Replaces the cv and docs minikube Deployments with ansible roles that
download release tarballs into ~/cv/content and ~/docs/content on indri.
Caddy now serves those directories directly via a new kind=static
service-block in the Caddy template; no daemon, no nginx pod, no
ProxyGroup ingress on the request path.
This commit adds the deploy-side artifacts only. Live cutover (delete
argocd apps, run ansible, verify) is staged manually after PR review;
the dead containers/{cv,quartz} and argocd manifests are removed in a
follow-up commit so each commit is internally consistent.
Workflows are simplified: the deploy step now bumps the role's pinned
version and pushes; running ansible + purging the Fly cache is manual
from gilbert (matches the devpi pattern).
service-versions.yaml: cv and docs are type=ansible. docs current-version
remains 1.28.2 for now to keep container-version-check passing while
containers/quartz still exists; will move to the docs release tag in the
cleanup commit.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3.1 KiB
| title | modified | last-reviewed | tags | ||
|---|---|---|---|---|---|
| CV on Indri | 2026-04-29 | 2026-04-29 |
|
CV on Indri
How the CV/resume static site (cv.eblu.me) is deployed on indri natively. Replaces the prior minikube Deployment; mirrors the rationale of devpi-on-indri.
Why native, not Kubernetes
CV is a tiny static site (HTML + CSS + PDF). It needs no daemon, no database, no auth. Caddy on indri can serve the extracted tarball directly via file_server. Removing the minikube Deployment shrinks the cluster's footprint and removes a network hop (Fly → indri Caddy → ProxyGroup ingress → minikube pod becomes Fly → indri Caddy → local files).
Layout
| Concern | Path / detail |
|---|---|
| Content dir | /Users/erichblume/cv/content/ |
| Version sentinel | /Users/erichblume/cv/.installed-version |
| Caddy entry | cv service in ansible/roles/caddy/defaults/main.yml (kind: static) |
| Public URL | https://cv.eblu.me (via flyio-proxy) |
| Private URL | https://cv.ops.eblu.me (Caddy on indri) |
| Tarball source | Forgejo generic package cv (forge.eblu.me/eblume/-/packages) |
The role is driven by cv_version in ansible/roles/cv/defaults/main.yml. The download and extract steps only fire when the on-disk sentinel doesn't match cv_version — i.e. after a version bump.
Deploy
Two paths:
From a release workflow (most common):
- Run the
Release CVworkflow in the cv repo → produces a new generic package - Run the blumeops
Deploy CVworkflow → bumpscv_versioninansible/roles/cv/defaults/main.ymland pushes to main - From gilbert:
mise run provision-indri -- --tags cv - From gilbert:
fly ssh console -a blumeops-proxy -C "sh -c 'rm -rf /tmp/cache && nginx -s reload'"to purge the public-edge cache
Manual (e.g., reverting): edit cv_version in the role defaults yourself, then steps 3–4.
Verify
ssh indri 'cat ~/cv/.installed-version'
ssh indri 'ls -la ~/cv/content/'
curl -fsSI https://cv.ops.eblu.me/ # private
curl -fsSI https://cv.eblu.me/ # public
curl -fsSI https://cv.eblu.me/resume.pdf | grep -i disposition
The PDF response should include content-disposition: attachment; filename="erich-blume-resume.pdf".
Bumping the cv version
Edit cv_version in ansible/roles/cv/defaults/main.yml and re-run mise run provision-indri -- --tags cv. The role recreates the content dir from the new tarball; the sentinel update triggers the next idempotent skip.
Backup
The content dir is not in borgmatic_source_directories. The tarball is re-downloadable from the Forgejo generic package store on every deploy, and the source is in the cv repo — recovery is just re-running the role.
Rollback
If a bad version is published, set cv_version back to the previous tag in ansible/roles/cv/defaults/main.yml and re-run the role. The full minikube manifest set is preserved in git history (commits prior to the migration cleanup) for the worst case.
Related
- devpi-on-indri — same shape, different upstream
- restart-indri — graceful indri restart procedure
- cv — service reference