Merge pull request #6866 from ambarve/cleanup_leaked_shim

Cleanup leaked shim process

runtime/v2/manager.go

@@ -38,6 +38,7 @@ import (
 	"github.com/containerd/containerd/protobuf"
 	"github.com/containerd/containerd/runtime"
 	shimbinary "github.com/containerd/containerd/runtime/v2/shim"
+	"github.com/containerd/containerd/sandbox"
 	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 )
 
@@ -78,6 +79,7 @@ func init() {
 				return nil, err
 			}
 			cs := metadata.NewContainerStore(m.(*metadata.DB))
+			ss := metadata.NewSandboxStore(m.(*metadata.DB))
 			events := ep.(*exchange.Exchange)
 
 			shimManager, err := NewShimManager(ic.Context, &ManagerConfig{
@@ -88,6 +90,7 @@ func init() {
 				Events:       events,
 				Store:        cs,
 				SchedCore:    config.SchedCore,
+				SandboxStore: ss,
 			})
 			if err != nil {
 				return nil, err
@@ -124,6 +127,7 @@ type ManagerConfig struct {
 	Address      string
 	TTRPCAddress string
 	SchedCore    bool
+	SandboxStore sandbox.Store
 }
 
 // NewShimManager creates a manager for v2 shims
@@ -143,6 +147,7 @@ func NewShimManager(ctx context.Context, config *ManagerConfig) (*ShimManager, e
 		events:                 config.Events,
 		containers:             config.Store,
 		schedCore:              config.SchedCore,
+		sandboxStore:           config.SandboxStore,
 	}
 
 	if err := m.loadExistingTasks(ctx); err != nil {
@@ -167,6 +172,7 @@ type ShimManager struct {
 	containers             containers.Store
 	// runtimePaths is a cache of `runtime names` -> `resolved fs path`
 	runtimePaths sync.Map
+	sandboxStore sandbox.Store
 }
 
 // ID of the shim manager

runtime/v2/shim_load.go

@@ -18,9 +18,11 @@ package v2
 
 import (
 	"context"
+	"errors"
 	"os"
 	"path/filepath"
 
+	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/log"
 	"github.com/containerd/containerd/mount"
 	"github.com/containerd/containerd/namespaces"
@@ -139,7 +141,26 @@ func (m *ShimManager) loadShims(ctx context.Context) error {
 			cleanupAfterDeadShim(ctx, id, ns, m.shims, m.events, binaryCall)
 			continue
 		}
-		m.shims.Add(ctx, shim)
+
+		// There are 3 possibilities for the loaded shim here:
+		// 1. It could be a shim that is running a task.
+		// 2. It could be a sandbox shim.
+		// 3. Or it could be a shim that was created for running a task but
+		// something happened (probably a containerd crash) and the task was never
+		// created. This shim process should be cleaned up here. Look at
+		// containerd/containerd#6860 for further details.
+
+		_, sgetErr := m.sandboxStore.Get(ctx, id)
+		pInfo, pidErr := shim.Pids(ctx)
+		if sgetErr != nil && errors.Is(sgetErr, errdefs.ErrNotFound) && (len(pInfo) == 0 || errors.Is(pidErr, errdefs.ErrNotFound)) {
+			log.G(ctx).WithField("id", id).Info("cleaning leaked shim process")
+			// We are unable to get Pids from the shim and it's not a sandbox
+			// shim. We should clean it up her.
+			// No need to do anything for removeTask since we never added this shim.
+			shim.delete(ctx, false, func(ctx context.Context, id string) {})
+		} else {
+			m.shims.Add(ctx, shim)
+		}
 	}
 	return nil
 }