study.

Diagnose Plugins, Disk, and Runtime Logs · containerd

Collect runtime and host evidence before considering a restart.

← containerd

Learn with progress and quizzes

First, understand the concepts.

Plugin state

plugins ls shows loaded capabilities and status. Optional plugins need not all be enabled; focus on the snapshotter and CRI features you actually use.

Content and snapshots

Image content storage differs from the filesystem snapshots used for execution. An image listing does not prove snapshot preparation or task execution will succeed.

Host resources and logs

Exhausted inodes can prevent writes even with free disk blocks. Correlate log timestamps, and do not restart services or delete data directories without understanding the cause.

Why it matters

Runtime disk usage can come from content, snapshots, logs, and other consumers. Deleting internal directories without identifying ownership can break reference consistency.

Work through the reasoning

Compare features with host state

Inspect required plugins and correlate daemon errors by time. Use df -h and df -i to distinguish bytes from inodes. One plugin error does not establish failure of every feature.

Understand references and reclamation

Images, containers, and snapshots have references. References or leases may retain content; deletion need not immediately release disk space. Use supported CLI operations on owned practice resources.

Learn by doing.

Use a personal Linux practice VM with containerd, runc, ctr, and sudo access. The nerdctl lab also needs nerdctl; the CRI lab needs crictl and a practice node with CRI enabled. Examples assume /run/containerd/containerd.sock; check the actual socket in other environments.

Run these read-only diagnostics on a personal systemd-based practice VM. If the unit name differs or systemd is absent, use your platform’s logging mechanism.

sudo ctr plugins ls
sudo ctr --namespace k8s.io images ls
sudo ctr --namespace k8s.io snapshots ls
df -h
df -i
sudo systemctl status containerd --no-pager
sudo journalctl -u containerd --since '10 minutes ago' --no-pager -n 30

How to check your result

Record the active snapshotter’s status, free disk blocks and inodes, and recent errors. A healthy system is a valid observation. Do not overwrite configuration defaults or run a global prune.

Apply what you learned

Deleting one image barely frees space. Should you immediately remove files manually?

Reveal the reasoning after trying

No. Investigate shared layers, remaining container or snapshot references, leases, garbage collection, and logs or other consumers. Manual rm in production runtime storage is not a substitute for supported reclamation.

Ready to check your understanding?

What else should you inspect when file creation fails despite free disk space?

  1. Inode usage and filesystem errors
  2. Only container name length
  3. Only image tag capitalization
  4. A Service DNS name
Reveal the reasoning after trying

Inode usage and filesystem errors

Inode exhaustion and filesystem issues can prevent writes. Compare df -h, df -i, and logs.

Official documentation · Diagnose Plugins, Disk, and Runtime Logs · Further reading · official documentation