Drop the repo-recall web container + adjust tailnet exposure for api-only service #159

New issue

Open

opened 2026-05-27 22:18:53 +00:00 by coilysiren · 0 comments

coilysiren commented

2026-05-27 22:18:53 +00:00

Owner

Problem

deploy/repo-recall.yml still ships a web container (Caddy + static SPA from ghcr.io/coilysiren/repo-recall-web:latest) plus a ts sidecar that exposes port 80 routed at the web container. As of repo-recall#90 the React frontend has been removed from coilysiren/repo-recall - the repo-recall-web image is no longer built by the docker workflow, and the web setter has been dropped from the kubectl set image call. The deployment still has a web container slot, currently holding whatever repo-recall-web:latest was last pulled.

Scope

Remove the web container from the Deployment template.
Drop the repo-recall-web GHCR reference and any "web" pull-policy bits.
Adjust the tailnet exposure so peers hit the api container directly. Either:
- Have the api container bind 80 (needs NET_BIND_SERVICE capability), or
- Keep api on 7777 and add a port-translating Service / tailscale serve config in front of it.
Update comments throughout the file to drop "Caddy fronts the static SPA" framing.
Update the docker workflow comment on rollout-status timeout (currently mentions "85 MB api + 24 MB web") and any other multi-container framing in coilysiren/repo-recall/.github/workflows/docker.yml if needed.

Out of scope

Restoring any web frontend. repo-recall is API + MCP only.

Acceptance

sudo k3s kubectl apply -f deploy/repo-recall.yml reconciles cleanly.
Pod has one app container (api) + the ts sidecar.
Tailnet peers can still reach repo-recall at http://repo-recall via MagicDNS.
Next push to coilysiren/repo-recall main rolls the api image successfully.

Context: coilysiren/repo-recall#90.

**Problem** `deploy/repo-recall.yml` still ships a `web` container (Caddy + static SPA from `ghcr.io/coilysiren/repo-recall-web:latest`) plus a `ts` sidecar that exposes port 80 routed at the web container. As of repo-recall#90 the React frontend has been removed from `coilysiren/repo-recall` - the `repo-recall-web` image is no longer built by the docker workflow, and the `web` setter has been dropped from the `kubectl set image` call. The deployment still has a `web` container slot, currently holding whatever `repo-recall-web:latest` was last pulled. **Scope** - Remove the `web` container from the Deployment template. - Drop the `repo-recall-web` GHCR reference and any "web" pull-policy bits. - Adjust the tailnet exposure so peers hit the `api` container directly. Either: - Have the api container bind 80 (needs `NET_BIND_SERVICE` capability), or - Keep `api` on 7777 and add a port-translating Service / `tailscale serve` config in front of it. - Update comments throughout the file to drop "Caddy fronts the static SPA" framing. - Update the docker workflow comment on rollout-status timeout (currently mentions "85 MB api + 24 MB web") and any other multi-container framing in `coilysiren/repo-recall/.github/workflows/docker.yml` if needed. **Out of scope** - Restoring any web frontend. repo-recall is API + MCP only. **Acceptance** - `sudo k3s kubectl apply -f deploy/repo-recall.yml` reconciles cleanly. - Pod has one app container (`api`) + the `ts` sidecar. - Tailnet peers can still reach repo-recall at `http://repo-recall` via MagicDNS. - Next push to `coilysiren/repo-recall` main rolls the api image successfully. Context: coilysiren/repo-recall#90.

coilysiren referenced this issue from coilyco-flight-deck/repo-recall

2026-05-27 22:19:31 +00:00

Remove the React frontend entirely #90

coilysiren added the

label

2026-05-31 01:54:32 +00:00

coilysiren added

and removed

labels

2026-05-31 07:00:36 +00:00