From 146f5038f551e4c7a9a6261d42b4799c3ae985bc Mon Sep 17 00:00:00 2001
From: Raj Singh <raj@tailscale.com>
Date: Sun, 26 Oct 2025 02:54:35 -0500
Subject: [PATCH] cmd/containerboot: exit with non-zero code on unexpected
 tailscaled death

When tailscaled exits unexpectedly (crashes, killed directly), containerboot now exits with a non-zero code to signal failure to the orchestrator. The reaper now distinguishes between graceful shutdowns which still exit 0, and unexpected exits which propagate the child's exit code or force 1 if the child exited cleanly on its own. Fixes #17650

Signed-off-by: Raj Singh <raj@tailscale.com>
---
 cmd/containerboot/main.go | 31 +++++++++++++++++++++++++++++--
 1 file changed, 29 insertions(+), 2 deletions(-)

diff --git a/cmd/containerboot/main.go b/cmd/containerboot/main.go
index f056d26f3..eafc77de2 100644
--- a/cmd/containerboot/main.go
+++ b/cmd/containerboot/main.go
@@ -728,8 +728,35 @@ runLoop:
 							if err != nil {
 								log.Fatalf("Waiting for tailscaled to exit: %v", err)
 							}
-							log.Print("tailscaled exited")
-							os.Exit(0)
+
+							// Distinguish between graceful shutdown and unexpected failure.
+							// Graceful shutdown (SIGTERM/SIGINT to containerboot) should exit 0
+							// so Kubernetes treats the pod as completed successfully.
+							// Unexpected failures should exit non-zero so Kubernetes restarts the pod.
+							select {
+							case <-ctx.Done():
+								// Graceful shutdown: containerboot received SIGTERM/SIGINT
+								// and told tailscaled to exit. This is expected, exit 0.
+								log.Print("tailscaled exited after graceful shutdown")
+								os.Exit(0)
+							default:
+								// Unexpected exit: tailscaled crashed or was killed directly.
+								// Exit non-zero to signal failure to the orchestrator.
+								exitCode := 1
+								if status.Exited() {
+									exitCode = status.ExitStatus()
+								} else if status.Signaled() {
+									log.Printf("tailscaled terminated by signal: %v", status.Signal())
+									exitCode = 128 + int(status.Signal())
+								}
+								log.Printf("tailscaled exited unexpectedly with code %d", exitCode)
+								if exitCode == 0 {
+									// If tailscaled exited cleanly on its own, this is still unexpected.
+									// Force non-zero to ensure the container orchestrator restarts us.
+									exitCode = 1
+								}
+								os.Exit(exitCode)
+							}
 						}
 					}
 					wg.Add(1)