Add timeout to plan actions

Author: ewoutp

pkg/deployment/reconcile/action.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 )
 
 // Action executes a single Plan item.
@@ -35,4 +36,6 @@ type Action interface {
 	// CheckProgress checks the progress of the action.
 	// Returns true if the action is completely finished, false otherwise.
 	CheckProgress(ctx context.Context) (bool, error)
+	// Timeout returns the amount of time after which this action will timeout.
+	Timeout() time.Duration
 }

pkg/deployment/reconcile/action_add_member.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1alpha"
 	"github.com/rs/zerolog"
@@ -64,3 +65,8 @@ func (a *actionAddMember) CheckProgress(ctx context.Context) (bool, error) {
 	// Nothing todo
 	return true, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *actionAddMember) Timeout() time.Duration {
+	return addMemberTimeout
+}

pkg/deployment/reconcile/action_cleanout_member.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1alpha"
 	"github.com/rs/zerolog"
@@ -114,3 +115,8 @@ func (a *actionCleanoutMember) CheckProgress(ctx context.Context) (bool, error)
 	// Cleanout completed
 	return true, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *actionCleanoutMember) Timeout() time.Duration {
+	return cleanoutMemberTimeout
+}

pkg/deployment/reconcile/action_remove_member.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	"github.com/pkg/errors"
 	"github.com/rs/zerolog"
@@ -94,3 +95,8 @@ func (a *actionRemoveMember) CheckProgress(ctx context.Context) (bool, error) {
 	// Nothing todo
 	return true, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *actionRemoveMember) Timeout() time.Duration {
+	return removeMemberTimeout
+}

pkg/deployment/reconcile/action_renew_tls_certificate.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1alpha"
 	"github.com/rs/zerolog"
@@ -69,3 +70,8 @@ func (a *renewTLSCertificateAction) Start(ctx context.Context) (bool, error) {
 func (a *renewTLSCertificateAction) CheckProgress(ctx context.Context) (bool, error) {
 	return true, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *renewTLSCertificateAction) Timeout() time.Duration {
+	return renewTLSCertificateTimeout
+}

pkg/deployment/reconcile/action_rotate_member.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1alpha"
 	"github.com/rs/zerolog"
@@ -116,3 +117,8 @@ func (a *actionRotateMember) CheckProgress(ctx context.Context) (bool, error) {
 	}
 	return true, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *actionRotateMember) Timeout() time.Duration {
+	return rotateMemberTimeout
+}

pkg/deployment/reconcile/action_shutdown_member.go

@@ -111,3 +111,8 @@ func (a *actionShutdownMember) CheckProgress(ctx context.Context) (bool, error)
 	// Member still not shutdown, retry soon
 	return false, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *actionShutdownMember) Timeout() time.Duration {
+	return shutdownMemberTimeout
+}

pkg/deployment/reconcile/action_upgrade_member.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1alpha"
 	"github.com/rs/zerolog"
@@ -126,3 +127,8 @@ func (a *actionUpgradeMember) CheckProgress(ctx context.Context) (bool, error) {
 	}
 	return isUpgrading, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *actionUpgradeMember) Timeout() time.Duration {
+	return upgradeMemberTimeout
+}

pkg/deployment/reconcile/action_wait_for_member_up.go

@@ -24,6 +24,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	driver "github.com/arangodb/go-driver"
 	"github.com/arangodb/go-driver/agency"
@@ -164,3 +165,8 @@ func (a *actionWaitForMemberUp) checkProgressArangoSync(ctx context.Context) (bo
 	}
 	return true, nil
 }
+
+// Timeout returns the amount of time after which this action will timeout.
+func (a *actionWaitForMemberUp) Timeout() time.Duration {
+	return waitForMemberUpTimeout
+}

pkg/deployment/reconcile/plan_executor.go

@@ -25,6 +25,7 @@ package reconcile
 import (
 	"context"
 	"fmt"
+	"time"
 
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
@@ -106,7 +107,20 @@ func (d *Reconciler) ExecutePlan(ctx context.Context) (bool, error) {
 			}
 			log.Debug().Bool("ready", ready).Msg("Action CheckProgress completed")
 			if !ready {
-				// Not ready check, come back soon
+				// Not ready yet, check timeout
+				deadline := planAction.CreationTime.Add(action.Timeout())
+				if time.Now().After(deadline) {
+					// Timeout has expired
+					log.Warn().Msg("Action not finished in time. Removing the entire plan")
+					status.Plan = api.Plan{}
+					// Save plan update
+					if err := d.context.UpdateStatus(status); err != nil {
+						log.Debug().Err(err).Msg("Failed to update CR status")
+						return false, maskAny(err)
+					}
+					return true, nil
+				}
+				// Timeout not yet expired, come back soon
 				return true, nil
 			}
 			// Continue with next action