rstillwell/Submatter
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions
Submatter/docs/Current Progress.md
varutasu 32f76abc9f
Some checks failed
Deploy homelab (CT 116) / Deploy to CT 116 (push) Failing after 7s
Details
CI / Lint, type-check, test (push) Failing after 20m19s
Details
fix: give Workspace Chat its own public hostname
2026-10-30 19:38:03 +00:00

6.4 KiB
Raw Permalink Blame History

Workspace Chat — Current Progress

Updated: 2026-10-30 (dedicated Chat hostname migration pending rebuild).

Hostname correction — pending deployment

chat.stillwell.cloud belongs to OpenMausBot on CT 117, not Workspace Chat. The operator reports routing to that app has been restored and echodo-chat.stillwell.cloud is already an alias on the Workspace Chat Traefik rule.

The Workspace Chat endpoint is now wss://echodo-chat.stillwell.cloud (socket path /chat). Set NEXT_PUBLIC_CHAT_SERVER_URL to that value in CT 116's /opt/submatter/.env, rebuild the web image, and recreate web as documented in docs/deployment-production.md. Repo examples now use only the dedicated hostname. Deployment and browser verification of the new endpoint are not yet confirmed; a push alone will not update the host's .env.

Previous go-live evidence (before hostname correction)

The following records the previous deployment, not verification of the new hostname. Operator reported 14/14 acceptance checks passed. Operator-confirmed CT 102 echodo ledger: 24 entries through 0023, with required columns/table verified and all chat_* tables owned by echodo. An earlier copy of this document said 0022–0024; no 0024 migration exists in this checkout, so that statement is not treated as verified.

The deployment evidence below was supplied by the operator and is present in this workspace copy. It was not independently rerun in this reconciliation session; read-only access through PVE stopped at unknown SSH host-key verification.

Deployed

  • CHAT_TICKET_SECRET generated on CT 116 (≥32), shared by web + collab (sha256 match). Not logged.
  • CHAT_PORT=1235; collab dual listeners 1234 (Hocuspocus) + 1235 (Chat).
  • Traefik echodo-chat → 192.168.68.116:1235 for chat.echodo.stillwell.cloud and chat.stillwell.cloud.
  • Public WSS URL baked into web: wss://chat.stillwell.cloud (nested *.echodo.stillwell.cloud Universal SSL fails publicly).
  • Compose validated; web/collab/worker rebuilt/recreated; opencode-v2 preserved.

Runtime fixes applied during go-live

  1. Ticket Origin check used request.url origin (https://0.0.0.0:3000) behind Docker — now prefers x-forwarded-host / host + x-forwarded-proto.
  2. channel-drizzle-adapter row() spread full DB columns into strict lifecycle receipt schema — project ChannelRecord fields only.
  3. redactCreatedOutbox jsonb_build_object(...) params caused could not determine data type of parameter $1 — set plain JSON payload instead.
  4. Empty AUTH_DEV_PASSWORD blocked credentials acceptance — set on CT 116 (not logged).

Source under /opt/submatter (+ opencode workspace copy) updated for (1–3). Image chunks patched in place for immediate runtime; rebuild web on next deploy to bake permanently.

Source reconciliation (2026-10-29)

  • Channel projection and plain-JSON redaction fixes were already present locally; preserved and covered with adapter regression tests.
  • Local ticket handling now pins the expected origin to configured AUTH_URL (or NEXTAUTH_URL), not caller-supplied forwarded headers. Production fails closed without a valid configured URL. This is a hardened replacement for the deployed forwarded-header hot patch, not a byte-for-byte copy. Deploy and retest ticket issuance before considering this replacement live.
  • Canonical browser endpoint is wss://chat.stillwell.cloud (socket path /chat). The Traefik example retains both hostname routes.
  • Next deploy must rebuild web from reconciled source; image hot patches are not durable. Re-run ticket issuance, cross-origin rejection and the acceptance suite afterward. No production changes, rebuild, commit or push were performed during this reconciliation.
  • Security follow-up: AUTH_DEV_PASSWORD is a shared credentials password for existing user emails, not a per-user password. Remove it after acceptance unless the operator explicitly intends this login mechanism; verify SSO access first and recreate web when removing the environment value. Awaiting operator decision; not changed remotely.
  • Remaining release gates: full local checks, disposable-Postgres coverage, final review and convoy lifecycle reconciliation. The reported live suite does not prove multi-instance fan-out or Redis outage behavior.

Main deployment preparation

The operator approved sending the reconciled branch to main for real-build testing. The branch includes the earlier foundation commit and its AI-shell/backlog changes. The main push triggers .forgejo/workflows/deploy-homelab.yml when the runner and deployment secrets are configured; a push alone does not prove deployment success. The deploy script no longer removes orphan containers, preserving separately managed services such as opencode-v2. Six static deployment checks cover the configuration and this safety guard.

Local validation after reconciliation

  • pnpm lint: passed with existing warnings.
  • pnpm type-check --concurrency=1: passed.
  • pnpm test: ticket-origin regressions (11), channel adapter tests (10) and outbox adapter test (1) passed. Full suite remains blocked by the worker remote-AG-UI pending-response polling test timing out after 5000 ms. PostgreSQL-gated tests remain skipped without a disposable test database.
  • pnpm build --concurrency=1: passed with a longer command allowance after the first attempt timed out during build tracing. This is a local build, not a Docker image rebuild or deployment.
  • git diff --check: passed.
  • Security review of this slice: ticket origin is pinned to server configuration; forwarded-header spoofing, missing/malformed origins and workspace denial are covered. Redaction remains workspace/channel/message/create-event scoped. Production shared-password sign-in remains an operational security concern until explicitly resolved. No secrets were read or printed.

Acceptance evidence (2026-10-29)

Two signed-in members (me@randallstillwell.com + chat-acceptance-b@stillwell.local) against workspace randall-stillwell:

Gate Result
Secret shared web/collab Pass
Compose validate + ports Pass
Public WSS reject without ticket Pass (connection fails)
Dual WS subscribe via wss://chat.stillwell.cloud/chat Pass
Send + Redis invalidate hint Pass
Reply Pass
Reconnect catch-up Pass
Unread mark + summary Pass
Edit / delete + mutation catch-up Pass (after redact fix)
Membership revocation (hint + history denied) Pass