feat: expand pool across full region to minimize capacity issues

Signed-off-by: Adrian Riobo <[email protected]>

Author: adrianriobo

cmd/mapt/cmd/aws/services/mac-pool.go

@@ -30,18 +30,12 @@ const (
 	paramMaxSizeDesc            = "max number of machines in the pool"
 	paramMaxSizeDefault         = 2
 	// Machines in the pool params
-	paramVPCID           = "vpcid"
-	paramVPCIDDesc       = "VPC Id to setup mac machines"
-	paramVPCIDDefault    = ""
-	paramAZID            = "azid"
-	paramAZIDDesc        = "AZ Id to setup mac machines"
-	paramAZIDDefault     = ""
-	paramSubnetID        = "subnetid"
-	paramSubnetIDDesc    = "Subnet Id to setup mac machines"
-	paramSubnetIDDefault = ""
-	paramSSHSGID         = "ssh-sgid"
-	paramSSHSGIDDesc     = "Security group Id to securize ssh access to machines. SSH can only be used from instances with this SG."
-	paramSSHSGIDDefault  = ""
+	paramVPCID          = "vpcid"
+	paramVPCIDDesc      = "VPC Id to setup mac machines"
+	paramVPCIDDefault   = ""
+	paramSSHSGID        = "ssh-sgid"
+	paramSSHSGIDDesc    = "Security group Id to securize ssh access to machines. SSH can only be used from instances with this SG."
+	paramSSHSGIDDefault = ""
 	// Request / Release
 	paramTicket        = "ticket"
 	paramTicketDesc    = "this is a unique identifier to tag the dedicated host meanwhile it is being locked to identify the request which locked it. It will be used on release to identify the machine being released"
@@ -164,10 +158,8 @@ func houseKeep() *cobra.Command {
 						MaxSize:         viper.GetInt(paramMaxSize),
 					},
 					Machine: &macpool.MachineRequestArgs{
-						VPCID:    viper.GetString(paramVPCID),
-						AZID:     viper.GetString(paramAZID),
-						SubnetID: viper.GetString(paramSubnetID),
-						SSHSGID:  viper.GetString(paramSSHSGID)},
+						VPCID:   viper.GetString(paramVPCID),
+						SSHSGID: viper.GetString(paramSSHSGID)},
 				}); err != nil {
 				logging.Error(err)
 			}
@@ -184,8 +176,6 @@ func houseKeep() *cobra.Command {
 	flagSet.StringP(awsParams.MACOSVersion, "", awsParams.MACOSVersion, awsParams.MACOSVersionDefault)
 	flagSet.Bool(params.Serverless, false, params.ServerlessDesc)
 	flagSet.StringP(paramVPCID, "", paramVPCIDDefault, paramVPCIDDesc)
-	flagSet.StringP(paramAZID, "", paramAZIDDefault, paramAZIDDesc)
-	flagSet.StringP(paramSubnetID, "", paramSubnetIDDefault, paramSubnetIDDesc)
 	flagSet.StringP(paramSSHSGID, "", paramSSHSGIDDefault, paramSSHSGIDDesc)
 	c.PersistentFlags().AddFlagSet(flagSet)
 	return c
@@ -237,10 +227,8 @@ func request() *cobra.Command {
 					Architecture: viper.GetString(awsParams.MACArch),
 					OSVersion:    viper.GetString(awsParams.MACOSVersion),
 					Machine: &macpool.MachineRequestArgs{
-						VPCID:    viper.GetString(paramVPCID),
-						AZID:     viper.GetString(paramAZID),
-						SubnetID: viper.GetString(paramSubnetID),
-						SSHSGID:  viper.GetString(paramSSHSGID),
+						VPCID:   viper.GetString(paramVPCID),
+						SSHSGID: viper.GetString(paramSSHSGID),
 					},
 					Ticket:  viper.GetString(paramTicket),
 					Timeout: viper.GetString(params.Timeout),
@@ -258,8 +246,6 @@ func request() *cobra.Command {
 	flagSet.StringP(awsParams.MACOSVersion, "", awsParams.MACOSVersion, awsParams.MACOSVersionDefault)
 	flagSet.StringP(params.Timeout, "", "", params.TimeoutDesc)
 	flagSet.StringP(paramVPCID, "", paramVPCIDDefault, paramVPCIDDesc)
-	flagSet.StringP(paramAZID, "", paramAZIDDefault, paramAZIDDesc)
-	flagSet.StringP(paramSubnetID, "", paramSubnetIDDefault, paramSubnetIDDesc)
 	flagSet.StringP(paramSSHSGID, "", paramSSHSGIDDefault, paramSSHSGIDDesc)
 	flagSet.StringP(paramTicket, "", paramTicketDefault, paramTicketDesc)
 	flagSet.Bool(params.Serverless, false, params.ServerlessDesc)
@@ -287,10 +273,8 @@ func release() *cobra.Command {
 					Remote:     viper.IsSet(params.Remote),
 				},
 				&macpool.MachineRequestArgs{
-					VPCID:    viper.GetString(paramVPCID),
-					AZID:     viper.GetString(paramAZID),
-					SubnetID: viper.GetString(paramSubnetID),
-					SSHSGID:  viper.GetString(paramSSHSGID),
+					VPCID:   viper.GetString(paramVPCID),
+					SSHSGID: viper.GetString(paramSSHSGID),
 				},
 				viper.GetString(paramTicket)); err != nil {
 				logging.Error(err)
@@ -300,8 +284,6 @@ func release() *cobra.Command {
 	}
 	flagSet := pflag.NewFlagSet(awsParams.MACReleaseCmd, pflag.ExitOnError)
 	flagSet.StringP(paramVPCID, "", paramVPCIDDefault, paramVPCIDDesc)
-	flagSet.StringP(paramAZID, "", paramAZIDDefault, paramAZIDDesc)
-	flagSet.StringP(paramSubnetID, "", paramSubnetIDDefault, paramSubnetIDDesc)
 	flagSet.StringP(paramSSHSGID, "", paramSSHSGIDDefault, paramSSHSGIDDesc)
 	flagSet.StringP(paramTicket, "", paramTicketDefault, paramTicketDesc)
 	flagSet.Bool(params.Serverless, false, params.ServerlessDesc)

pkg/provider/aws/action/kind/kind.go

@@ -195,6 +195,7 @@ func (r *kindRequest) deploy(ctx *pulumi.Context) error {
 		c.GetHostIP(true))
 	if len(*r.timeout) > 0 {
 		if err = serverless.OneTimeDelayedTask(ctx,
+			fmt.Sprintf("kind-destroy-%s", maptContext.RunID()),
 			*r.allocationData.Region, *r.prefix,
 			awsKindID,
 			fmt.Sprintf("aws %s destroy --project-name %s --backed-url %s --serverless --force-destroy",

pkg/provider/aws/action/mac-pool/housekeeper.go

@@ -89,10 +89,8 @@ func (r *HouseKeepRequestArgs) fillHostRequest() *macHost.PoolMacDedicatedHostRe
 			Prefix:       r.Pool.Prefix,
 			Architecture: r.Pool.Architecture,
 			// FixedLocation: r.FixedLocation,
-			VPCID:    &r.Machine.VPCID,
-			AZID:     &r.Machine.AZID,
-			SubnetID: &r.Machine.SubnetID,
-			SSHSGID:  &r.Machine.SSHSGID,
+			VPCID:   &r.Machine.VPCID,
+			SSHSGID: &r.Machine.SSHSGID,
 		},
 		PoolID: &macHost.PoolID{
 			PoolName:  r.Pool.Name,

pkg/provider/aws/action/mac-pool/mac-pool.go

@@ -42,7 +42,7 @@ func Create(ctx *maptContext.ContextArgs, r *PoolRequestArgs) error {
 }
 
 func Destroy(ctx *maptContext.ContextArgs) (err error) {
-	logging.Debug("Run fedora destroy")
+	logging.Debug("Run mac pool destroy")
 	// Create mapt Context
 	if err := maptContext.Init(ctx, aws.Provider()); err != nil {
 		return err
@@ -97,10 +97,6 @@ func (r *PoolRequestArgs) deploy(ctx *pulumi.Context) error {
 			MaxSize:         r.MaxSize,
 		})
 	return err
-	// if err != nil {
-	// 	return err
-	// }
-	// return p.RunRemoteHouseKeep(&r.Name, &r.Architecture, &r.OSVersion, &r.OfferedCapacity, &r.MaxSize)
 }
 
 func (r *PoolRequestArgs) results(stackResult auto.UpResult) error {

pkg/provider/aws/action/mac-pool/request.go

@@ -42,12 +42,8 @@ func request(ctx *maptContext.ContextArgs, r *RequestMachineArgs) error {
 		Version:      *hi.OSVersion,
 		Architecture: *hi.Arch,
 		VPCID:        &r.Machine.VPCID,
-		// Availability zone is not needed cause it will be picked
-		// from dedicated host
-		// AvailabilityZone: &r.Machine.AZID,
-		SubnetID: &r.Machine.SubnetID,
-		SSHSGID:  &r.Machine.SSHSGID,
-		Timeout:  r.Timeout,
+		SSHSGID:      &r.Machine.SSHSGID,
+		Timeout:      r.Timeout,
 	}
 
 	// TODO here we would change based on the integration-mode requested
@@ -94,8 +90,8 @@ func (r *HouseKeepRequestArgs) fillMacRequest() *macMachine.Request {
 		Architecture: r.Pool.Architecture,
 		Version:      r.Pool.OSVersion,
 		// Network and Security
-		VPCID:    &r.Machine.VPCID,
-		SubnetID: &r.Machine.SubnetID,
-		SSHSGID:  &r.Machine.SSHSGID,
+		VPCID: &r.Machine.VPCID,
+		// SubnetID: &r.Machine.SubnetID,
+		SSHSGID: &r.Machine.SSHSGID,
 	}
 }

pkg/provider/aws/action/mac-pool/types.go

@@ -28,10 +28,11 @@ type PoolRequestArgs struct {
 
 // Custom values to setup within machines in the cluster
 type MachineRequestArgs struct {
-	VPCID    string
-	AZID     string
-	SubnetID string
-	SSHSGID  string
+	VPCID string
+	// This values now will be calculated
+	// AZID     string
+	// SubnetID string
+	SSHSGID string
 }
 
 type HouseKeepRequestArgs struct {

pkg/provider/aws/action/openshift-snc/openshift-snc.go

@@ -241,6 +241,7 @@ func (r *openshiftSNCRequest) deploy(ctx *pulumi.Context) error {
 		c.GetHostIP(true))
 	if len(*r.timeout) > 0 {
 		if err = serverless.OneTimeDelayedTask(ctx,
+			fmt.Sprintf("openshift-snc-destroy-%s", maptContext.RunID()),
 			*r.allocationData.region, *r.prefix,
 			awsOCPSNCID,
 			fmt.Sprintf("aws %s destroy --project-name %s --backed-url %s --serverless",

pkg/provider/aws/aws.go

@@ -26,15 +26,11 @@ import (
 	"github.com/redhat-developer/mapt/pkg/util/maps"
 )
 
-<<<<<<< HEAD
 const pulumiLocksPath = ".pulumi/locks"
 
-type AWS struct{}
-=======
 type AWS struct {
 	cp *pulumi.ProviderResource
 }
->>>>>>> ac071298 (chore: aws mac pool allow orchestrate serverless stack, to allow several task specs)
 
 func (a *AWS) Init(backedURL string) error {
 	// Manage remote state requirements, if backedURL
@@ -53,6 +49,7 @@ func (a *AWS) Custom(ctx *pulumi.Context) (*pulumi.ProviderResource, error) {
 			&awsConfig.ProviderArgs{
 				SkipCredentialsValidation: pulumi.Bool(true),
 				SkipRequestingAccountId:   pulumi.Bool(true),
+				MaxRetries:                pulumi.Int(1),
 			})
 		if err != nil {
 			return nil, err

pkg/provider/aws/data/azs.go

@@ -26,8 +26,8 @@ func GetRandomAvailabilityZone(region string, excludedAZs []string) (*string, er
 	return azs[util.Random(len(azs)-1, 0)].ZoneName, nil
 }
 
-func GetAvailabilityZones() []string {
-	azs, err := describeAvailabilityZones("")
+func GetAvailabilityZones(regionName string) []string {
+	azs, err := describeAvailabilityZones(regionName)
 	if err != nil {
 		logging.Error(err)
 		return nil

pkg/provider/aws/data/network.go

@@ -117,3 +117,32 @@ func isPublic(client *ec2.Client, subnetID string) error {
 	}
 	return fmt.Errorf("no public subnet setup found")
 }
+
+func GetSubnetID(region, vpcID, azID *string) (*string, error) {
+	cfg, err := getConfig(*region)
+	if err != nil {
+		return nil, err
+	}
+	ec2Client := ec2.NewFromConfig(cfg)
+	output, err := ec2Client.DescribeSubnets(
+		context.TODO(),
+		&ec2.DescribeSubnetsInput{
+			Filters: []ec2types.Filter{
+				{
+					Name:   aws.String("vpc-id"),
+					Values: []string{*vpcID},
+				},
+				{
+					Name:   aws.String("availability-zone-id"),
+					Values: []string{*azID},
+				},
+			},
+		})
+	if err != nil {
+		return nil, err
+	}
+	if len(output.Subnets) != 1 {
+		return nil, fmt.Errorf("expected one subnet, found %d", len(output.Subnets))
+	}
+	return output.Subnets[0].SubnetId, nil
+}