Zero-downtime WooCommerce migration at 400 orders per day

A multi-brand apparel retailer, approximately 12,000 SKUs, 400+ orders on peak days, INR 2–3 crore monthly GMV, needed off shared hosting before festival season. Checkout timeouts, MySQL lock waits, and manual inventory CSV exports were already costing operations hours daily. Downtime during migration was unacceptable: prior internal attempt on a staging clone had taken the cart offline for 40 minutes.

Target architecture: WordPress + WooCommerce on EC2 Auto Scaling (min 2), RDS MySQL 8 with read replica, ElastiCache Redis for sessions and object cache, ALB with stickiness for admin, CloudFront for static assets, and S3 for media offload.

Problems we anticipated and hit: payment gateway (Razorpay) webhook URLs were IP-allowlisted to the old host, had to pre-register ALB hostname and run dual-delivery during cutover; WooCommerce session cookies used a domain attribute that broke when we tested on a grey hostname; serialized PHP in wp_options broke when search-replace ran without serialization-aware tooling; a custom plugin stored absolute filesystem paths from the old cPanel layout.

Migration strategy: weeks 1–4, rebuild on AWS with Infrastructure-as-Code (Terraform), nightly DB sync via mysqldump + binlog catch-up on replica; weeks 5–6, parallel order flow on staging with synthetic load (k6) at 2× peak checkout concurrency; week 7, blue-green: green stack receives read traffic via weighted Route 53, blue remains write-primary until replication lag < 1s for 48 hours.

Cutover window (Tuesday 02:00–06:00 IST): lower DNS TTL to 60s five days prior; enable maintenance mode only for wp-admin; freeze catalog edits; final binlog apply; flip write traffic to green; re-point webhooks; smoke-test 15 payment methods and refund path; run WP-CLI order count reconciliation against Razorpay settlement API.

Zero-downtime mechanics: cart and checkout stayed on blue ALB target until green passed health checks; Redis session replication meant in-flight carts survived the flip; media served from CloudFront already, no origin path change for buyers.

Post-migration: HPOS (High-Performance Order Storage) enabled after two weeks of stable ops; autoscaling policy on CPU and checkout latency p95; AMC runbook for plugin updates with visual regression on checkout.

Outcome: zero lost orders in cutover window; checkout p95 latency 3.8s → 1.1s on mobile; Black Friday week handled 620 orders/day without manual scaling. Client described as a D2C apparel collective, name withheld under NDA.