Merge pull request rancher#45039 from markusewalker/markusv28/fix-rel…

…ease-issues

[v2.8] Fix flaky nodescaling test cases + specify cluster types

go.mod

@@ -166,7 +166,7 @@ require (
 	github.com/containers/image/v5 v5.25.0
 	github.com/google/gnostic-models v0.6.8
 	github.com/rancher/rancher/pkg/apis v0.0.0-20240213233515-935d309ebad4
-	github.com/rancher/shepherd v0.0.0-20240405212128-578908d4308a
+	github.com/rancher/shepherd v0.0.0-20240408151625-d0c3b8dbe5dd
 	go.qase.io/client v0.0.0-20231114201952-65195ec001fa
 )
 

go.sum

@@ -1641,8 +1641,8 @@ github.com/rancher/remotedialer v0.3.0 h1:y1EO8JCsgZo0RcqTUp6U8FXcBAv27R+TLnWRcp
 github.com/rancher/remotedialer v0.3.0/go.mod h1:BwwztuvViX2JrLLUwDlsYt5DiyUwHLlzynRwkZLAY0Q=
 github.com/rancher/rke v1.5.7 h1:pCVziDwgulQc2WgRkisY6sEo3DFGgu1StE66UbkuF2c=
 github.com/rancher/rke v1.5.7/go.mod h1:vojhOf8U8VCmw7y17OENWXSIfEFPEbXCMQcmI7xN7i8=
-github.com/rancher/shepherd v0.0.0-20240405212128-578908d4308a h1:VV4AyNyCQrkPtvSVa1rQL69+A/gyWTYFgWSkjuR4JGQ=
-github.com/rancher/shepherd v0.0.0-20240405212128-578908d4308a/go.mod h1:LNI7nH1BptYMvJmuqsLgmkMytGBBTpW4jk4vAHCxfF4=
+github.com/rancher/shepherd v0.0.0-20240408151625-d0c3b8dbe5dd h1:Nog4ViMD04zg6GP+5IvsthOIWqps7m6mdMnhIwWkPxA=
+github.com/rancher/shepherd v0.0.0-20240408151625-d0c3b8dbe5dd/go.mod h1:LNI7nH1BptYMvJmuqsLgmkMytGBBTpW4jk4vAHCxfF4=
 github.com/rancher/steve v0.0.0-20240305150728-3943409601f1 h1:6wNYy3q9jget45syTN6K2uOLSYaptLYCHscY2WRmhDI=
 github.com/rancher/steve v0.0.0-20240305150728-3943409601f1/go.mod h1:o4vLBzMTKbHHhIiAcbgOiaN3aK1vIjL6ZTgaGxQYpsY=
 github.com/rancher/system-upgrade-controller/pkg/apis v0.0.0-20210727200656-10b094e30007 h1:ru+mqGnxMmKeU0Q3XIDxkARvInDIqT1hH2amTcsjxI4=

tests/v2/validation/deleting/delete_cluster_rke1_test.go

@@ -34,11 +34,20 @@ func (c *RKE1ClusterDeleteTestSuite) SetupSuite() {
 }
 
 func (c *RKE1ClusterDeleteTestSuite) TestDeletingRKE1Cluster() {
-	clusterID, err := clusters.GetClusterIDByName(c.client, c.client.RancherConfig.ClusterName)
-	require.NoError(c.T(), err)
-
-	clusters.DeleteRKE1Cluster(c.client, clusterID)
-	provisioning.VerifyDeleteRKE1Cluster(c.T(), c.client, clusterID)
+	tests := []struct {
+		name   string
+		client *rancher.Client
+	}{
+		{"Deleting cluster", c.client},
+	}
+
+	for _, tt := range tests {
+		clusterID, err := clusters.GetClusterIDByName(c.client, c.client.RancherConfig.ClusterName)
+		require.NoError(c.T(), err)
+
+		clusters.DeleteRKE1Cluster(tt.client, clusterID)
+		provisioning.VerifyDeleteRKE1Cluster(c.T(), tt.client, clusterID)
+	}
 }
 
 // In order for 'go test' to run this suite, we need to create

tests/v2/validation/deleting/delete_cluster_test.go

@@ -3,9 +3,12 @@
 package deleting
 
 import (
+	"strings"
 	"testing"
 
+	apisV1 "github.com/rancher/rancher/pkg/apis/provisioning.cattle.io/v1"
 	"github.com/rancher/shepherd/clients/rancher"
+	v1 "github.com/rancher/shepherd/clients/rancher/v1"
 	"github.com/rancher/shepherd/extensions/clusters"
 	"github.com/rancher/shepherd/extensions/provisioning"
 	"github.com/rancher/shepherd/pkg/session"
@@ -34,11 +37,35 @@ func (c *ClusterDeleteTestSuite) SetupSuite() {
 }
 
 func (c *ClusterDeleteTestSuite) TestDeletingCluster() {
-	clusterID, err := clusters.GetV1ProvisioningClusterByName(c.client, c.client.RancherConfig.ClusterName)
-	require.NoError(c.T(), err)
+	tests := []struct {
+		name   string
+		client *rancher.Client
+	}{
+		{"cluster", c.client},
+	}
+
+	for _, tt := range tests {
+		clusterID, err := clusters.GetV1ProvisioningClusterByName(c.client, c.client.RancherConfig.ClusterName)
+		require.NoError(c.T(), err)
+
+		cluster, err := tt.client.Steve.SteveType(clusters.ProvisioningSteveResourceType).ByID(clusterID)
+		require.NoError(c.T(), err)
+
+		updatedCluster := new(apisV1.Cluster)
+		err = v1.ConvertToK8sType(cluster, &updatedCluster)
+		require.NoError(c.T(), err)
+
+		if strings.Contains(updatedCluster.Spec.KubernetesVersion, "rke2") {
+			tt.name = "Deleting RKE2 " + tt.name
+		} else {
+			tt.name = "Deleting K3S " + tt.name
+		}
 
-	clusters.DeleteK3SRKE2Cluster(c.client, clusterID)
-	provisioning.VerifyDeleteRKE2K3SCluster(c.T(), c.client, clusterID)
+		c.Run(tt.name, func() {
+			clusters.DeleteK3SRKE2Cluster(tt.client, clusterID)
+			provisioning.VerifyDeleteRKE2K3SCluster(c.T(), tt.client, clusterID)
+		})
+	}
 }
 
 // In order for 'go test' to run this suite, we need to create

tests/v2/validation/nodescaling/README.md

@@ -21,50 +21,42 @@ rancher:
 ```
 
 ## Node Replacing
-Node replacement tests require that the given pools have unique, distinct roles and more than 1 node per pool. You can run a subset of the tests, but still need more than 1 node for the role you would like to run the test for. i.e. for `-run ^TestScaleDownAndUp/TestWorkerScaleDownAndUp$` you would need at least 1 pool with 2 or more dedicaated workers in it. The last node in the pool will be replaced. 
-
-Typically, a cluster with the following 3 pools is used for testing:
+Node replacement tests require that the given pools have unique, distinct roles and more than 1 node per pool. Typically, a cluster with the following 3 pools is used for testing:
 ```yaml
-{
-  {
-    ControlPlane: true,
-    Quantity:     2,
-  },
-  {
-    Etcd:     true,
-    Quantity: 3,
-  },
-  {
-    Worker:   true,
-    Quantity: 2,
-  },
-}
+provisioningInput:
+  nodePools:      # nodePools is specific for RKE1 clusters.
+  - nodeRoles:
+      etcd: true
+      quantity: 3
+  - nodeRoles:
+      controlplane: true
+      quantity: 2
+  - nodeRoles:
+      worker: true
+      quantity: 3
+  machinePools:   # machienPools is specific for RKE2/K3s clusters.
+  - machinePoolConfig:
+      etcd: true
+      quantity: 3
+  - machinePoolConfig:
+      controlplane: true
+      quantity: 2
+  - machinePoolConfig:
+      worker: true
+      quantity: 3
   ```
 
 These tests utilize Go build tags. Due to this, see the below examples on how to run the tests:
 
 ### RKE1
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1NodeScaleDownAndUp/TestEtcdScaleDownAndUp"` \
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1NodeScaleDownAndUp/TestControlPlaneScaleDownAndUp"` \
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1NodeScaleDownAndUp/TestWorkerScaleDownAndUp"`
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1NodeReplacingTestSuite/TestReplacingRKE1Nodes"`
 
 ### RKE2 | K3S
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestNodeScaleDownAndUp/TestEtcdScaleDownAndUp"` \
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestNodeScaleDownAndUp/TestControlPlaneScaleDownAndUp"` \
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestNodeScaleDownAndUp/TestWorkerScaleDownAndUp"`
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestNodeReplacingTestSuite/TestReplacingNodes"`
 
 ## Scaling Existing Node Pools
-Similar to the `provisioning` tests, the node scaling tests have static test cases as well as dynamicInput tests you can specify. In order to run the dynamicInput tests, you will need to define the `scalingInput` block in your config file. This block defines the quantity you would like the pool to be scaled up/down to. See an example below:
+Similar to the `provisioning` tests, the node scaling tests have static test cases as well as dynamicInput tests you can specify. In order to run the dynamicInput tests, you will need to define the `scalingInput` block in your config file. This block defines the quantity you would like the pool to be scaled up/down to. See an example below that accounts for node drivers, custom clusters and hosted clusters:
 ```yaml
-scalingInput:
-  nodePools:
-    nodeRoles:
-      worker: true
-      quantity: 2
-  machinePools:
-    nodeRoles:
-      etcd: true
-      quantity: 1
 scalingInput:
   nodeProvider: "ec2"
   nodePools:
@@ -76,23 +68,29 @@ scalingInput:
       etcd: true
       quantity: 1
   aksNodePool:
-    nodeCount: 1
+    nodeCount: 3
   eksNodePool:
-    desiredSize: 1
+    desiredSize: 6
   gkeNodePool:
-    initialNodeCount: 1
+    initialNodeCount: 3
 ```
-NOTE: When scaling AKS and EKS, you will need to make sure that the `maxCount` and `maxSize` parameter is greater than the desired scale amount, respectively. For example, if you wish to have 6 total EKS nodes, then the `maxSize` parameter needs to be at least 7. This is not a limitation of the automation, but rather how EkS specifically handles nodegroups.
+NOTE: When scaling AKS and EKS, you will need to make sure that the `maxCount` and `maxSize` parameter is greater than the desired scale amount, respectively. For example, if you wish to have 6 total EKS nodes, then the `maxSize` parameter needs to be at least 7. This is not a limitation of the automation, but rather how EKS specifically handles nodegroups.
+
+Additionally, for AKS, you must have `enableAutoScaling` set to true if you specify `maxCount` and `minCount`.
 
 These tests utilize Go build tags. Due to this, see the below examples on how to run the tests:
 
 ### RKE1
 `gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1NodeScalingTestSuite/TestScalingRKE1NodePools"` \
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1NodeScalingTestSuite/TestScalingRKE1NodePoolsDynamicInput"`
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1NodeScalingTestSuite/TestScalingRKE1NodePoolsDynamicInput"` \
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1CustomClusterNodeScalingTestSuite/TestScalingRKE1CustomClusterNodes"` \
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE1CustomClusterNodeScalingTestSuite/TestScalingRKE1CustomClusterNodesDynamicInput"`
 
 ### RKE2 | K3S
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE2NodeScalingTestSuite/TestRKE2NodeScalingTestSuite"` \
-`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestRKE2NodeScalingTestSuite/TestScalingRKE2NodePoolsDynamicInput"`
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestNodeScalingTestSuite/TestScalingNodePools"` \
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestNodeScalingTestSuite/TestScalingNodePoolsDynamicInput"` \
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestCustomClusterNodeScalingTestSuite/TestScalingCustomClusterNodes"` \
+`gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestCustomClusterNodeScalingTestSuite/TestScalingCustomClusterNodesDynamicInput"`
 
 ### AKS
 `gotestsum --format standard-verbose --packages=github.com/rancher/rancher/tests/v2/validation/nodescaling --junitfile results.xml -- -timeout=60m -tags=validation -v -run "TestAKSNodeScalingTestSuite/TestScalingAKSNodePools"` \

tests/v2/validation/nodescaling/scale_replace_rke1_test.go

@@ -7,25 +7,26 @@ import (
 
 	"github.com/rancher/shepherd/clients/rancher"
 	"github.com/rancher/shepherd/extensions/provisioninginput"
+	nodepools "github.com/rancher/shepherd/extensions/rke1/nodepools"
 	"github.com/rancher/shepherd/pkg/config"
 	"github.com/rancher/shepherd/pkg/session"
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
 )
 
-type NodeScaleRKE1DownAndUp struct {
+type RKE1NodeReplacingTestSuite struct {
 	suite.Suite
 	session        *session.Session
 	client         *rancher.Client
 	ns             string
 	clustersConfig *provisioninginput.Config
 }
 
-func (s *NodeScaleRKE1DownAndUp) TearDownSuite() {
+func (s *RKE1NodeReplacingTestSuite) TearDownSuite() {
 	s.session.Cleanup()
 }
 
-func (s *NodeScaleRKE1DownAndUp) SetupSuite() {
+func (s *RKE1NodeReplacingTestSuite) SetupSuite() {
 	testSession := session.NewSession()
 	s.session = testSession
 
@@ -40,22 +41,42 @@ func (s *NodeScaleRKE1DownAndUp) SetupSuite() {
 	s.client = client
 }
 
-func (s *NodeScaleRKE1DownAndUp) TestEtcdScaleDownAndUp() {
-	s.Run("rke1-etcd-node-scale-down-and-up", func() {
-		ReplaceRKE1Nodes(s.T(), s.client, s.client.RancherConfig.ClusterName, true, false, false)
-	})
-}
-func (s *NodeScaleRKE1DownAndUp) TestWorkerScaleDownAndUp() {
-	s.Run("rke1-worker-node-scale-down-and-up", func() {
-		ReplaceRKE1Nodes(s.T(), s.client, s.client.RancherConfig.ClusterName, false, false, true)
-	})
-}
-func (s *NodeScaleRKE1DownAndUp) TestControlPlaneScaleDownAndUp() {
-	s.Run("rke1-controlplane-node-scale-down-and-up", func() {
-		ReplaceRKE1Nodes(s.T(), s.client, s.client.RancherConfig.ClusterName, false, true, false)
-	})
+func (s *RKE1NodeReplacingTestSuite) TestReplacingRKE1Nodes() {
+	nodeRolesEtcd := nodepools.NodeRoles{
+		Etcd:         true,
+		ControlPlane: false,
+		Worker:       false,
+	}
+
+	nodeRolesControlPlane := nodepools.NodeRoles{
+		Etcd:         false,
+		ControlPlane: true,
+		Worker:       false,
+	}
+
+	nodeRolesWorker := nodepools.NodeRoles{
+		Etcd:         false,
+		ControlPlane: false,
+		Worker:       true,
+	}
+
+	tests := []struct {
+		name      string
+		nodeRoles nodepools.NodeRoles
+		client    *rancher.Client
+	}{
+		{"Replacing control plane nodes", nodeRolesControlPlane, s.client},
+		{"Replacing etcd nodes", nodeRolesEtcd, s.client},
+		{"Replacing worker nodes", nodeRolesWorker, s.client},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			ReplaceRKE1Nodes(s.T(), s.client, s.client.RancherConfig.ClusterName, tt.nodeRoles.Etcd, tt.nodeRoles.ControlPlane, tt.nodeRoles.Worker)
+		})
+	}
 }
 
-func TestRKE1NodeScaleDownAndUp(t *testing.T) {
-	suite.Run(t, new(NodeScaleRKE1DownAndUp))
+func TestRKE1NodeReplacingTestSuite(t *testing.T) {
+	suite.Run(t, new(RKE1NodeReplacingTestSuite))
 }