|
| 1 | +package hibernation |
| 2 | + |
| 3 | +import ( |
| 4 | + "fmt" |
| 5 | + |
| 6 | + "github.com/aws/aws-sdk-go/aws" |
| 7 | + "github.com/aws/aws-sdk-go/service/ec2" |
| 8 | + log "github.com/sirupsen/logrus" |
| 9 | + |
| 10 | + "k8s.io/apimachinery/pkg/util/sets" |
| 11 | + "sigs.k8s.io/controller-runtime/pkg/client" |
| 12 | + |
| 13 | + hivev1 "github.com/openshift/hive/pkg/apis/hive/v1" |
| 14 | + awsclient "github.com/openshift/hive/pkg/awsclient" |
| 15 | +) |
| 16 | + |
| 17 | +var ( |
| 18 | + runningStates = sets.NewString("running") |
| 19 | + stoppedStates = sets.NewString("stopped") |
| 20 | + pendingStates = sets.NewString("pending") |
| 21 | + stoppingStates = sets.NewString("stopping", "shutting-down") |
| 22 | + runningOrPendingStates = runningStates.Union(pendingStates) |
| 23 | + stoppedOrStoppingStates = stoppedStates.Union(stoppingStates) |
| 24 | + notRunningStates = stoppedOrStoppingStates.Union(pendingStates) |
| 25 | + notStoppedStates = runningOrPendingStates.Union(stoppingStates) |
| 26 | +) |
| 27 | + |
| 28 | +func init() { |
| 29 | + RegisterActuator(&awsActuator{awsClientFn: getAWSClient}) |
| 30 | +} |
| 31 | + |
| 32 | +type awsActuator struct { |
| 33 | + // awsClientFn is the function to build an AWS client, here for testing |
| 34 | + awsClientFn func(*hivev1.ClusterDeployment, client.Client, log.FieldLogger) (awsclient.Client, error) |
| 35 | +} |
| 36 | + |
| 37 | +// CanHandle returns true if the actuator can handle a particular ClusterDeployment |
| 38 | +func (a *awsActuator) CanHandle(cd *hivev1.ClusterDeployment) bool { |
| 39 | + return cd.Spec.Platform.AWS != nil |
| 40 | +} |
| 41 | + |
| 42 | +// StopMachines will stop machines belonging to the given ClusterDeployment |
| 43 | +func (a *awsActuator) StopMachines(cd *hivev1.ClusterDeployment, c client.Client, logger log.FieldLogger) error { |
| 44 | + logger = logger.WithField("cloud", "aws") |
| 45 | + awsClient, err := a.awsClientFn(cd, c, logger) |
| 46 | + if err != nil { |
| 47 | + return err |
| 48 | + } |
| 49 | + instanceIDs, err := getClusterInstanceIDs(cd, awsClient, runningOrPendingStates, logger) |
| 50 | + if err != nil { |
| 51 | + return err |
| 52 | + } |
| 53 | + if len(instanceIDs) == 0 { |
| 54 | + logger.Warning("No instances were found to stop") |
| 55 | + return nil |
| 56 | + } |
| 57 | + logger.WithField("instanceIDs", instanceIDs).Info("Stopping cluster instances") |
| 58 | + _, err = awsClient.StopInstances(&ec2.StopInstancesInput{ |
| 59 | + InstanceIds: instanceIDs, |
| 60 | + }) |
| 61 | + if err != nil { |
| 62 | + logger.WithError(err).Error("failed to stop instances") |
| 63 | + } |
| 64 | + return err |
| 65 | +} |
| 66 | + |
| 67 | +// StartMachines will select machines belonging to the given ClusterDeployment |
| 68 | +func (a *awsActuator) StartMachines(cd *hivev1.ClusterDeployment, c client.Client, logger log.FieldLogger) error { |
| 69 | + logger = logger.WithField("cloud", "aws") |
| 70 | + awsClient, err := a.awsClientFn(cd, c, logger) |
| 71 | + if err != nil { |
| 72 | + return err |
| 73 | + } |
| 74 | + instanceIDs, err := getClusterInstanceIDs(cd, awsClient, stoppedOrStoppingStates, logger) |
| 75 | + if err != nil { |
| 76 | + return err |
| 77 | + } |
| 78 | + if len(instanceIDs) == 0 { |
| 79 | + logger.Warning("No instances were found to start") |
| 80 | + return nil |
| 81 | + } |
| 82 | + logger.WithField("instanceIDs", instanceIDs).Info("Starting cluster instances") |
| 83 | + _, err = awsClient.StartInstances(&ec2.StartInstancesInput{ |
| 84 | + InstanceIds: instanceIDs, |
| 85 | + }) |
| 86 | + if err != nil { |
| 87 | + logger.WithError(err).Error("failed to start instances") |
| 88 | + } |
| 89 | + return err |
| 90 | +} |
| 91 | + |
| 92 | +// MachinesRunning will return true if the machines associated with the given |
| 93 | +// ClusterDeployment are in a running state. |
| 94 | +func (a *awsActuator) MachinesRunning(cd *hivev1.ClusterDeployment, c client.Client, logger log.FieldLogger) (bool, error) { |
| 95 | + logger = logger.WithField("cloud", "aws") |
| 96 | + logger.Infof("checking whether machines are running") |
| 97 | + awsClient, err := a.awsClientFn(cd, c, logger) |
| 98 | + if err != nil { |
| 99 | + return false, err |
| 100 | + } |
| 101 | + instanceIDs, err := getClusterInstanceIDs(cd, awsClient, notRunningStates, logger) |
| 102 | + if err != nil { |
| 103 | + return false, err |
| 104 | + } |
| 105 | + return len(instanceIDs) == 0, nil |
| 106 | +} |
| 107 | + |
| 108 | +// MachinesStopped will return true if the machines associated with the given |
| 109 | +// ClusterDeployment are in a stopped state. |
| 110 | +func (a *awsActuator) MachinesStopped(cd *hivev1.ClusterDeployment, c client.Client, logger log.FieldLogger) (bool, error) { |
| 111 | + logger = logger.WithField("cloud", "aws") |
| 112 | + logger.Infof("checking whether machines are stopped") |
| 113 | + awsClient, err := a.awsClientFn(cd, c, logger) |
| 114 | + if err != nil { |
| 115 | + return false, err |
| 116 | + } |
| 117 | + instanceIDs, err := getClusterInstanceIDs(cd, awsClient, notStoppedStates, logger) |
| 118 | + if err != nil { |
| 119 | + return false, err |
| 120 | + } |
| 121 | + return len(instanceIDs) == 0, nil |
| 122 | +} |
| 123 | + |
| 124 | +func getAWSClient(cd *hivev1.ClusterDeployment, c client.Client, logger log.FieldLogger) (awsclient.Client, error) { |
| 125 | + awsClient, err := awsclient.NewClient(c, cd.Spec.Platform.AWS.CredentialsSecretRef.Name, cd.Namespace, cd.Spec.Platform.AWS.Region) |
| 126 | + if err != nil { |
| 127 | + logger.WithError(err).Error("failed to get AWS client") |
| 128 | + } |
| 129 | + return awsClient, err |
| 130 | +} |
| 131 | + |
| 132 | +func getClusterInstanceIDs(cd *hivev1.ClusterDeployment, c awsclient.Client, states sets.String, logger log.FieldLogger) ([]*string, error) { |
| 133 | + infraID := cd.Spec.ClusterMetadata.InfraID |
| 134 | + logger = logger.WithField("infraID", infraID) |
| 135 | + logger.Debug("listing cluster instances") |
| 136 | + out, err := c.DescribeInstances(&ec2.DescribeInstancesInput{ |
| 137 | + Filters: []*ec2.Filter{ |
| 138 | + { |
| 139 | + Name: aws.String(fmt.Sprintf("tag:kubernetes.io/cluster/%s", infraID)), |
| 140 | + Values: []*string{aws.String("owned")}, |
| 141 | + }, |
| 142 | + }, |
| 143 | + }) |
| 144 | + if err != nil { |
| 145 | + logger.WithError(err).Error("failed to list instances") |
| 146 | + return nil, err |
| 147 | + } |
| 148 | + result := []*string{} |
| 149 | + for _, r := range out.Reservations { |
| 150 | + for _, i := range r.Instances { |
| 151 | + if states.Has(aws.StringValue(i.State.Name)) { |
| 152 | + result = append(result, i.InstanceId) |
| 153 | + } |
| 154 | + } |
| 155 | + } |
| 156 | + logger.WithField("count", len(result)).WithField("states", states).Debug("result of listing instances") |
| 157 | + return result, nil |
| 158 | +} |
0 commit comments