From 5a72e5e87e72130373cf19510b943529f2c599f7 Mon Sep 17 00:00:00 2001
From: Valentin Rothberg <vrothberg@redhat.com>
Date: Tue, 28 Mar 2023 14:54:26 +0200
Subject: [PATCH] auto-update: stop+start instead of restart sytemd units

It turns out the restart is _not_ a stop+start but keeps certain
resources open and is subject to some timeouts that may differ across
distributions' default settings.

[NO NEW TESTS NEEDED] as I have absolutely no idea how to reliably cause
the failure/flake/race.

Fixes: #17607
Signed-off-by: Valentin Rothberg <vrothberg@redhat.com>
---
 pkg/autoupdate/autoupdate.go | 33 +++++++++++++++++++++++++++++++--
 1 file changed, 31 insertions(+), 2 deletions(-)

diff --git a/pkg/autoupdate/autoupdate.go b/pkg/autoupdate/autoupdate.go
index 10258701f8d9..ccd81d8cdbce 100644
--- a/pkg/autoupdate/autoupdate.go
+++ b/pkg/autoupdate/autoupdate.go
@@ -334,8 +334,16 @@ func (t *task) rollbackImage() error {
 
 // restartSystemdUnit restarts the systemd unit the container is running in.
 func (u *updater) restartSystemdUnit(ctx context.Context, unit string) error {
+	if err := u.stopSystemdUnit(ctx, unit); err != nil {
+		return err
+	}
+	return u.startSystemdUnit(ctx, unit)
+}
+
+// startSystemdUnit starts the systemd unit the container is running in.
+func (u *updater) startSystemdUnit(ctx context.Context, unit string) error {
 	restartChan := make(chan string)
-	if _, err := u.conn.RestartUnitContext(ctx, unit, "replace", restartChan); err != nil {
+	if _, err := u.conn.StartUnitContext(ctx, unit, "replace", restartChan); err != nil {
 		return err
 	}
 
@@ -349,7 +357,28 @@ func (u *updater) restartSystemdUnit(ctx context.Context, unit string) error {
 		return nil
 
 	default:
-		return fmt.Errorf("expected %q but received %q", "done", result)
+		return fmt.Errorf("error starting systemd unit %q expected %q but received %q", unit, "done", result)
+	}
+}
+
+// stopSystemdUnit stop the systemd unit the container is running in.
+func (u *updater) stopSystemdUnit(ctx context.Context, unit string) error {
+	restartChan := make(chan string)
+	if _, err := u.conn.StopUnitContext(ctx, unit, "replace", restartChan); err != nil {
+		return err
+	}
+
+	// Wait for the restart to finish and actually check if it was
+	// successful or not.
+	result := <-restartChan
+
+	switch result {
+	case "done":
+		logrus.Infof("Successfully stopped systemd unit %q", unit)
+		return nil
+
+	default:
+		return fmt.Errorf("error stopping systemd unit %q expected %q but received %q", unit, "done", result)
 	}
 }