Merge cockroachdb#113741

113741: CRDB-31650 : Add Spot instance support to roachtest r=BabuSrithar a=BabuSrithar Fixes CRDB-31650 Add UseSpot instance option. Co-authored-by: babusrithar <babusrithar@cockroachlabs.com>
nicktrav · Nov 10, 2023 · aa7c338 · aa7c338
2 parents 70157c7 + 4099752
commit aa7c338
Show file tree

Hide file tree

Showing 3 changed files with 24 additions and 8 deletions.
diff --git a/pkg/cmd/roachtest/spec/cluster_spec.go b/pkg/cmd/roachtest/spec/cluster_spec.go
@@ -94,7 +94,8 @@ type ClusterSpec struct {
 	ReusePolicy          clusterReusePolicy
 	TerminateOnMigration bool
 	UbuntuVersion        vm.UbuntuVersion
-
+	// Use a spot instance or equivalent of a cloud provider.
+	UseSpot bool
 	// FileSystem determines the underlying FileSystem
 	// to be used. The default is ext4.
 	FileSystem fileSystemType
@@ -196,6 +197,7 @@ func getGCEOpts(
 	minCPUPlatform string,
 	arch vm.CPUArch,
 	volumeType string,
+	useSpot bool,
 ) vm.ProviderOpts {
 	opts := gce.DefaultProviderOpts()
 	opts.MachineType = machineType
@@ -220,6 +222,7 @@ func getGCEOpts(
 		opts.UseMultipleDisks = !RAID0
 	}
 	opts.TerminateOnMigration = terminateOnMigration
+	opts.UseSpot = useSpot
 	if volumeType != "" {
 		opts.PDVolumeType = volumeType
 	}
@@ -426,7 +429,7 @@ func (s *ClusterSpec) RoachprodOpts(
 	case GCE:
 		providerOpts = getGCEOpts(machineType, zones, s.VolumeSize, ssdCount,
 			createVMOpts.SSDOpts.UseLocalSSD, s.RAID0, s.TerminateOnMigration,
-			s.GCE.MinCPUPlatform, vm.ParseArch(createVMOpts.Arch), s.GCE.VolumeType,
+			s.GCE.MinCPUPlatform, vm.ParseArch(createVMOpts.Arch), s.GCE.VolumeType, s.UseSpot,
 		)
 	case Azure:
 		providerOpts = getAzureOpts(machineType, zones)

diff --git a/pkg/cmd/roachtest/spec/option.go b/pkg/cmd/roachtest/spec/option.go
@@ -178,6 +178,19 @@ func TerminateOnMigration() Option {
 	}
 }
 
+// UseSpotInstances creates a spot instance or equivalent of a cloud provider.
+// Using this option creates SpotVMs instead of on demand VMS. SpotVMS are
+// cheaper but can be terminated at any time by the cloud provider.
+// This option is only supported by GCE for now.
+// See https://cloud.google.com/compute/docs/instances/spot,
+// https://azure.microsoft.com/en-in/products/virtual-machines/spot
+// and https://aws.amazon.com/ec2/spot/ for more details.
+func UseSpotInstances() Option {
+	return func(spec *ClusterSpec) {
+		spec.UseSpot = true
+	}
+}
+
 // SetFileSystem is an Option which can be used to set
 // the underlying file system to be used.
 func SetFileSystem(fs fileSystemType) Option {

diff --git a/pkg/roachprod/vm/gce/gcloud.go b/pkg/roachprod/vm/gce/gcloud.go
@@ -271,9 +271,9 @@ func DefaultProviderOpts() *ProviderOpts {
 		PDVolumeType:         "pd-ssd",
 		PDVolumeSize:         500,
 		TerminateOnMigration: false,
+		UseSpot:              false,
 		useSharedUser:        true,
 		preemptible:          false,
-		useSpot:              false,
 	}
 }
 
@@ -295,17 +295,17 @@ type ProviderOpts struct {
 	PDVolumeType     string
 	PDVolumeSize     int
 	UseMultipleDisks bool
+	// use spot instances (i.e., latest version of preemptibles which can run > 24 hours)
+	UseSpot bool
+
 	// GCE allows two availability policies in case of a maintenance event (see --maintenance-policy via gcloud),
 	// 'TERMINATE' or 'MIGRATE'. The default is 'MIGRATE' which we denote by 'TerminateOnMigration == false'.
 	TerminateOnMigration bool
-
 	// useSharedUser indicates that the shared user rather than the personal
 	// user should be used to ssh into the remote machines.
 	useSharedUser bool
 	// use preemptible instances
 	preemptible bool
-	// use spot instances (i.e., latest version of preemptibles which can run > 24 hours)
-	useSpot bool
 }
 
 // Provider is the GCE implementation of the vm.Provider interface.
@@ -835,7 +835,7 @@ func (o *ProviderOpts) ConfigureCreateFlags(flags *pflag.FlagSet) {
 			strings.Join(defaultZones, ",")))
 	flags.BoolVar(&o.preemptible, ProviderName+"-preemptible", false,
 		"use preemptible GCE instances (lifetime cannot exceed 24h)")
-	flags.BoolVar(&o.useSpot, ProviderName+"-use-spot", false,
+	flags.BoolVar(&o.UseSpot, ProviderName+"-use-spot", false,
 		"use spot GCE instances (like preemptible but lifetime can exceed 24h)")
 	flags.BoolVar(&o.TerminateOnMigration, ProviderName+"-terminateOnMigration", false,
 		"use 'TERMINATE' maintenance policy (for GCE live migrations)")
@@ -1046,7 +1046,7 @@ func (p *Provider) Create(
 		// Preemptible instances require the following arguments set explicitly
 		args = append(args, "--maintenance-policy", "TERMINATE")
 		args = append(args, "--no-restart-on-failure")
-	} else if providerOpts.useSpot {
+	} else if providerOpts.UseSpot {
 		args = append(args, "--provisioning-model", "SPOT")
 	} else {
 		if providerOpts.TerminateOnMigration {