Can I haz Kubernetes?

An image board demonstrating Codemowers Cloud: direct browser uploads, searchable OCR, object detection, and asynchronous image processing on Kubernetes.

Demo: can-i-haz-kubernetes.codemowers.io

Codemowers Cloud sandbox

Obtain a Codemowers Cloud sandbox at trial.codemowers.io. Follow the instructions on that site to:

• Install Skaffold.
• Install kubectl.
• Install the OIDC authentication plugin for kubectl.
• Configure your Kubernetes client with the sandbox kubeconfig.
• Set up skaffold.env in the project root.
• Refer to mcp.codemowers.io for platform guidance.
• Connect to mcp.driftmower.aws-us-west-2-bravo.codemowers.io.

Proceed to build locally using Docker and deploy to sandbox with:

skaffold dev

Open the URL in the frontend's startup log: Lolcatz available at https://…. Skaffold streams this log after the application starts.

To fit the sandbox quota, Skaffold leaves admin, EXIF, OCR, the tagger and the thumbnail worker off. Build and deploy every component with:

skaffold dev -p full

Once deployed, continue with the exercises, such as real-time voting, an LLM caption-correction worker, and face similarity search.

Docker Compose

docker compose up --build

Open localhost:3000. Compose supplies a local development identity (developer@localhost) with upload, comment, and board-admin access. Browser uploads require minio.localhost to resolve to 127.0.0.1; add it to /etc/hosts if needed. Enable the optional YOLO worker with docker compose build tagger-model && docker compose --profile ai up --build. Reset local data with docker compose down -v.

Architecture

Service	Language	Role
uploader	Go	Authorize direct S3 uploads, confirm posts, publish Kafka events
browse	Go	Board listing, thread view, comments
search	Rust / Rocket	Public title/OCR/tag search in Postgres
comments	Go	Post/delete comments
admin	Go	Authorize board administration and create/delete boards
tagger	Python/YOLO	Kafka consumer → auto-tag images via YOLOv8
ocr	Python/Tesseract	Kafka consumer → extract caption text into image metadata
thumbnailer	Python/libjpeg-turbo	Kafka + on-demand JPEG thumbnails cached in S3
exif	Go	Kafka consumer → extract EXIF camera, exposure, dimensions and GPS metadata
frontend	Next.js	UI and OIDC session handling

Ingress routes /api/<service> straight to each backend; only /api/auth belongs to the frontend. Browsers transfer image bytes directly to S3 with presigned URLs. Metadata lives in PostgreSQL (with PostGIS and pgvector), caches in Dragonfly, and events in Redpanda. The uploader creates the complete schema; workers only write enrichment results.

Optional workers are toggled with exif.enabled, ocr.enabled, tagger.enabled and thumbnailerWorker.enabled in the chart values. OCR and the tagger default to off.

Authentication

Passmower provisions the OIDC client registration. NextAuth owns the authorization-code/PKCE login and keeps refresh and ID tokens in its encrypted HttpOnly cookie; the browser calls each API directly with the access token. APIs verify the token's audience (public origin plus /api) and operation scopes:

Operation	Required scope
List own uploads	lolcatz:images:read
Upload or delete own images	lolcatz:images:write
Post comments	lolcatz:comments:write
Manage boards	lolcatz:boards:write and github.com:codemowers:admins membership

Browse and search are public.

Checks

docker compose up --build -d
docker compose run --build --rm frontend-tests
docker compose run --build --rm go-tests
docker compose run --build --rm search-tests
docker compose run --rm node-tools node test/integration.mjs

CI runs the complete checks and publishes images and the Helm chart to ghcr.io/<repository-owner> on v* tags.