Add a command to kill couchbase on nodes present in a cluster (#145)

* first commit

* SDKQE-3683 Add a command to kill coubhase server on nodes present in cluster

Author: CosmicSaaurabh

cmd/chaos-killcouchbase.go

@@ -0,0 +1,35 @@
+package cmd
+
+import (
+	"github.com/spf13/cobra"
+	"go.uber.org/zap"
+)
+
+var chaosKillCouchbaseCmd = &cobra.Command{
+	Use:   "kill-couchbase <cluster-id> [<node-id-or-ip> ...]",
+	Short: "Kills couchbase service on node/s present in the cluster.",
+	Args:  cobra.MinimumNArgs(1),
+	Run: func(cmd *cobra.Command, args []string) {
+		helper := CmdHelper{}
+		logger := helper.GetLogger()
+		ctx := helper.GetContext()
+
+		_, deployer, cluster := helper.IdentifyCluster(ctx, args[0])
+		nodeIdents := args[1:]
+
+		var nodeIds []string
+		for _, nodeIdent := range nodeIdents {
+			node := helper.IdentifyNode(ctx, cluster, nodeIdent)
+			nodeIds = append(nodeIds, node.GetID())
+		}
+
+		err := deployer.KillCouchbase(ctx, cluster.GetID(), nodeIds)
+		if err != nil {
+			logger.Fatal("failed to kill couchbase", zap.Error(err))
+		}
+	},
+}
+
+func init() {
+	chaosCmd.AddCommand(chaosKillCouchbaseCmd)
+}

deployment/caodeploy/deployer.go

@@ -780,3 +780,7 @@ func (d *Deployer) SetNodeRecovery(ctx context.Context, clusterID string, nodeID
 func (d *Deployer) RebalanceCluster(ctx context.Context, clusterID string, nodesToEject []string) error {
 	return errors.New("caodeploy does not support rebalance cluster")
 }
+
+func (d *Deployer) KillCouchbase(ctx context.Context, clusterID string, nodes []string) error {
+	return errors.New("caodeploy does not support killing couchbase process")
+}

deployment/clouddeploy/deployer.go

@@ -2479,3 +2479,7 @@ func (d *Deployer) SetNodeRecovery(ctx context.Context, clusterID string, nodeID
 func (d *Deployer) RebalanceCluster(ctx context.Context, clusterID string, nodesToEject []string) error {
 	return errors.New("clouddeploy does not support rebalance cluster")
 }
+
+func (d *Deployer) KillCouchbase(ctx context.Context, clusterID string, nodeIDs []string) error {
+	return errors.New("clouddeploy does not support killing couchbase process")
+}

deployment/deployer.go

@@ -150,4 +150,5 @@ type Deployer interface {
 	CreateS3Link(ctx context.Context, columnarID, linkName, region, endpoint, accessKey, secretKey string) error
 	DropLink(ctx context.Context, columnarID, linkName string) error
 	EnableDataApi(ctx context.Context, clusterID string) error
+	KillCouchbase(ctx context.Context, clusterID string, nodes []string) error
 }

deployment/dockerdeploy/deployer.go

@@ -5,6 +5,8 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"github.com/docker/docker/api/types/container"
+	"github.com/docker/docker/pkg/stdcopy"
 	"io"
 	"os"
 	"path"
@@ -1301,6 +1303,74 @@ func (d *Deployer) getNodeOTP(ctx context.Context, clusterID string, nodeId stri
 	return "", nil
 }
 
+func (d *Deployer) execInContainer(ctx context.Context, containerID string, cmd []string) error {
+	execOpts := container.ExecOptions{
+		Cmd:          cmd,
+		AttachStdout: true,
+		AttachStderr: true,
+		Tty:          false,
+		Privileged:   false,
+	}
+
+	execResp, err := d.dockerCli.ContainerExecCreate(ctx, containerID, execOpts)
+	if err != nil {
+		return fmt.Errorf("ContainerExecCreate failed: %w", err)
+	}
+
+	attachResp, err := d.dockerCli.ContainerExecAttach(ctx, execResp.ID, container.ExecAttachOptions{})
+	if err != nil {
+		return fmt.Errorf("ContainerExecAttach failed: %w", err)
+	}
+	defer attachResp.Close()
+
+	_, err = stdcopy.StdCopy(os.Stdout, os.Stderr, attachResp.Reader)
+	if err != nil {
+		return fmt.Errorf("copying output failed: %w", err)
+	}
+
+	return nil
+}
+
+/*
+KillCouchbase stops the couchbase-server process on the specified nodes.
+The runSv supervisor will automatically restart couchbase-server within 10 seconds.
+This method is used to simulate a stop and restart of the couchbase server.
+*/
+func (d *Deployer) KillCouchbase(ctx context.Context, clusterID string, nodeIDs []string) error {
+	var nodeContainerIDs []string
+
+	for _, nodeId := range nodeIDs {
+		node, err := d.getNode(ctx, clusterID, nodeId)
+		if err != nil {
+			return errors.Wrap(err, "failed to get node")
+		}
+
+		nodeContainerIDs = append(nodeContainerIDs, node.ContainerID)
+	}
+	if len(nodeIDs) == 0 {
+		clusterInfo, err := d.getCluster(ctx, clusterID)
+		if err != nil {
+			return errors.Wrap(err, "failed to get cluster info")
+		}
+
+		for _, node := range clusterInfo.Nodes {
+			nodeContainerIDs = append(nodeContainerIDs, node.ContainerID)
+		}
+	}
+
+	for _, nodeContainerID := range nodeContainerIDs {
+		d.logger.Info("killing couchbase process on node",
+			zap.String("containerID", nodeContainerID))
+
+		err := d.execInContainer(ctx, nodeContainerID, []string{"pkill", "-f", "couchbase-server"})
+		if err != nil {
+			return errors.Wrapf(err, "failed to kill couchbase process on node %s", nodeContainerID)
+		}
+	}
+
+	return nil
+}
+
 func (d *Deployer) RedeployCluster(ctx context.Context, clusterID string) error {
 	return errors.New("docker deploy does not support redeploy cluster")
 }

deployment/localdeploy/deployer.go

@@ -271,3 +271,7 @@ func (d *Deployer) SetNodeRecovery(ctx context.Context, clusterID string, nodeID
 func (d *Deployer) RebalanceCluster(ctx context.Context, clusterID string, nodesToEject []string) error {
 	return errors.New("localdeploy does not support rebalance cluster")
 }
+
+func (d *Deployer) KillCouchbase(ctx context.Context, clusterID string, nodes []string) error {
+	return errors.New("localdeploy does not support killing couchbase process")
+}