From 111856f26e3e78ea2eaaf846422edc953af2b804 Mon Sep 17 00:00:00 2001
From: Manuel Buil <mbuil@suse.com>
Date: Tue, 30 Jan 2024 15:36:48 +0100
Subject: [PATCH] Avoid race condition when deleting HNS networks

Signed-off-by: Manuel Buil <mbuil@suse.com>
---
 pkg/windows/calico.go |  2 +-
 pkg/windows/utils.go  | 22 ++++++++++++++++++++++
 2 files changed, 23 insertions(+), 1 deletion(-)

diff --git a/pkg/windows/calico.go b/pkg/windows/calico.go
index c90f4ce132..1851e0e9ac 100644
--- a/pkg/windows/calico.go
+++ b/pkg/windows/calico.go
@@ -295,7 +295,7 @@ func (c *Calico) Start(ctx context.Context) error {
 // generateCalicoNetworks creates the overlay networks for internode networking
 func (c *Calico) generateCalicoNetworks() error {
 	if err := deleteAllNetworks(); err != nil {
-		return err
+		return errors.Wrapf(err, "failed to delete all networks before bootstrapping calico")
 	}
 
 	// There are four ways to select the vxlan interface. In order of priority:
diff --git a/pkg/windows/utils.go b/pkg/windows/utils.go
index cf021709c4..6a6ed2e3ea 100644
--- a/pkg/windows/utils.go
+++ b/pkg/windows/utils.go
@@ -19,6 +19,7 @@ import (
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
 	opv1 "github.com/tigera/operator/api/v1"
+	"k8s.io/apimachinery/pkg/util/wait"
 )
 
 // createHnsNetwork creates the network that will connect nodes and returns its managementIP
@@ -101,8 +102,12 @@ func deleteAllNetworks() error {
 		return err
 	}
 
+	var ips []string
+
 	for _, network := range networks {
 		if network.Name != "nat" {
+			logrus.Debugf("Deleting network: %s before starting calico", network.Name)
+			ips = append(ips, network.ManagementIP)
 			_, err = network.Delete()
 			if err != nil {
 				return err
@@ -110,6 +115,23 @@ func deleteAllNetworks() error {
 		}
 	}
 
+	// HNS overlay networks restart the physical interface when they are deleted. Wait until it comes back before returning
+	// TODO: Replace with non-deprecated PollUntilContextTimeout when our and Kubernetes code migrate to it
+	waitErr := wait.Poll(2*time.Second, 20*time.Second, func() (bool, error) {
+		for _, ip := range ips {
+			logrus.Debugf("Calico is waiting for the interface with ip: %s to come back", ip)
+			_, err := findInterface(ip)
+			if err != nil {
+				return false, nil
+			}
+		}
+		return true, nil
+	})
+
+	if waitErr == wait.ErrWaitTimeout {
+		return fmt.Errorf("timed out waiting for the network interfaces to come back")
+	}
+
 	return nil
 }