Skip to content

Image Comparison

All three images share the same base toolkit. The only difference is which cloud CLIs sit on top.

At a glance

all-devops aws-devops gcp-devops
Shared base toolkit ✅ ✅ ✅
AWS tools ✅ ✅ —
Google Cloud tools ✅ — ✅
Compressed download ~1.6 GB ~1.55 GB ~1.5 GB
Unpacked (amd64) ~5.0 GB ~4.6 GB ~4.6 GB
Best for Multi-cloud and platform teams AWS-only work GCP-only work

Size isn't a big factor

The base is most of each image, so picking a single-cloud image saves only about 0.05 to 0.1 GB of download. Choose by the tools you need.

Tool matrix

Shared by every image

Category Tools all aws gcp
Terraform IaC Terraform, tfswitch, Terragrunt, TFLint, Packer ✅ ✅ ✅
Kubernetes Kubernetes kubectl, Helm 3, k9s ✅ ✅ ✅
Ansible Automation Ansible, ansible-lint, pre-commit, Task ✅ ✅ ✅
Trivy Security Trivy ✅ ✅ ✅
AI agents claude, codex, copilot, agy ✅ ✅ ✅
Python Languages Python 3.14, Node.js LTS (npm, npx) ✅ ✅ ✅
GitHub Git git, gh, ghorg ✅ ✅ ✅
Databases mongosh, psql 17, mysql 8.4 ✅ ✅ ✅
Shells & utilities zsh, bash, fish, jq, curl, wget, nmap, ncat, dig, vim ✅ ✅ ✅

Cloud-specific

Tool Purpose all aws gcp
AWS CLI v2 AWS operations ✅ ✅ —
Session Manager plugin Shell and port forwarding to EC2 ✅ ✅ —
boto3 AWS SDK for Python ✅ ✅ —
cfn-lint CloudFormation linter ✅ ✅ —
s3cmd, crcmod S3 client, CRC32c helper ✅ ✅ —
pytest, requests, bs4, lxml Scripting helpers ✅ ✅ —
gcloud (+ beta) Google Cloud operations ✅ — ✅
gsutil, bq Cloud Storage and BigQuery ✅ — ✅
docker-credential-gcr Registry auth helper ✅ — ✅
gke-gcloud-auth-plugin kubectl auth for GKE ✅ — ✅

Not in any image

kustomize (use kubectl kustomize or kubectl apply -k), yq, pipx, gcloud alpha, the Docker CLI/daemon, Go and Java. Install them in a derived image if you need them.

Decision tree

flowchart TD
  Q{"Which clouds?"} -->|AWS only| W["aws-devops"]
  Q -->|Google Cloud only| G["gcp-devops"]
  Q -->|Both, or not sure| A["all-devops"]
  Q -->|Neither| A

  classDef neutral fill:#334155,stroke:#1e293b,color:#fff
  classDef all fill:#059669,stroke:#047857,color:#fff
  classDef aws fill:#ea7a0c,stroke:#c2410c,color:#fff
  classDef gcp fill:#2563eb,stroke:#1d4ed8,color:#fff
  class Q neutral
  class A all
  class W aws
  class G gcp

Need no cloud CLIs at all? Any image works, or build the unpublished base yourself with docker build --target base -t devops-base:local ..

  • all-devops

    Both clouds in one image. The safe default for platform teams, migrations and mixed estates.

    Guide

  • aws-devops

    EKS, CloudFormation, Session Manager and boto3 scripting, with no Google Cloud SDK.

    Guide

  • Google Cloud gcp-devops

    GKE, Cloud Storage, BigQuery and Artifact Registry, with no AWS tooling.

    Guide

Architectures and registries

linux/amd64 linux/arm64
all-devops ✅ ✅
aws-devops ✅ ✅
gcp-devops ✅ ✅

Both architectures are built natively and in parallel, then merged into one multi-arch tag, so Docker pulls the right one automatically. Check with:

docker buildx imagetools inspect ghcr.io/jinalshah/devops/images/all-devops:latest

The three images are published to all three registries. The base is not published anywhere.

Registry Image path
GitHub GHCR (recommended) ghcr.io/jinalshah/devops/images/<image>
GitLab GitLab registry.gitlab.com/jinal-shah/devops/images/<image>
Docker Docker Hub js01/<image>

Switching images

Moving from all-devops to a single-cloud image is just a tag change. Check what your pipelines actually call first:

grep -rE "\baws |gcloud |gsutil |bq " .github/ .gitlab-ci.yml 2>/dev/null

container:
  image: ghcr.io/jinalshah/devops/images/aws-devops:latest

For reproducible pipelines, pin by digest (image@sha256:...), because 1.0.<sha> tags are refreshed by scheduled rebuilds.

Next steps