From 68cf778a4c1ea03c787cca003713d6c29475952a Mon Sep 17 00:00:00 2001
From: Cyrill Troxler <cyrilltroxler@gmail.com>
Date: Sat, 1 Jun 2024 18:03:18 +0200
Subject: [PATCH] fix: return last known pid when scaled down

When the pod is scaled down we need to return the last known pid, else
runc will be called, which will error as the checkpointed container
cannot be found. This fixes the shim being restarted during a containerd
restart.
---
 runc/task/service_zeropod.go | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/runc/task/service_zeropod.go b/runc/task/service_zeropod.go
index 83b3075..1f022e1 100644
--- a/runc/task/service_zeropod.go
+++ b/runc/task/service_zeropod.go
@@ -13,6 +13,7 @@ import (
 
 	"github.com/containerd/cgroups"
 	taskAPI "github.com/containerd/containerd/api/runtime/task/v2"
+	"github.com/containerd/containerd/api/types/task"
 	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/pkg/cri/annotations"
 	"github.com/containerd/containerd/pkg/oom"
@@ -198,6 +199,18 @@ func (w *wrapper) Exec(ctx context.Context, r *taskAPI.ExecProcessRequest) (*emp
 	return w.service.Exec(ctx, r)
 }
 
+func (w *wrapper) Pids(ctx context.Context, r *taskAPI.PidsRequest) (*taskAPI.PidsResponse, error) {
+	zeropodContainer, ok := w.getZeropodContainer(r.ID)
+	if ok && zeropodContainer.ScaledDown() {
+		log.G(ctx).Debugf("pids of scaled down container, returning last known pid: %v", zeropodContainer.Pid())
+		return &taskAPI.PidsResponse{
+			Processes: []*task.ProcessInfo{{Pid: uint32(zeropodContainer.Pid())}},
+		}, nil
+	}
+
+	return w.service.Pids(ctx, r)
+}
+
 func (w *wrapper) Delete(ctx context.Context, r *taskAPI.DeleteRequest) (*taskAPI.DeleteResponse, error) {
 	zeropodContainer, ok := w.getZeropodContainer(r.ID)
 	if !ok {