dilim/containerd
1
0 Fork 0
mirror of https://github.com/containerd/containerd.git synced 2026-11-06 05:17:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #11634 from ningmingxiao/v1.7_fix_event

Browse Source 
[release/1.7] cri:fix lost container exit events if they arrive before info is cached
This commit is contained in:
Samuel Karp authored and GitHub committed 2026-06-30 11:47:21 -07:00
2 parents 71bbb8c80d 2fe076ea7e
commit 7df7d64767
2 files changed
+32 -20
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

pkg/cri/sbserver/restart.go
+16 -10
View File
@@ -150,7 +150,7 @@ func (c *criService) recover(ctx context.Context) error {
for _, container := range containers {
container := container
eg.Go(func() error {
cntr, err := c.loadContainer(ctx2, container)
cntr, exitCh, pid, err := c.loadContainer(ctx2, container)
if err != nil {
log.G(ctx2).WithError(err).Errorf("Failed to load container %q", container.ID())
return nil
@@ -159,6 +159,10 @@ func (c *criService) recover(ctx context.Context) error {
if err := c.containerStore.Add(cntr); err != nil {
return fmt.Errorf("failed to add container %q to store: %w", container.ID(), err)
}
if exitCh != nil {
// Start the exit monitor. This should run after that container has been added to the container store.
c.eventMonitor.startContainerExitMonitor(context.Background(), cntr.ID, pid, exitCh)
}
if err := c.containerNameIndex.Reserve(cntr.Name, cntr.ID); err != nil {
return fmt.Errorf("failed to reserve container name %q: %w", cntr.Name, err)
}
@@ -228,7 +232,9 @@ func (c *criService) recover(ctx context.Context) error {
const loadContainerTimeout = 10 * time.Second
// loadContainer loads container from containerd and status checkpoint.
func (c *criService) loadContainer(ctx context.Context, cntr containerd.Container) (containerstore.Container, error) {
func (c *criService) loadContainer(ctx context.Context, cntr containerd.Container) (containerstore.Container, <-chan containerd.ExitStatus, uint32, error) {
var exitCh <-chan containerd.ExitStatus
var statusPid uint32
ctx, cancel := context.WithTimeout(ctx, loadContainerTimeout)
defer cancel()
id := cntr.ID()
@@ -238,15 +244,15 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
// Load container metadata.
exts, err := cntr.Extensions(ctx)
if err != nil {
return container, fmt.Errorf("failed to get container extensions: %w", err)
return container, nil, 0, fmt.Errorf("failed to get container extensions: %w", err)
}
ext, ok := exts[containerMetadataExtension]
if !ok {
return container, fmt.Errorf("metadata extension %q not found", containerMetadataExtension)
return container, nil, 0, fmt.Errorf("metadata extension %q not found", containerMetadataExtension)
}
data, err := typeurl.UnmarshalAny(ext)
if err != nil {
return container, fmt.Errorf("failed to unmarshal metadata extension %q: %w", ext, err)
return container, nil, 0, fmt.Errorf("failed to unmarshal metadata extension %q: %w", ext, err)
}
meta := data.(*containerstore.Metadata)
@@ -352,11 +358,13 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
// before status is checkpointed.
status.StartedAt = time.Now().UnixNano()
status.Pid = t.Pid()
statusPid = t.Pid()
}
// Wait for the task for exit monitor.
// wait is a long running background request, no timeout needed.
exitCh, err := t.Wait(ctrdutil.NamespacedContext())
exitCh, err = t.Wait(ctrdutil.NamespacedContext())
if err != nil {
exitCh = nil
if !errdefs.IsNotFound(err) {
return fmt.Errorf("failed to wait for task: %w", err)
}
@@ -365,9 +373,6 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
status.FinishedAt = time.Now().UnixNano()
status.ExitCode = unknownExitCode
status.Reason = unknownExitReason
} else {
// Start exit monitor.
c.eventMonitor.startContainerExitMonitor(context.Background(), id, status.Pid, exitCh)
}
case containerd.Stopped:
// Task is stopped. Update status and delete the task.
@@ -396,7 +401,8 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
if containerIO != nil {
opts = append(opts, containerstore.WithContainerIO(containerIO))
}
return containerstore.NewContainer(*meta, opts...)
container, err = containerstore.NewContainer(*meta, opts...)
return container, exitCh, statusPid, err
}
// podSandboxRecover is an additional interface implemented by podsandbox/ controller to handle
pkg/cri/server/restart.go
+16 -10
View File
@@ -90,7 +90,7 @@ func (c *criService) recover(ctx context.Context) error {
for _, container := range containers {
container := container
eg.Go(func() error {
cntr, err := c.loadContainer(ctx2, container)
cntr, exitCh, pid, err := c.loadContainer(ctx2, container)
if err != nil {
log.G(ctx2).WithError(err).Errorf("Failed to load container %q", container.ID())
return nil
@@ -99,6 +99,10 @@ func (c *criService) recover(ctx context.Context) error {
if err := c.containerStore.Add(cntr); err != nil {
return fmt.Errorf("failed to add container %q to store: %w", container.ID(), err)
}
if exitCh != nil {
// Start the exit monitor. This should run after that container has been added to the container store.
c.eventMonitor.startContainerExitMonitor(context.Background(), cntr.ID, pid, exitCh)
}
if err := c.containerNameIndex.Reserve(cntr.Name, cntr.ID); err != nil {
return fmt.Errorf("failed to reserve container name %q: %w", cntr.Name, err)
}
@@ -168,7 +172,9 @@ func (c *criService) recover(ctx context.Context) error {
const loadContainerTimeout = 10 * time.Second
// loadContainer loads container from containerd and status checkpoint.
func (c *criService) loadContainer(ctx context.Context, cntr containerd.Container) (containerstore.Container, error) {
func (c *criService) loadContainer(ctx context.Context, cntr containerd.Container) (containerstore.Container, <-chan containerd.ExitStatus, uint32, error) {
var exitCh <-chan containerd.ExitStatus
var statusPid uint32
ctx, cancel := context.WithTimeout(ctx, loadContainerTimeout)
defer cancel()
id := cntr.ID()
@@ -178,15 +184,15 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
// Load container metadata.
exts, err := cntr.Extensions(ctx)
if err != nil {
return container, fmt.Errorf("failed to get container extensions: %w", err)
return container, nil, 0, fmt.Errorf("failed to get container extensions: %w", err)
}
ext, ok := exts[containerMetadataExtension]
if !ok {
return container, fmt.Errorf("metadata extension %q not found", containerMetadataExtension)
return container, nil, 0, fmt.Errorf("metadata extension %q not found", containerMetadataExtension)
}
data, err := typeurl.UnmarshalAny(ext)
if err != nil {
return container, fmt.Errorf("failed to unmarshal metadata extension %q: %w", ext, err)
return container, nil, 0, fmt.Errorf("failed to unmarshal metadata extension %q: %w", ext, err)
}
meta := data.(*containerstore.Metadata)
@@ -292,11 +298,13 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
// before status is checkpointed.
status.StartedAt = time.Now().UnixNano()
status.Pid = t.Pid()
statusPid = t.Pid()
}
// Wait for the task for exit monitor.
// wait is a long running background request, no timeout needed.
exitCh, err := t.Wait(ctrdutil.NamespacedContext())
exitCh, err = t.Wait(ctrdutil.NamespacedContext())
if err != nil {
exitCh = nil
if !errdefs.IsNotFound(err) {
return fmt.Errorf("failed to wait for task: %w", err)
}
@@ -305,9 +313,6 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
status.FinishedAt = time.Now().UnixNano()
status.ExitCode = unknownExitCode
status.Reason = unknownExitReason
} else {
// Start exit monitor.
c.eventMonitor.startContainerExitMonitor(context.Background(), id, status.Pid, exitCh)
}
case containerd.Stopped:
// Task is stopped. Update status and delete the task.
@@ -336,7 +341,8 @@ func (c *criService) loadContainer(ctx context.Context, cntr containerd.Containe
if containerIO != nil {
opts = append(opts, containerstore.WithContainerIO(containerIO))
}
return containerstore.NewContainer(*meta, opts...)
container, err = containerstore.NewContainer(*meta, opts...)
return container, exitCh, statusPid, err
}
// loadSandbox loads sandbox from containerd.