Merged
Conversation
The old deploy.sh assumed \`railway up --config template.yaml\` (doesn't exist) and other commands that never shipped. Replaced with a phased script that uses only verified CLI 4.40 commands: - \`railway init --name\` for project create - \`railway add --database\` / \`--image\` / \`--repo\` / \`--service\` - \`railway variable set K=V --service N --skip-deploys\` - \`railway domain --service N\` for public URL - \`railway up\` from services/mcp-router to ship the router (CLI has no rootDirectory flag for \`add --repo\`, so we create empty + push local) Phases — each independently re-runnable: init project + Postgres/FalkorDB/Qdrant services nakatomi, automem, mcp-router (env vars, public domain) seed admin user + Rev A pipeline/custom-field overlay finalize print public URL + signup token (what admins hand to PMs) Generated secrets (Nakatomi SECRET_KEY, AutoMem API/admin tokens, signup token) persist in railway/.deploy-state across phases. Gitignored. README rewritten to match: removed the aspirational \"one-click\" template button claim, added the CLI phase flow as the recommended path and a web-UI fallback for when things go sideways. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Two fixes after a live run hit real edge cases: 1. \`railway add\` happily creates duplicates on re-run (made two Postgres services when we re-ran phase 1). Added \`have_service <name>\` helper that queries \`railway status --json\` and skips the add if the name already exists. Every service add in phases 1 and 2 now gated by it. 2. \`railway add --repo\` returns "Unauthorized" silently if Railway's GitHub App isn't authorized on the target org — which surprised me mid-deploy on mrdulasolutions/NakatomiCRM and automem. Added a pre-flight warning block at the top of phase 2 pointing at the exact fix (Railway dashboard → Integrations → GitHub → grant org access), and called it out as a prerequisite in railway/README.md. These make the script safe to re-run after any failure without leaving orphaned or duplicated services. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The Railway GitHub App is installed inline when you add the first repo-sourced service in the web UI — there's no standalone Integrations page to click. My previous README step sent users to a page that doesn't exist. Corrected both README and deploy.sh pre-flight warning to walk through the actual UI path: + Create → GitHub Repo → Configure GitHub App → grant org access → name the service. Called out that the script is safe to re-run after the UI step because have_service() will skip the existing service. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Shaken out by actually running the phased deploy against a fresh Railway
project. Every fix here comes from a concrete failure mode hit during the
first reva-ops bring-up.
mcp-router
- router/main.py: wire FastMCP session_manager.run() into FastAPI's
lifespan. Without it, mounting mcp.streamable_http_app() leaves the
anyio task group uninitialized and every /mcp/ request 500s with
"Task group is not initialized. Make sure to use run()."
- requirements.txt: add email-validator — signup.py uses pydantic
EmailStr which fails ImportError at import time without it.
- railway.toml: drop startCommand. Railway runs it exec-style without
shell, so "--port $PORT" stays literal and uvicorn rejects it. The
Dockerfile CMD already wraps in `sh -c` which expands correctly.
Nakatomi seed overlay (services/nakatomi-backend/seed/reva.py)
- Rewrite against Nakatomi's real pipeline API: stages are nested in
the POST /pipelines body (there is no /pipelines/{id}/stages route);
pipeline + stage both require a `slug` matching [a-z0-9][a-z0-9_-]*.
- Custom fields live at POST /custom-fields, not /schema/custom-fields.
Schema is {entity_type, name, label, field_type, description} with
field_type restricted to scalars (string|text|number|bool|date|url|
email|select) — no object/array. JSON-shaped fields now ride as text.
- Idempotent by (entity_type, name) pre-check on GET.
railway/deploy.sh
- phase_seed: switch from `railway run` (which injects env vars locally
rather than running in the container and can't reach the in-container
DB) to `railway ssh` and execute `python -m scripts.seed` against
/app directly. Ship reva.py into the container via base64 pipe since
the upstream Nakatomi image doesn't vendor the overlay; run it
against http://localhost:8000.
- discover_nakatomi_service(): tolerate UI rename drift
(Nakatomi-backend vs nakatomi-backend) by matching case-insensitively
from railway status --json.
- Cross-service wiring: stop hardcoding `<service>.railway.internal`
hostnames — Railway derives those from the *original* service name
at creation, so a later rename leaves the DNS stale. Use
`${{service.RAILWAY_PRIVATE_DOMAIN}}` refs which Railway resolves to
the real current domain at deploy time.
- Migrate variable writes from the deprecated `railway variable set`
flag-form to the batched `railway variables --set k=v --set k=v`.
- Fix `railway domain --json` parsing: shape is {domains: [url, ...]},
not [{domain}]. Strip the https:// prefix to get the bare host.
- Extend save_state to persist ADMIN_EMAIL/PASSWORD/WORKSPACE_* so a
seed retry doesn't lose the generated password.
railway/README.md
- Update rotation hint to the new `railway variables --set ...` syntax.
Verified against live deploy: /health 200, /signup 200 (HTML page
renders), POST /mcp/ with Accept: text/event-stream returns a valid
initialize response advertising REVA Router + crm_/mem_/reva_ tool
namespaces.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Owner
Author
|
approved |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.