Production setup

Everything in test works identically in live - the difference is operational, not functional. This page is the checklist we recommend completing before routing production traffic.

Credential checklist

• Live secret key (mrnd_live_…) stored in your production secret manager - not in .env files, CI logs, or Slack.
• Each service uses a restricted key scoped to what it calls (e.g. projects:write, not *).
• IP allowlists set on keys used by fixed infrastructure.
• A rotation runbook exists: create → deploy → verify → revoke. (Practice it in test.)
• No key appears in client-side code, mobile bundles, or public repositories - search your codebase for mrnd_ before launch.

Webhook checklist

• Endpoints registered in the live environment (easy to miss - check the environment badge).
• Signature verification enabled with the live endpoint's whsec_ secret - verification code is identical, the secret differs.
• Handlers respond within 10 seconds; heavy work is queued async.
• Handlers are idempotent - dedupe on event.id.
• webhook.endpoint.disabled alerts route to your on-call channel.
• HTTPS endpoints only - live rejects plain http:// URLs.

Rate limits and resilience

• SDK retry defaults reviewed (maxRetries, timeout) for your latency budget.
• Idempotency keys on POST calls that must not double-execute.
• Queue/buffer for burst writes - live limit is 5,000 req/min per key.
• Alert on sustained 429s - they mean you're approaching the ceiling.

Observability

• request_id logged on every API call and error - it's the support trace key.
• Delivery failures monitored via GET /webhooks/:id/deliveries or dashboard alerts.
• Event history retention understood (90 days in live) - export what you must keep longer.

curl "https://api.meridian.dev/v2/webhooks/wh_2nQ8kLm4xP/deliveries?limit=25" \
  -H "Authorization: Bearer $MERIDIAN_LIVE_KEY"

Rollout strategy

Recommended sequence:

1. Shadow in live - create the project and webhook endpoints in live but keep traffic off. Verify event flow with POST test mutations.
2. Canary - point one service instance at the live key; watch errors and delivery latency for a few hours.
3. Ramp - roll the key to the rest of the fleet; keep the rollback path (old config) one deploy away.
4. Verify - confirm dashboard metrics match expectations and audit log entries look clean.

Go-live checklist