constellation create azure: use custom poller to check for scale set creation (#394)

Signed-off-by: Malte Poll <mp@edgeless.systems>

cli/internal/azure/client/api.go

@@ -36,6 +36,9 @@ type loadBalancersAPI interface {
 }
 
 type scaleSetsAPI interface {
+	Get(ctx context.Context, resourceGroupName string, vmScaleSetName string,
+		options *armcomputev2.VirtualMachineScaleSetsClientGetOptions,
+	) (armcomputev2.VirtualMachineScaleSetsClientGetResponse, error)
 	BeginCreateOrUpdate(ctx context.Context, resourceGroupName string,
 		vmScaleSetName string, parameters armcomputev2.VirtualMachineScaleSet,
 		options *armcomputev2.VirtualMachineScaleSetsClientBeginCreateOrUpdateOptions) (

cli/internal/azure/client/api_test.go

@@ -136,6 +136,16 @@ type stubScaleSetsAPI struct {
 	createErr    error
 	stubResponse armcomputev2.VirtualMachineScaleSetsClientCreateOrUpdateResponse
 	pollErr      error
+	getResponse  armcomputev2.VirtualMachineScaleSet
+	getErr       error
+}
+
+func (a stubScaleSetsAPI) Get(ctx context.Context, resourceGroupName string, vmScaleSetName string,
+	options *armcomputev2.VirtualMachineScaleSetsClientGetOptions,
+) (armcomputev2.VirtualMachineScaleSetsClientGetResponse, error) {
+	return armcomputev2.VirtualMachineScaleSetsClientGetResponse{
+		VirtualMachineScaleSet: a.getResponse,
+	}, a.getErr
 }
 
 func (a stubScaleSetsAPI) BeginCreateOrUpdate(ctx context.Context, resourceGroupName string,

cli/internal/azure/client/compute.go

@@ -4,15 +4,22 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"net/http"
 	"strconv"
 	"sync"
+	"time"
 
+	"github.com/Azure/azure-sdk-for-go/sdk/azcore"
 	"github.com/Azure/azure-sdk-for-go/sdk/azcore/runtime"
 	"github.com/Azure/azure-sdk-for-go/sdk/resourcemanager/resources/armresources"
 	"github.com/edgelesssys/constellation/cli/internal/azure"
+	"github.com/edgelesssys/constellation/cli/internal/azure/internal/poller"
 	"github.com/edgelesssys/constellation/internal/cloud/cloudtypes"
 )
 
+// scaleSetCreateTimeout maximum timeout to wait for scale set creation.
+const scaleSetCreateTimeout = 5 * time.Minute
+
 func (c *Client) CreateInstances(ctx context.Context, input CreateInstancesInput) error {
 	// Create worker scale set
 	createWorkerInput := CreateScaleSetInput{
@@ -211,7 +218,7 @@ func (c *Client) createScaleSet(ctx context.Context, input CreateScaleSetInput)
 		LoadBalancerBackendAddressPool: input.LoadBalancerBackendAddressPool,
 	}.Azure()
 
-	poller, err := c.scaleSetsAPI.BeginCreateOrUpdate(
+	_, err = c.scaleSetsAPI.BeginCreateOrUpdate(
 		ctx, c.resourceGroup, input.Name,
 		scaleSet,
 		nil,
@@ -220,14 +227,18 @@ func (c *Client) createScaleSet(ctx context.Context, input CreateScaleSetInput)
 		return err
 	}
 
-	_, err = poller.PollUntilDone(ctx, &runtime.PollUntilDoneOptions{
-		Frequency: c.pollFrequency,
+	// use custom poller to wait for resource creation but skip waiting for OS provisioning.
+	// OS provisioning does not work reliably without the azure guest agent installed.
+	poller := poller.New[bool](&scaleSetCreationPollingHandler{
+		resourceGroup: c.resourceGroup,
+		scaleSet:      input.Name,
+		scaleSetsAPI:  c.scaleSetsAPI,
 	})
-	if err != nil {
-		return err
-	}
 
-	return nil
+	pollCtx, cancel := context.WithTimeout(ctx, scaleSetCreateTimeout)
+	defer cancel()
+	_, err = poller.PollUntilDone(pollCtx, nil)
+	return err
 }
 
 func (c *Client) getInstanceIPs(ctx context.Context, scaleSet string, count int) (cloudtypes.Instances, error) {
@@ -324,3 +335,41 @@ func (c *Client) TerminateResourceGroup(ctx context.Context) error {
 	c.controlPlaneScaleSet = ""
 	return nil
 }
+
+// scaleSetCreationPollingHandler is a custom poller used to check if a scale set was created successfully.
+type scaleSetCreationPollingHandler struct {
+	done          bool
+	resourceGroup string
+	scaleSet      string
+	scaleSetsAPI  scaleSetsAPI
+}
+
+// Done returns true if the condition is met.
+func (h *scaleSetCreationPollingHandler) Done() bool {
+	return h.done
+}
+
+// Poll checks if the scale set resource was created successfully.
+func (h *scaleSetCreationPollingHandler) Poll(ctx context.Context) error {
+	_, err := h.scaleSetsAPI.Get(ctx, h.resourceGroup, h.scaleSet, nil)
+	if err == nil {
+		h.done = true
+		return nil
+	}
+	var respErr *azcore.ResponseError
+	if errors.As(err, &respErr) && respErr.StatusCode == http.StatusNotFound {
+		// resource does not exist yet - retry later
+		return nil
+	}
+	return err
+}
+
+// Result returns the result of the poller if the condition is met.
+// If the condition is not met, an error is returned.
+func (h *scaleSetCreationPollingHandler) Result(ctx context.Context, out *bool) error {
+	if !h.done {
+		return fmt.Errorf("failed to create scale set")
+	}
+	*out = h.done
+	return nil
+}

cli/internal/azure/client/compute_test.go

@@ -167,7 +167,7 @@ func TestCreateInstances(t *testing.T) {
 		"error when polling create scale set response": {
 			publicIPAddressesAPI: stubPublicIPAddressesAPI{},
 			networkInterfacesAPI: stubNetworkInterfacesAPI{},
-			scaleSetsAPI:         stubScaleSetsAPI{pollErr: someErr},
+			scaleSetsAPI:         stubScaleSetsAPI{getErr: someErr},
 			resourceGroupAPI:     newSuccessfulResourceGroupStub(),
 			roleAssignmentsAPI:   &stubRoleAssignmentsAPI{},
 			createInstancesInput: CreateInstancesInput{

cli/internal/azure/internal/poller/poller.go

@@ -0,0 +1,121 @@
+// Package poller implements a poller that can be used to wait for a condition to be met.
+// The poller is designed to be a replacement for the azure-sdk-for-go poller
+// with exponential backoff and an injectable clock.
+// reference: https://pkg.go.dev/github.com/Azure/azure-sdk-for-go/sdk/azcore@v1.1.1/runtime#Poller .
+package poller
+
+import (
+	"context"
+	"errors"
+	"time"
+
+	"k8s.io/utils/clock"
+)
+
+// PollUntilDoneOptions provides options for the Poller.
+// Used to specify backoff and clock options.
+type PollUntilDoneOptions struct {
+	StartingBackoff time.Duration
+	MaxBackoff      time.Duration
+	clock.Clock
+}
+
+// NewPollUntilDoneOptions creates a new PollUntilDoneOptions with the default values and a real clock.
+func NewPollUntilDoneOptions() *PollUntilDoneOptions {
+	return &PollUntilDoneOptions{
+		Clock: clock.RealClock{},
+	}
+}
+
+// Poller is a poller that can be used to wait for a condition to be met.
+// The poller is designed to be a replacement for the azure-sdk-for-go poller
+// with exponential backoff and an injectable clock.
+// reference: https://pkg.go.dev/github.com/Azure/azure-sdk-for-go/sdk/azcore@v1.1.1/runtime#Poller .
+type Poller[T any] struct {
+	handler PollingHandler[T]
+	err     error
+	result  *T
+	done    bool
+}
+
+// New creates a new Poller.
+func New[T any](handler PollingHandler[T]) *Poller[T] {
+	return &Poller[T]{
+		handler: handler,
+		result:  new(T),
+	}
+}
+
+// PollUntilDone polls the handler until the condition is met or the context is cancelled.
+func (p *Poller[T]) PollUntilDone(ctx context.Context, options *PollUntilDoneOptions) (T, error) {
+	if options == nil {
+		options = NewPollUntilDoneOptions()
+	}
+	if options.MaxBackoff == 0 {
+		options.MaxBackoff = time.Minute
+	}
+	if options.StartingBackoff < time.Second {
+		options.StartingBackoff = time.Second
+	}
+	backoff := options.StartingBackoff
+	for {
+		timer := options.Clock.NewTimer(backoff)
+		err := p.Poll(ctx)
+		if err != nil {
+			return *new(T), err
+		}
+		if p.Done() {
+			return p.Result(ctx)
+		}
+		select {
+		case <-ctx.Done():
+			return *new(T), ctx.Err()
+		case <-timer.C():
+		}
+		if backoff >= options.MaxBackoff/2 {
+			backoff = options.MaxBackoff
+		} else {
+			backoff *= 2
+		}
+	}
+}
+
+// Poll polls the handler.
+func (p *Poller[T]) Poll(ctx context.Context) error {
+	return p.handler.Poll(ctx)
+}
+
+// Done returns true if the condition is met.
+func (p *Poller[T]) Done() bool {
+	return p.handler.Done()
+}
+
+// Result returns the result of the poller if the condition is met.
+// If the condition is not met, an error is returned.
+func (p *Poller[T]) Result(ctx context.Context) (T, error) {
+	if !p.Done() {
+		return *new(T), errors.New("poller is in a non-terminal state")
+	}
+	if p.done {
+		// the result has already been retrieved, return the cached value
+		if p.err != nil {
+			return *new(T), p.err
+		}
+		return *p.result, nil
+	}
+	err := p.handler.Result(ctx, p.result)
+	p.done = true
+	if err != nil {
+		p.err = err
+		return *new(T), p.err
+	}
+
+	return *p.result, nil
+}
+
+// PollingHandler is a handler that can be used to poll for a condition to be met.
+type PollingHandler[T any] interface {
+	Done() bool
+	Poll(ctx context.Context) error
+	Result(ctx context.Context, out *T) error
+}

cli/internal/azure/internal/poller/poller_test.go

@@ -0,0 +1,205 @@
+package poller
+
+import (
+	"context"
+	"errors"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/Azure/azure-sdk-for-go/sdk/azcore/to"
+	"github.com/stretchr/testify/assert"
+	testclock "k8s.io/utils/clock/testing"
+)
+
+func TestResult(t *testing.T) {
+	testCases := map[string]struct {
+		done       bool
+		pollErr    error
+		resultErr  error
+		wantErr    bool
+		wantResult int
+	}{
+		"result called before poller is done": {
+			wantErr: true,
+		},
+		"result returns error": {
+			done:      true,
+			resultErr: errors.New("result error"),
+			wantErr:   true,
+		},
+		"result succeeds": {
+			done: true,
+		},
+	}
+
+	for name, tc := range testCases {
+		t.Run(name, func(t *testing.T) {
+			assert := assert.New(t)
+
+			poller := New[int](&stubPoller[int]{
+				result:    &tc.wantResult,
+				done:      tc.done,
+				pollErr:   tc.pollErr,
+				resultErr: tc.resultErr,
+			})
+			_, firstErr := poller.Result(context.Background())
+			if tc.wantErr {
+				assert.Error(firstErr)
+				// calling Result again should return the same error
+				_, secondErr := poller.Result(context.Background())
+				assert.Equal(firstErr, secondErr)
+				return
+			}
+			assert.NoError(firstErr)
+			// calling Result again should still not return an error
+			_, secondErr := poller.Result(context.Background())
+			assert.NoError(secondErr)
+		})
+	}
+}
+
+func TestPollUntilDone(t *testing.T) {
+	testCases := map[string]struct {
+		messages   []message
+		maxBackoff time.Duration
+		resultErr  error
+		wantErr    bool
+		wantResult int
+	}{
+		"poll succeeds on first try": {
+			messages: []message{
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(true)},
+				{done: to.Ptr(true)}, // Result() will call Done() after the last poll
+			},
+			wantResult: 1,
+		},
+		"poll succeeds on fourth try": {
+			messages: []message{
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(false), backoff: time.Second},
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(false), backoff: 2 * time.Second},
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(false), backoff: 4 * time.Second},
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(true)},
+				{done: to.Ptr(true)}, // Result() will call Done() after the last poll
+			},
+			wantResult: 1,
+		},
+		"max backoff reached": {
+			messages: []message{
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(false), backoff: time.Second},
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(false), backoff: time.Second},
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(false), backoff: time.Second},
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(true)},
+				{done: to.Ptr(true)}, // Result() will call Done() after the last poll
+			},
+			maxBackoff: time.Second,
+			wantResult: 1,
+		},
+		"poll errors": {
+			messages: []message{
+				{pollErr: to.Ptr(errors.New("poll error"))},
+			},
+			wantErr: true,
+		},
+		"result errors": {
+			messages: []message{
+				{pollErr: to.Ptr[error](nil), done: to.Ptr(true)},
+				{done: to.Ptr(true)}, // Result() will call Done() after the last poll
+			},
+			resultErr: errors.New("result error"),
+			wantErr:   true,
+		},
+	}
+
+	for name, tc := range testCases {
+		t.Run(name, func(t *testing.T) {
+			assert := assert.New(t)
+
+			doneC := make(chan bool)
+			pollC := make(chan error)
+			poller := New[int](&fakePoller[int]{
+				result:    &tc.wantResult,
+				resultErr: tc.resultErr,
+				doneC:     doneC,
+				pollC:     pollC,
+			})
+			clock := testclock.NewFakeClock(time.Now())
+			var gotResult int
+			var gotErr error
+			wg := sync.WaitGroup{}
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				gotResult, gotErr = poller.PollUntilDone(context.Background(), &PollUntilDoneOptions{
+					MaxBackoff: tc.maxBackoff,
+					Clock:      clock,
+				})
+			}()
+
+			for _, msg := range tc.messages {
+				if msg.pollErr != nil {
+					pollC <- *msg.pollErr
+				}
+				if msg.done != nil {
+					doneC <- *msg.done
+				}
+				clock.Step(msg.backoff)
+			}
+
+			wg.Wait()
+			if tc.wantErr {
+				assert.Error(gotErr)
+				return
+			}
+			assert.NoError(gotErr)
+			assert.Equal(tc.wantResult, gotResult)
+		})
+	}
+}
+
+type stubPoller[T any] struct {
+	result    *T
+	done      bool
+	pollErr   error
+	resultErr error
+}
+
+func (s *stubPoller[T]) Poll(ctx context.Context) error {
+	return s.pollErr
+}
+
+func (s *stubPoller[T]) Done() bool {
+	return s.done
+}
+
+func (s *stubPoller[T]) Result(ctx context.Context, out *T) error {
+	*out = *s.result
+	return s.resultErr
+}
+
+type message struct {
+	pollErr *error
+	done    *bool
+	backoff time.Duration
+}
+
+type fakePoller[T any] struct {
+	result    *T
+	resultErr error
+
+	doneC chan bool
+	pollC chan error
+}
+
+func (s *fakePoller[T]) Poll(ctx context.Context) error {
+	return <-s.pollC
+}
+
+func (s *fakePoller[T]) Done() bool {
+	return <-s.doneC
+}
+
+func (s *fakePoller[T]) Result(ctx context.Context, out *T) error {
+	*out = *s.result
+	return s.resultErr
+}