Supabase Region Migration

Moving CE Pro to a new Supabase project is not just a database copy.

The production stack currently depends on Supabase through:

• web runtime envs and service-role helpers
• hosted Auth callbacks and redirect allow-lists
• storage buckets and signed upload/download flows
• Supabase Edge Functions
• mobile Supabase config and mobile function invocation

CE Pro moved production Supabase traffic from Oregon to East US / North Virginia (us-east-1) on April 26, 2026. The active production Supabase project is yoztbemocmsljwgqqtgr, which keeps Supabase close to the Vercel Washington, D.C. / Virginia (iad1) runtime.

Current Production State

• Production Vercel is deployed from the East-backed build rolling-suds-estimator-lp7cuq1as-mattclawdbot-9929s-projects.vercel.app.
• Final cutover validation showed row-count parity across critical public, auth, and storage metadata tables.
• Post-cutover audit validation repaired East constraint drift for holiday follow-up sequences, Meta/Google Ads connector providers, lead assignment foreign keys, and the lead intake timestamp index.
• Storage object validation copied and verified 80 objects totaling 34,782,472 bytes; the audit also downloaded all 80 East objects and matched the expected byte total.
• Supabase Auth hosted settings, redirect allow-list, SMTP sender settings, Edge Functions, and function secrets were recreated on the East project.
• Production health checks now return status: ok with Supabase reachable from Vercel iad1.
• Local web .env.local was realigned to the East Supabase URL and keys; direct DB migration pushes still need the East SUPABASE_DB_PASSWORD available for the Supabase CLI.
• Mobile EAS hosted environments for development, preview, and production now set EXPO_PUBLIC_SUPABASE_URL and EXPO_PUBLIC_SUPABASE_ANON_KEY to the East project; ignored local mobile .env files were realigned too.
• The deployed ai-assistant Supabase Edge Function now falls back to OPENAI_API_KEY when GOOGLE_AI_API_KEY is not configured, matching the active web admin AI route's OpenAI-backed provider.

Cutover Checklist

Before switching traffic:

1. Create the new hosted project in East US / North Virginia (us-east-1) and capture the new project ref, API URL, anon key, service-role key, and DB/pooler strings.
2. Recreate the hosted Auth settings on the new project, including the production site_url, redirect allow-list, session limits, refresh rotation, and OTP expiry settings CE Pro expects.
3. Apply the full schema and deploy the required Supabase Edge Functions to the new project.
4. Recreate the storage buckets CE Pro needs and copy storage objects, not just tables.
5. Update the web app envs, mobile envs, and any local ops/scripts that still point to the old project.
6. Drain or pause background jobs and webhook delivery work before the final delta copy so you do not process the same workload in two places.
7. Run smoke checks, then rerun the stress harness against the us-east-1 stack before calling the move complete.

Repo Audit Highlights

The repo has a few migration-sensitive surfaces worth calling out explicitly.

Web app

The main web runtime is env-driven, so the core cutover values are:

• NEXT_PUBLIC_SUPABASE_URL
• NEXT_PUBLIC_SUPABASE_ANON_KEY
• SUPABASE_SERVICE_ROLE_KEY

Those values drive the shared Supabase bootstrap paths, auth callback route, middleware session handling, direct proposal-view tracking, and script helpers.

Mobile

The mobile app has its own Supabase coupling and should be treated as a first-class migration surface.

• The runtime reads EXPO_PUBLIC_SUPABASE_URL and EXPO_PUBLIC_SUPABASE_ANON_KEY.
• mobile/eas.json now selects the matching EAS environment instead of committing a Supabase project URL or anon key.
• Verify EXPO_PUBLIC_SUPABASE_URL and EXPO_PUBLIC_SUPABASE_ANON_KEY are present in the EAS development, preview, and production environments before rebuilding.
• Update ignored local mobile .env files during future cutovers so device testing does not keep talking to the previous region.
• The mobile app invokes Supabase Functions directly, including send-proposal.

If the mobile EAS environments or local mobile env files stay pointed at the previous project, the migration is only half-done.

Local CLI state

Before running migrations against the new project, confirm local Supabase CLI metadata is internally consistent.

• supabase/.temp/project-ref
• supabase/.temp/pooler-url
• supabase/.temp/linked-project.json

A mismatch between those files means the local CLI has not been cleanly relinked yet.

Auth redirects

CE Pro generates hosted auth links for:

• password reset
• team invite/setup
• customer magic-link login
• super-admin reset flows

That means the new Supabase project must have the correct hosted redirect allow-list for the production app origin and the /auth/callback and /auth/complete flows CE Pro uses.

Storage buckets

At minimum, verify the currently active hosted buckets:

• bug-report-screenshots
• proposal-photos
• proposals

Some of these are called out in repo migrations as manual bucket-creation steps on a fresh project, so do not assume db push alone makes storage ready.

Post-migration Validation

Run this in order:

1. GET /api/health
2. admin login, customer login, forgot-password, and invite flows
3. admin Jobs, Invoices, Services, Lead Sources, and Tags reads
4. job-photo and attachment upload/download
5. proposal send and proposal-view tracking
6. webhook smoke tests for the providers you actively use

Also verify:

• background-job queue health
• webhook-delivery health
• no missing Edge Functions on the new project

Stress Test Rerun

After the smoke checks, rerun the same stress harness used during the scale-hardening work.

Small shakeout:

STRESS_BASE_URL=https://cleanestimate.pro \
STRESS_ADMIN_COOKIE_FILE=.secrets/va-admin-cookies.txt \
STRESS_CRON_SECRET=replace-me \
npm run test:stress -- --stages='[{"name":"shakeout","sessions":10,"durationSec":120}]'

Then the full 1k-ready profile:

STRESS_BASE_URL=https://cleanestimate.pro \
STRESS_ADMIN_COOKIE_FILE=.secrets/va-admin-cookies.txt \
STRESS_CRON_SECRET=replace-me \
npm run test:stress

The March 22, 2026 stabilized Oregon-backed baseline was:

• error rate 1.70%
• overall p95 2465ms
• health p95 351ms

The us-east-1 cutover should at minimum preserve correctness and improve that latency ceiling.

Related Guides

• Platform Reliability
• Operations Runbook
• Stress Testing