From f20faa8ec34a20401334dc377f87b3b015b8886c Mon Sep 17 00:00:00 2001
From: Manuel Buil <mbuil@suse.com>
Date: Mon, 27 May 2024 07:29:52 +0200
Subject: [PATCH] Improve rke2-uninstall.ps1 script (#5961)

Signed-off-by: Manuel Buil <mbuil@suse.com>
---
 bundle/bin/rke2-uninstall.ps1    | 42 +++++++++++++++++++++++++++++---
 pkg/pebinaryexecutor/pebinary.go | 15 ++++++++++++
 2 files changed, 53 insertions(+), 4 deletions(-)

diff --git a/bundle/bin/rke2-uninstall.ps1 b/bundle/bin/rke2-uninstall.ps1
index 50b8b83818..4134e42871 100644
--- a/bundle/bin/rke2-uninstall.ps1
+++ b/bundle/bin/rke2-uninstall.ps1
@@ -272,6 +272,13 @@ function Remove-Containerd () {
     }
 
     if (ctr) {
+        # We create a lockfile to prevent rke2 service from starting kubelet again
+        Create-Lockfile
+        Stop-Process -Name "kubelet"
+        while (-Not(Get-Process -Name "kubelet").HasExited) {
+            Write-LogInfo "Waiting for kubelet process to stop"
+            Start-Sleep -s 5
+        }
         $namespaces = $(Find-Namespaces)
         if (-Not($namespaces)) {
             $ErrorActionPreference = 'SilentlyContinue'
@@ -292,10 +299,23 @@ function Remove-Containerd () {
             foreach ($image in $images) {
                 Remove-Image $ns $image
             }
-            Remove-Namespace $ns
-            # TODO
-            # clean pods with crictl
-            # $CONTAINER_RUNTIME_ENDPOINT = "npipe:\\.\\pipe\\containerd-containerd"
+        }
+
+        # Some resources in the namespace take a while to disappear. Try several times to remove the namespace and give up after 30s 
+        $endTime = (Get-Date).AddSeconds(30)
+        while ((Get-Date) -lt $endTime) {
+            $namespaces = $(Find-Namespaces)
+            if ($namespaces) {
+                foreach ($ns in $namespaces) {
+                    Remove-Namespace $ns
+                }
+            } else {
+                break
+            }
+            Start-Sleep -Seconds 5
+            if ((Get-Date) -ge $endTime) {
+                Write-Output "Unable to remove all namespaces"
+            }
         }
     }
     else {
@@ -345,6 +365,20 @@ function Remove-Namespace() {
     Invoke-Ctr -cmd "namespace remove $namespace"
 }
 
+function Create-Lockfile() {
+    # We fetch ctr.exe path and place the lock there
+    $command = Get-Command ctr -ErrorAction SilentlyContinue
+    if ($command) {
+        $executablePath = $command.Source
+        $dataBinDirDirectory = Split-Path -Parent $executablePath
+        $lockFilePath = Join-Path -Path $dataBinDirDirectory -ChildPath "rke2-uninstall.lock"
+        New-Item -ItemType File -Path $lockFilePath -Force
+    } else {
+	# ctr should exist at this point but just in case we add this log
+        Write-Host "ctr.exe not found, container cleanup and RKE2 uninstallation is unlikely to succeed."
+    }
+}
+
 function Invoke-Rke2Uninstall () {
     $env:PATH += ";$env:CATTLE_AGENT_BIN_PREFIX/bin/;c:\var\lib\rancher\rke2\bin"
     Remove-Containerd
diff --git a/pkg/pebinaryexecutor/pebinary.go b/pkg/pebinaryexecutor/pebinary.go
index d45ed81a46..ab1e9683b4 100644
--- a/pkg/pebinaryexecutor/pebinary.go
+++ b/pkg/pebinaryexecutor/pebinary.go
@@ -165,6 +165,15 @@ func (p *PEBinaryConfig) Kubelet(ctx context.Context, args []string) error {
 		cleanArgs = append(cleanArgs, arg)
 	}
 
+	// It should never happen but just in case, we make sure the rke2-uninstall.lock does not exist before starting kubelet
+	lockFile := filepath.Join(p.DataDir, "bin", "rke2-uninstall.lock")
+	if _, err := os.Stat(lockFile); err == nil {
+		// If the file exists, delete it
+		if err := os.Remove(lockFile); err != nil {
+			logrus.Errorf("Failed to remove the %s file: %v", lockFile, err)
+		}
+	}
+
 	win.ProcessWaitGroup.StartWithContext(ctx, func(ctx context.Context) {
 		for {
 			logrus.Infof("Running RKE2 kubelet %v", cleanArgs)
@@ -185,6 +194,12 @@ func (p *PEBinaryConfig) Kubelet(ctx context.Context, args []string) error {
 			}
 			cancel()
 
+			// If the rke2-uninstall.ps1 script created the lock file, we are removing rke2 and thus we don't restart kubelet
+			if _, err := os.Stat(lockFile); err == nil {
+				logrus.Infof("rke2-uninstall.lock exists. kubelet is not restarted")
+				return
+			}
+
 			select {
 			case <-ctx.Done():
 				return