One Paperless-ngx instance, SQLite, and Valkey in the paperless-ngx namespace. Documents, database, index, and inbox are on one CephFS PVC. The app is served at https://paper.brunner.ninja through Traefik. Back up the PVC and the paperless-secret-key Secret together.

Install

Run ./install.sh from any directory. It creates the namespace and random, persistent application and OIDC secrets on first install, creates the Authentik applications, checks the manifests with a server dry run, applies them, then waits for the deployments. Open the URL and sign in with the brunner.ninja Auth button. Drop scanned files in the web UI or into /usr/src/paperless/consume on the PVC.

Change settings in app.yaml, then run ./test.sh and ./install.sh. The Paperless image is pinned there. The CephFS inbox uses polling because network filesystems may not deliver file notifications.

Paperless prefers the 24-core arschrock node for OCR. This is a scheduler preference, so it can run on an Odroid when arschrock is unavailable. OCR is capped at four concurrent pages and the pod has a 4 GiB memory limit; using the host's default 24 OCR workers caused an out-of-memory restart while editing a large PDF. The control-plane taint is tolerated only by Paperless, not Valkey. A node failure may briefly interrupt OCR while Kubernetes reschedules the single Paperless pod.

The 20 Gi PVC uses rook-cephfs-ec, whose CephFS data pool is erasure coded with three data and two coding chunks across hosts. CephFS metadata uses three replicas. The claim can be expanded if high-resolution scans outgrow it; erasure coding does not replace backups.

Scanner

The ScanservJS web UI is at https://scanner.brunner.ninja. Point its DNS CNAME at the same target as paper.brunner.ninja. Traefik requires Authentik login and access to the Paperless Users group. The scanner's eSCL address is set once as AIRSCAN_DEVICES in app.yaml; update it there if the printer gets a new IP. The UI pulls scans directly from the HP OfficeJet Pro 8020 at 192.168.4.189:8080, without HP cloud services or a running laptop.

Load the automatic document feeder, choose the HP scanner and its feeder source, select Auto batch mode, then press Scan. Review the staged result in the Files view, select its row, then choose Action selected → Send to Paperless. ScanservJS may save a multipage JPG batch as one ZIP row; the action converts its ordered images into one multipage PDF before placing it in the existing Paperless consume directory. PDF and single-image results also work. Paperless polls the directory every 30 seconds and handles OCR. The scan remains in staging until you send or delete it. ScanservJS does not offer reliable individual-page deletion or retry before sending: rescan the batch, or edit/split the PDF after import in Paperless. The OfficeJet Pro 8020's feeder is single-sided. This workflow starts scans from the web UI rather than a button on the printer panel.

ScanservJS stores staged scans in the scanner-staging directory on the same retained CephFS claim. The deployment uses the pinned upstream image; its config.local.js action is in app.yaml and has a local handoff test in test.sh. The Authentik proxy application is created by setup-authentik.sh along with Paperless's OIDC application.

To merge separate Paperless documents imported from JPG scans, enable Try to include archive version in merge for non-PDF files in the merge dialog. Paperless's default merge tries to open the original JPG as a PDF and skips it. Enable Delete original documents after successful merge if you want only the merged entry in the active document list; the source entries go to Paperless's trash. A single feeder batch sent through the scanner action already becomes one document, without a later merge.

CI/CD

./test.sh checks syntax, the deployment manifest, and the scanner PDF handoff. ./deploy.sh updates the existing workloads, services, ConfigMap, and ingress using app.yaml. The GitLab and Gitea workflows call these same scripts. Bootstrap once with ./create-ci-kubeconfig.sh, then save its single output line as the protected CI variable / Gitea Actions secret KUBE_CONFIG_BASE64. Do not commit it. The deployer has named-object read and patch permissions and cannot read Secrets. Run ./install.sh manually for first installation or changes to storage.yaml or Authentik setup.

The Git remote is gitea@brunner.ninja:feedc0de/paperless-ngx-deployment.git. No credentials or generated kubeconfig are committed. To publish changes: git add . && git commit -m 'Deploy Paperless-ngx' && git push origin main.

./setup-authentik.sh creates the OAuth2/OpenID Connect provider, application, and Paperless Users and Paperless Admins groups in Authentik, adds feedc0de to both groups, and stores the client secret in the paperless-oidc Kubernetes Secret. It preserves that secret on later runs. ./install.sh invokes it automatically. To grant others access, add them to Paperless Users in Authentik's web UI; use Paperless Admins only for people who should see and administer all documents. Paperless syncs those Authentik group claims at login, including administrator status. Existing Paperless users can link their Authentik account from My Profile. Local login stays enabled as a recovery option. The callback is https://paper.brunner.ninja/accounts/oidc/authentik/login/callback/ per Authentik's Paperless guide.

The sign-in button says brunner.ninja Auth; the internal OIDC provider ID remains authentik so existing account links keep working.

The first-visit create-user form is Paperless's bootstrap screen. Since the initial local feedc0de account was created, first sign in at /accounts/login/ with its local password, then choose My Profile → Connect new social account and authenticate with Authentik. Starting with the Authentik button before linking opens a new-account registration screen and conflicts with the existing username. After linking, use the Authentik button for future logins; the Paperless Admins claim grants administrator status.

Recovery

The PVC has a Retain storage class. Do not delete it on uninstall. Restore the PVC and the application Secret together. SQLite requires a consistent backup; stop the Paperless deployment or use its document exporter before copying the database.