17_check_container_images.yml - Usage Guide

Read-Only Docker

Purpose

Inventories container images installed on each host:

Column Meaning
IN_USE yes = at least one container on this host (running or stopped) uses this image; no = no current reference
CREATED_AT/AGE Docker: absolute create time. containerd/k3s: most versions have no createdAt → ctr content AGE (e.g. 6 months, 5 weeks)
SIZE Disk size
IMAGE Repo:tag (and short image id)

Runtime is detected automatically:

  • Docker if present → docker images + docker ps -a
  • crictl / containerd (K8s/k3s) if present → crictl images + crictl ps -a

If both exist, both sections are printed. If neither exists, the host is skipped / a “not found” message is shown.

Read-only; does not delete or prune images.

Host scope (important)

The inventory is only for the host(s) the playbook runs on. Each node has its own image store; seeing images on master does not show worker images. For the whole cluster, include every node in inventory and run with hosts: all (or an appropriate --limit). The same rule applies for deletion: 18_prune_unused_images.

Shared task: tasks/container_images_inventory.yml

Difference from 03

03_check_docker_containers is about container processes and port links.
17 is the image layer: which images are on disk, which are in use, when they arrived — for prune / drift / disk-full diagnosis.

Requirements

  • hosts: all (master, worker, singlenode, datanode…)
  • become: true (sudo) — crictl/containerd usually need root
  • python3 on the target (for report assembly; default on modern Ubuntu)

How to run 

ansible-playbook -i inventories/cagatayuresincom/hosts.ini playbooks/17_check_container_images.yml

# Workers only:
ansible-playbook -i inventories/musteri_a/hosts.ini playbooks/17_check_container_images.yml --limit workers

Multi-host

Each report starts with:

# HOST: <inventory name/IP>
# hostname: <server hostname>

How to read the output

  • IN_USE=yes is listed first — live / recently stopped workloads.
  • IN_USE=no → prune candidate; before deleting, check whether another node still uses the same image (separate cluster-wide kubectl get pods -A -o jsonpath='{..image}').
  • The same image can have several tags; the id column (abcdef123456) helps matching.
  • <none>:<none> / <untagged> → dangling image; usually safe to clean.
  • CREATED_AT is a hint for “when it was pulled/built”; it is not always the registry push time.

Sample (shortened) 

################################################################################
# HOST: 188.240.81.253
# hostname: cagatayuresincom
################################################################################
Runtime: crictl/containerd

=== containerd / crictl (K8s) ===
IN_USE  CREATED_AT                SIZE        IMAGE
------  ------------------------  ----------  -----
yes     2026-01-15 10:00:00 UTC   77.0MB      rancher/mirrored-metrics-server:v0.8.0  (7b9c9c4b9c)
no      2025-06-01 08:00:00 UTC   120.0MB     old/unused:1.0  (aabbccddeeff)
Summary: total=40 in_use=18 unused=22

Notes

  • Collecting dates with crictl inspecti can take a while on hosts with many images.
  • The report is host-local; to answer “is this image used anywhere in the cluster?” run 17 on every node and combine the IN_USE rows.

Deleting unused images

Candidate list / delete: 18_prune_unused_images

# Candidates only (does not delete)
ansible-playbook -i inventories/cagatayuresincom/hosts.ini playbooks/18_prune_unused_images.yml

# Delete
ansible-playbook -i inventories/cagatayuresincom/hosts.ini playbooks/18_prune_unused_images.yml \
  --extra-vars 'image_prune_confirm=true'