Add the notion of a "permanent" error to the kubelet plugin

For now just mark one well-known erro as permanent. Future commits will
abstract this better and mark more errors as permanaent.

Signed-off-by: Kevin Klues <[email protected]>

cmd/compute-domain-kubelet-plugin/device_state.go

@@ -382,7 +382,7 @@ func (s *DeviceState) applyComputeDomainChannelConfig(ctx context.Context, confi
 	for _, r := range results {
 		channel := s.allocatable[r.Device].Channel
 		if err := s.computeDomainManager.AssertComputeDomainNamespace(ctx, claim.Namespace, config.DomainID); err != nil {
-			return nil, fmt.Errorf("error asserting ComputeDomain's namespace: %w", err)
+			return nil, permanentError{fmt.Errorf("error asserting ComputeDomain's namespace: %w", err)}
 		}
 		if err := s.computeDomainManager.AddNodeLabel(ctx, config.DomainID); err != nil {
 			return nil, fmt.Errorf("error adding Node label for ComputeDomain: %w", err)

cmd/compute-domain-kubelet-plugin/driver.go

@@ -18,6 +18,7 @@ package main
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"sync"
 	"time"
@@ -31,6 +32,15 @@ import (
 	"github.com/NVIDIA/k8s-dra-driver-gpu/pkg/workqueue"
 )
 
+// ErrorRetryMaxTimeout is the max amount of time we retry when errors are
+// returned before giving up.
+const ErrorRetryMaxTimeout = 45 * time.Second
+
+// permanentError defines an error indicating that it is permanent.
+// By default, every error will be retried up to ErrorRetryMaxTimeout.
+// Errors marked as permament will not be retried.
+type permanentError struct{ error }
+
 var _ drapbv1.DRAPluginServer = &driver{}
 
 type driver struct {
@@ -98,19 +108,19 @@ func (d *driver) NodePrepareResources(ctx context.Context, req *drapbv1.NodePrep
 	preparedResources := &drapbv1.NodePrepareResourcesResponse{Claims: map[string]*drapbv1.NodePrepareResourceResponse{}}
 
 	var wg sync.WaitGroup
-	ctx, cancel := context.WithTimeout(ctx, 45*time.Second)
+	ctx, cancel := context.WithTimeout(ctx, ErrorRetryMaxTimeout)
 	workQueue := workqueue.New(workqueue.DefaultControllerRateLimiter())
 
 	for _, claim := range req.Claims {
 		wg.Add(1)
 		workQueue.EnqueueRaw(claim, func(ctx context.Context, obj any) error {
-			prepared := d.nodePrepareResource(ctx, claim)
-			if prepared.Error != "" {
-				return fmt.Errorf("%s", prepared.Error)
+			done, prepared := d.nodePrepareResource(ctx, claim)
+			if done {
+				preparedResources.Claims[claim.UID] = prepared
+				wg.Done()
+				return nil
 			}
-			preparedResources.Claims[claim.UID] = prepared
-			wg.Done()
-			return nil
+			return fmt.Errorf("%s", prepared.Error)
 		})
 	}
 
@@ -129,19 +139,19 @@ func (d *driver) NodeUnprepareResources(ctx context.Context, req *drapbv1.NodeUn
 	unpreparedResources := &drapbv1.NodeUnprepareResourcesResponse{Claims: map[string]*drapbv1.NodeUnprepareResourceResponse{}}
 
 	var wg sync.WaitGroup
-	ctx, cancel := context.WithTimeout(ctx, 45*time.Second)
+	ctx, cancel := context.WithTimeout(ctx, ErrorRetryMaxTimeout)
 	workQueue := workqueue.New(workqueue.DefaultControllerRateLimiter())
 
 	for _, claim := range req.Claims {
 		wg.Add(1)
 		workQueue.EnqueueRaw(claim, func(ctx context.Context, obj any) error {
-			unprepared := d.nodeUnprepareResource(ctx, claim)
-			if unprepared.Error != "" {
-				return fmt.Errorf("%s", unprepared.Error)
+			done, unprepared := d.nodeUnprepareResource(ctx, claim)
+			if done {
+				unpreparedResources.Claims[claim.UID] = unprepared
+				wg.Done()
+				return nil
 			}
-			unpreparedResources.Claims[claim.UID] = unprepared
-			wg.Done()
-			return nil
+			return fmt.Errorf("%s", unprepared.Error)
 		})
 	}
 
@@ -155,7 +165,7 @@ func (d *driver) NodeUnprepareResources(ctx context.Context, req *drapbv1.NodeUn
 	return unpreparedResources, nil
 }
 
-func (d *driver) nodePrepareResource(ctx context.Context, claim *drapbv1.Claim) *drapbv1.NodePrepareResourceResponse {
+func (d *driver) nodePrepareResource(ctx context.Context, claim *drapbv1.Claim) (bool, *drapbv1.NodePrepareResourceResponse) {
 	d.Lock()
 	defer d.Unlock()
 
@@ -164,33 +174,43 @@ func (d *driver) nodePrepareResource(ctx context.Context, claim *drapbv1.Claim)
 		claim.Name,
 		metav1.GetOptions{})
 	if err != nil {
-		return &drapbv1.NodePrepareResourceResponse{
+		ret := &drapbv1.NodePrepareResourceResponse{
 			Error: fmt.Sprintf("failed to fetch ResourceClaim %s in namespace %s", claim.Name, claim.Namespace),
 		}
+		return isPermanentError(err), ret
 	}
 
 	prepared, err := d.state.Prepare(ctx, resourceClaim)
 	if err != nil {
-		return &drapbv1.NodePrepareResourceResponse{
+		ret := &drapbv1.NodePrepareResourceResponse{
 			Error: fmt.Sprintf("error preparing devices for claim %v: %v", claim.UID, err),
 		}
+		return isPermanentError(err), ret
 	}
 
 	klog.Infof("Returning newly prepared devices for claim '%v': %v", claim.UID, prepared)
-	return &drapbv1.NodePrepareResourceResponse{Devices: prepared}
+	return true, &drapbv1.NodePrepareResourceResponse{Devices: prepared}
 }
 
-func (d *driver) nodeUnprepareResource(ctx context.Context, claim *drapbv1.Claim) *drapbv1.NodeUnprepareResourceResponse {
+func (d *driver) nodeUnprepareResource(ctx context.Context, claim *drapbv1.Claim) (bool, *drapbv1.NodeUnprepareResourceResponse) {
 	d.Lock()
 	defer d.Unlock()
 
 	if err := d.state.Unprepare(ctx, claim.UID); err != nil {
-		return &drapbv1.NodeUnprepareResourceResponse{
+		ret := &drapbv1.NodeUnprepareResourceResponse{
 			Error: fmt.Sprintf("error unpreparing devices for claim %v: %v", claim.UID, err),
 		}
+		return isPermanentError(err), ret
 	}
 
-	return &drapbv1.NodeUnprepareResourceResponse{}
+	return true, &drapbv1.NodeUnprepareResourceResponse{}
+}
+
+func isPermanentError(err error) bool {
+	if errors.As(err, &permanentError{}) {
+		return true
+	}
+	return false
 }
 
 // TODO: implement loop to remove CDI files from the CDI path for claimUIDs