageK8sRestarter) Filter(
 
 	r.prepareK8sState(r.Opts.kubeconfigPath, storageLabelSelector, r.Opts.namespace)
 
-	allStorageNodes := FilterStorageNodes(cluster.AllNodes)
+	allStorageNodes := FilterStorageNodes(cluster.AllNodes, spec.MaxStaticNodeId)
 
 	selectedNodes := populateWithK8sRules(allStorageNodes, spec, r.FQDNToPodName)
 

diff --git a/pkg/rolling/restarters/storage_ssh.go b/pkg/rolling/restarters/storage_ssh.go
@@ -38,7 +38,7 @@ func (r StorageSSHRestarter) Filter(
 	spec FilterNodeParams,
 	cluster ClusterNodesInfo,
 ) []*Ydb_Maintenance.Node {
-	storageNodes := FilterStorageNodes(cluster.AllNodes)
+	storageNodes := FilterStorageNodes(cluster.AllNodes, spec.MaxStaticNodeId)
 
 	preSelectedNodes := PopulateByCommonFields(storageNodes, spec)
 

diff --git a/pkg/rolling/rolling.go b/pkg/rolling/rolling.go
@@ -108,6 +108,7 @@ func (r *Rolling) DoRestart() error {
 			StartedTime:     r.opts.StartedTime,
 			Version:         r.opts.VersionSpec,
 			ExcludeHosts:    r.opts.ExcludeHosts,
+			MaxStaticNodeId: uint32(r.opts.MaxStaticNodeId),
 		},
 		restarters.ClusterNodesInfo{
 			TenantToNodeIds: r.state.tenantNameToNodeIds,

diff --git a/tests/filtering_test.go b/tests/filtering_test.go
@@ -40,6 +40,7 @@ var _ = Describe("Test storage Filter", func() {
 		nodes := mock.CreateNodesFromShortConfig(nodeGroups, nodeInfoMap)
 
 		filterSpec := restarters.FilterNodeParams{
+			MaxStaticNodeId: options.DefaultMaxStaticNodeId,
 			StartedTime: &options.StartedTime{
 				Direction: '<',
 				Timestamp: fiveMinutesAgoTimestamp,
@@ -89,8 +90,9 @@ var _ = Describe("Test storage Filter", func() {
 
 		nodes := mock.CreateNodesFromShortConfig(nodeGroups, nodeInfoMap)
 
-		// empty params equivalent to no arguments
-		filterSpec := restarters.FilterNodeParams{}
+		filterSpec := restarters.FilterNodeParams{
+			MaxStaticNodeId: options.DefaultMaxStaticNodeId,
+		}
 
 		clusterInfo := restarters.ClusterNodesInfo{
 			AllNodes: nodes,

diff --git a/tests/run_e2e_test.go b/tests/run_e2e_test.go
@@ -112,7 +112,6 @@ func RunTestCase(tc TestCase) {
 					),
 				)
 				maintenanceTaskId = uuidOnlyRegexp.FindString(output)
-				fmt.Printf("DETERMINED taskId %s\n", maintenanceTaskId)
 			}
 
 			r := regexp.MustCompile(expectedOutputRegexp)
-Original file line number
+Diff line change
@@ Expand Up / @@ -112,7 +112,6 @@ func RunTestCase(tc TestCase) { @@
     					),
     				)
     				maintenanceTaskId = uuidOnlyRegexp.FindString(output)
-    				fmt.Printf("DETERMINED taskId %s\n", maintenanceTaskId)
     			}
     			r := regexp.MustCompile(expectedOutputRegexp)
@@ Expand Down @@