Merge pull request #1312 from zimnx/mz/replace-node-uuid

Replace ScyllaCluster nodes using Host ID

Author: zimnx

examples/common/operator.yaml

@@ -2965,7 +2965,7 @@ spec:
                       replace_address_first_boot:
                         additionalProperties:
                           type: string
-                        description: replace_address_first_boot holds addresses which should be replaced by new nodes.
+                        description: 'replace_address_first_boot holds addresses which should be replaced by new nodes. DEPRECATED: since Scylla Operator 1.10 it''s only used for deprecated replace node procedure (ScyllaDB OS <5.2, Enterprise <2023.1). With Scylla Operator 1.11+ this field may be empty.'
                         type: object
                       stale:
                         description: stale indicates if the current rack status is collected for a previous generation. stale should eventually become false when the appropriate controller writes a fresh status.

pkg/api/scylla/v1/scylla.scylladb.com_scyllaclusters.yaml

@@ -2009,7 +2009,7 @@ spec:
                       replace_address_first_boot:
                         additionalProperties:
                           type: string
-                        description: replace_address_first_boot holds addresses which should be replaced by new nodes.
+                        description: 'replace_address_first_boot holds addresses which should be replaced by new nodes. DEPRECATED: since Scylla Operator 1.10 it''s only used for deprecated replace node procedure (ScyllaDB OS <5.2, Enterprise <2023.1). With Scylla Operator 1.11+ this field may be empty.'
                         type: object
                       stale:
                         description: stale indicates if the current rack status is collected for a previous generation. stale should eventually become false when the appropriate controller writes a fresh status.

pkg/api/scylla/v1/types_cluster.go

@@ -493,10 +493,9 @@ type RackStatus struct {
 	// conditions are the latest available observations of a rack's state.
 	Conditions []RackCondition `json:"conditions,omitempty"`
 
-	// FIXME: The json value should have been a camelCase string.
-	//        We need to deprecate this value and introduce a new one.
-
 	// replace_address_first_boot holds addresses which should be replaced by new nodes.
+	// DEPRECATED: since Scylla Operator 1.10 it's only used for deprecated replace node procedure (ScyllaDB OS <5.2, Enterprise <2023.1).
+	//             With Scylla Operator 1.11+ this field may be empty.
 	ReplaceAddressFirstBoot map[string]string `json:"replace_address_first_boot,omitempty"`
 }
 

pkg/cmd/tests/tests_run.go

@@ -49,15 +49,15 @@ var suites = ginkgotest.TestSuites{
 		Description: templates.LongDesc(`
 		Runs all tests.
 		`),
-		DefaultParallelism: 15,
+		DefaultParallelism: 8,
 	},
 	{
 		Name: "scylla-operator/conformance/parallel",
 		Description: templates.LongDesc(`
 		Tests that ensure an Scylla Operator is working properly.
 		`),
 		LabelFilter:        fmt.Sprintf("!%s", framework.SerialLabelName),
-		DefaultParallelism: 15,
+		DefaultParallelism: 8,
 	},
 	{
 		Name: "scylla-operator/conformance/serial",

pkg/controller/scyllacluster/sync_services.go

@@ -19,6 +19,9 @@ const (
 	// ReplaceLabel express the intent to replace pod under the specific member.
 	ReplaceLabel = "scylla/replace"
 
+	// ReplacingNodeHostIDLabel contains the Host ID of node labelled node is replacing.
+	ReplacingNodeHostIDLabel = "internal.scylla-operator.scylladb.com/replacing-node-hostid"
+
 	// NodeMaintenanceLabel means that node is under maintenance.
 	// Readiness check will always fail when this label is added to member service.
 	NodeMaintenanceLabel = "scylla/node-maintenance"

pkg/naming/constants.go

@@ -0,0 +1,73 @@
+// Copyright (c) 2023 ScyllaDB.
+
+package scyllaclient
+
+import (
+	"context"
+	"fmt"
+	"net"
+	"net/http"
+	"time"
+
+	api "github.com/go-openapi/runtime/client"
+	"github.com/go-openapi/strfmt"
+	"github.com/scylladb/scylla-operator/pkg/auth"
+	scyllaclient "github.com/scylladb/scylladb-swagger-go-client/scylladb/gen/v2/client"
+	"github.com/scylladb/scylladb-swagger-go-client/scylladb/gen/v2/client/config"
+)
+
+const (
+	agentPort      = "10001"
+	defaultTimeout = 30 * time.Second
+)
+
+type ConfigClient struct {
+	client *scyllaclient.ScylladbV2
+}
+
+func NewConfigClient(host, authToken string) *ConfigClient {
+	var transport http.RoundTripper = DefaultTransport()
+	transport = fixContentType(transport)
+	transport = auth.AddToken(transport, authToken)
+
+	client := &http.Client{
+		Timeout:   defaultTimeout,
+		Transport: transport,
+	}
+
+	host = net.JoinHostPort(host, agentPort)
+
+	scyllaV2Runtime := api.NewWithClient(
+		host, scyllaclient.DefaultBasePath, scyllaclient.DefaultSchemes, client,
+	)
+
+	return &ConfigClient{
+		client: scyllaclient.New(scyllaV2Runtime, strfmt.Default),
+	}
+}
+
+func (c *ConfigClient) BroadcastAddress(ctx context.Context) (string, error) {
+	resp, err := c.client.Config.FindConfigBroadcastAddress(config.NewFindConfigBroadcastAddressParamsWithContext(ctx))
+	if err != nil {
+		return "", fmt.Errorf("can't get broadcast_address: %w", err)
+	}
+	return resp.Payload, nil
+}
+
+// ReplaceAddressFirstBoot returns value of "replace_address_first_boot" config parameter.
+func (c *ConfigClient) ReplaceAddressFirstBoot(ctx context.Context) (string, error) {
+	resp, err := c.client.Config.FindConfigReplaceAddressFirstBoot(config.NewFindConfigReplaceAddressFirstBootParamsWithContext(ctx))
+	if err != nil {
+		return "", fmt.Errorf("can't get replace_address_first_boot: %w", err)
+	}
+	return resp.Payload, nil
+}
+
+// ReplaceNodeFirstBoot returns value of "replace_node_first_boot" config parameter.
+func (c *ConfigClient) ReplaceNodeFirstBoot(ctx context.Context) (string, error) {
+	resp, err := c.client.Config.FindConfigReplaceNodeFirstBoot(config.NewFindConfigReplaceNodeFirstBootParamsWithContext(ctx))
+	if err != nil {
+		return "", fmt.Errorf("can't get replace_node_first_boot: %w", err)
+	}
+	return resp.Payload, nil
+}

pkg/scyllaclient/config_client.go

@@ -0,0 +1,62 @@
+// Copyright (c) 2023 ScyllaDB.
+
+package scyllafeatures
+
+import (
+	"fmt"
+
+	"github.com/blang/semver"
+	scyllav1 "github.com/scylladb/scylla-operator/pkg/api/scylla/v1"
+)
+
+var (
+	scyllaEnterpriseMinimalVersion = semver.MustParse("2000.0.0")
+)
+
+type ScyllaFeature string
+
+const (
+	ReplacingNodeUsingHostID ScyllaFeature = "ReplacingNodeUsingHostID"
+)
+
+type scyllaDBVersionMinimalConstraint struct {
+	openSource semver.Version
+	enterprise semver.Version
+}
+
+var featureMinimalVersionConstraints = map[ScyllaFeature]scyllaDBVersionMinimalConstraint{
+	ReplacingNodeUsingHostID: {
+		openSource: semver.MustParse("5.2.0"),
+		enterprise: semver.MustParse("2023.1.0"),
+	},
+}
+
+func Supports(sc *scyllav1.ScyllaCluster, feature ScyllaFeature) (bool, error) {
+	constraints, ok := featureMinimalVersionConstraints[feature]
+	if !ok {
+		return false, fmt.Errorf("unable to find minimal version constraints, unknown feature %q", feature)
+	}
+
+	version, err := semver.Parse(sc.Spec.Version)
+	if err != nil {
+		return false, fmt.Errorf("can't parse ScyllaCluster version %q: %w", sc.Spec.Version, err)
+	}
+
+	if isOpenSource(version) && version.GTE(constraints.openSource) {
+		return true, nil
+	}
+
+	if isEnterprise(version) && version.GTE(constraints.enterprise) {
+		return true, nil
+	}
+
+	return false, nil
+}
+
+func isEnterprise(v semver.Version) bool {
+	return v.GTE(scyllaEnterpriseMinimalVersion)
+}
+
+func isOpenSource(v semver.Version) bool {
+	return v.LT(scyllaEnterpriseMinimalVersion)
+}

pkg/scyllafeatures/scyllafeatures.go

@@ -246,8 +246,20 @@ func (s *ScyllaConfig) setupEntrypoint(ctx context.Context) (*exec.Cmd, error) {
 	}
 	// If node is being replaced
 	if addr, ok := m.ServiceLabels[naming.ReplaceLabel]; ok {
-		args["replace-address-first-boot"] = pointer.StringPtr(addr)
+		if len(addr) == 0 {
+			klog.Warningf("Service %q have unexpectedly empty label %q, skipping replace", m.Name, naming.ReplaceLabel)
+		} else {
+			args["replace-address-first-boot"] = pointer.StringPtr(addr)
+		}
+	}
+	if hostID, ok := m.ServiceLabels[naming.ReplacingNodeHostIDLabel]; ok {
+		if len(hostID) == 0 {
+			klog.Warningf("Service %q have unexpectedly empty label %q, skipping replace", m.Name, naming.ReplacingNodeHostIDLabel)
+		} else {
+			args["replace-node-first-boot"] = pointer.String(hostID)
+		}
 	}
+
 	// See if we need to use cpu-pinning
 	// TODO: Add more checks to make sure this is valid.
 	// eg. parse the cpuset and check the number of cpus is the same as cpu limits

pkg/sidecar/config/config.go

@@ -10,6 +10,7 @@ import (
 	g "github.com/onsi/ginkgo/v2"
 	o "github.com/onsi/gomega"
 	"github.com/scylladb/scylla-operator/pkg/naming"
+	"github.com/scylladb/scylla-operator/pkg/scyllaclient"
 	scyllafixture "github.com/scylladb/scylla-operator/test/e2e/fixture/scylla"
 	"github.com/scylladb/scylla-operator/test/e2e/framework"
 	"github.com/scylladb/scylla-operator/test/e2e/utils"
@@ -18,17 +19,61 @@ import (
 	"k8s.io/apimachinery/pkg/types"
 )
 
-var _ = g.Describe("ScyllaCluster replace", func() {
+var _ = g.Describe("ScyllaCluster", func() {
 	defer g.GinkgoRecover()
 
 	f := framework.NewFramework("scyllacluster")
 
-	g.It("should replace a node", func() {
+	const (
+		scyllaOSImageRepository         = "docker.io/scylladb/scylla"
+		scyllaEnterpriseImageRepository = "docker.io/scylladb/scylla-enterprise"
+	)
+
+	validateReplaceViaClusterIPAddress := func(ctx context.Context, configClient *scyllaclient.ConfigClient, preReplaceService *corev1.Service) error {
+		replaceAddressFirstBoot, err := configClient.ReplaceAddressFirstBoot(ctx)
+		if err != nil {
+			return fmt.Errorf("can't get replace_address_first_boot config parameter: %w", err)
+		}
+
+		if replaceAddressFirstBoot != preReplaceService.Spec.ClusterIP {
+			return fmt.Errorf("unexpected value of replace_address_first_boot scylla config, expected %q, got %q", preReplaceService.Spec.ClusterIP, replaceAddressFirstBoot)
+		}
+
+		return nil
+	}
+
+	validateReplaceViaHostID := func(ctx context.Context, configClient *scyllaclient.ConfigClient, preReplaceService *corev1.Service) error {
+		replaceNodeFirstBoot, err := configClient.ReplaceNodeFirstBoot(ctx)
+		if err != nil {
+			return fmt.Errorf("can't get replace_node_first_boot config parameter: %w", err)
+		}
+
+		if replaceNodeFirstBoot != preReplaceService.Annotations[naming.HostIDAnnotation] {
+			return fmt.Errorf("unexpected value of replace_node_first_boot scylla config, expected %q, got %q", preReplaceService.Annotations[naming.HostIDAnnotation], replaceNodeFirstBoot)
+		}
+
+		return nil
+	}
+
+	type entry struct {
+		procedure             string
+		scyllaImageRepository string
+		scyllaVersion         string
+		validateScyllaConfig  func(context.Context, *scyllaclient.ConfigClient, *corev1.Service) error
+	}
+
+	describeEntry := func(e *entry) string {
+		return fmt.Sprintf(`using %s based procedure when version of ScyllaDB is "%s:%s"`, e.procedure, e.scyllaImageRepository, e.scyllaVersion)
+	}
+
+	g.DescribeTable("should replace a node", func(e *entry) {
 		ctx, cancel := context.WithTimeout(context.Background(), testTimeout)
 		defer cancel()
 
 		sc := scyllafixture.BasicScyllaCluster.ReadOrFail()
-		sc.Spec.Datacenter.Racks[0].Members = 2
+		sc.Spec.Repository = e.scyllaImageRepository
+		sc.Spec.Version = e.scyllaVersion
+		sc.Spec.Datacenter.Racks[0].Members = 3
 
 		framework.By("Creating a ScyllaCluster")
 		sc, err := f.ScyllaClient().ScyllaV1().ScyllaClusters(f.Namespace()).Create(ctx, sc, metav1.CreateOptions{})
@@ -42,10 +87,15 @@ var _ = g.Describe("ScyllaCluster replace", func() {
 
 		verifyScyllaCluster(ctx, f.KubeClient(), sc)
 		hosts := getScyllaHostsAndWaitForFullQuorum(ctx, f.KubeClient().CoreV1(), sc)
-		o.Expect(hosts).To(o.HaveLen(2))
+		o.Expect(hosts).To(o.HaveLen(int(utils.GetMemberCount(sc))))
 		di := insertAndVerifyCQLData(ctx, hosts)
 		defer di.Close()
 
+		replacedNodeService, err := f.KubeClient().CoreV1().Services(sc.Namespace).Get(ctx, utils.GetNodeName(sc, 0), metav1.GetOptions{})
+		o.Expect(err).NotTo(o.HaveOccurred())
+
+		preReplaceService := replacedNodeService.DeepCopy()
+
 		framework.By("Replacing a node #0")
 		pod, err := f.KubeClient().CoreV1().Pods(f.Namespace()).Get(
 			ctx,
@@ -91,7 +141,7 @@ var _ = g.Describe("ScyllaCluster replace", func() {
 		client, _, err := utils.GetScyllaClient(ctx, f.KubeClient().CoreV1(), sc)
 		o.Expect(err).NotTo(o.HaveOccurred())
 
-		replacedNodeService, err := f.KubeClient().CoreV1().Services(sc.Namespace).Get(ctx, utils.GetNodeName(sc, 0), metav1.GetOptions{})
+		replacedNodeService, err = f.KubeClient().CoreV1().Services(sc.Namespace).Get(ctx, utils.GetNodeName(sc, 0), metav1.GetOptions{})
 		o.Expect(err).NotTo(o.HaveOccurred())
 
 		otherNodeService, err := f.KubeClient().CoreV1().Services(sc.Namespace).Get(ctx, utils.GetNodeName(sc, 1), metav1.GetOptions{})
@@ -106,9 +156,43 @@ var _ = g.Describe("ScyllaCluster replace", func() {
 		oldHosts := hosts
 		hosts = getScyllaHostsAndWaitForFullQuorum(ctx, f.KubeClient().CoreV1(), sc)
 		o.Expect(hosts).To(o.HaveLen(len(oldHosts)))
-		o.Expect(hosts).NotTo(o.ConsistOf(oldHosts))
 		err = di.SetClientEndpoints(hosts)
 		o.Expect(err).NotTo(o.HaveOccurred())
 		verifyCQLData(ctx, di)
-	})
+
+		framework.By("Verifying ScyllaDB config")
+
+		configClient, err := utils.GetScyllaConfigClient(ctx, f.KubeClient().CoreV1(), sc, replacedNodeService.Spec.ClusterIP)
+		o.Expect(err).NotTo(o.HaveOccurred())
+
+		err = e.validateScyllaConfig(ctx, configClient, preReplaceService)
+		o.Expect(err).NotTo(o.HaveOccurred())
+	},
+		g.Entry(describeEntry, &entry{
+			procedure:             "ClusterIP",
+			scyllaImageRepository: scyllaOSImageRepository,
+			scyllaVersion:         "5.1.15",
+			validateScyllaConfig:  validateReplaceViaClusterIPAddress,
+		}),
+		g.Entry(describeEntry, &entry{
+			procedure:             "ClusterIP",
+			scyllaImageRepository: scyllaEnterpriseImageRepository,
+			scyllaVersion:         "2022.2.12",
+			validateScyllaConfig:  validateReplaceViaClusterIPAddress,
+		}),
+		g.Entry(describeEntry, &entry{
+			procedure:             "HostID",
+			scyllaImageRepository: scyllaOSImageRepository,
+			scyllaVersion:         "5.2.0",
+			validateScyllaConfig:  validateReplaceViaHostID,
+		}),
+		// TODO: Enable test when ScyllaDB Enterprise 2023.1 is released
+		//       Ref: https://github.com/scylladb/scylla-operator/issues/1325
+		g.PEntry(describeEntry, &entry{
+			procedure:             "HostID",
+			scyllaImageRepository: scyllaEnterpriseImageRepository,
+			scyllaVersion:         "2023.1.0",
+			validateScyllaConfig:  validateReplaceViaHostID,
+		}),
+	)
 })