The unified image uses nginx on port 80 as the entrypoint, which
reverse-proxies to the Go backend on 127.0.0.1:8080 internally.
K8s was exposing 8080 directly, bypassing nginx entirely.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- k8s/backend.yaml → k8s/switchboard.yaml (already renamed)
- Deployment/Service name: switchboard-be → switchboard
- Removed component:backend labels (single image, no component split)
- Ingress: all paths route to switchboard service on :8080
- CI: updated manifest render, apply, and rollout references
- Removed stale extractor sidecar and extraction env vars
- Updated CHANGELOG and ROADMAP with CI fixes and design decisions
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
k8s/backend.yaml referenced ${BE_IMAGE} but the workflow env
defines ${IMAGE}, causing envsubst to produce an empty image
name (InvalidImageName). Also consolidated the summary table
to single image line since FE/BE are merged.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The k8s/backend.yaml template used BE_-prefixed env vars
(BE_MEMORY_REQUEST, etc.) but the CI workflow outputs them
without the prefix, causing envsubst to produce empty strings
and K8s to reject the deployment.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CI pipeline: replace separate FE + BE image builds with single unified
Dockerfile. Remove FE_IMAGE, FE_REPLICAS, FE_* resource vars. Simplify
build/push/deploy to single IMAGE. Remove frontend.yaml k8s manifest.
Remove provider integration test env vars (PROVIDER, PROVIDER_KEY,
VENICE_API_KEY). Remove EXTRACTION_CONCURRENCY.
Delete server/Dockerfile (backend-only image, superseded by root
Dockerfile). Delete k8s/frontend.yaml (no separate FE deployment).
Update ROADMAP: steps 5-7 and 9 marked complete. Step 8 (ICD) partial.
Step 10 (smoke test + tag) is all that remains for v0.1.0.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
