Zero Downtime Migration with 2X Performance Growth - Veppy's Transformation Story

Growth is exciting until complexity begins to slow it down. Veppy had built a compelling headless commerce platform with an ambitious roadmap and a rapidly growing UAE customer base. Averaging over 37 million CDN requests per month, the platform was gaining serious traction. But beneath that momentum, operational strain was quietly building.

The existing infrastructure was a single-server setup on ReadyMage's managed platform, all services sharing the same resources. Varnish, PHP, MariaDB, OpenSearch, RabbitMQ, Redis, and the Magento admin panel were co-located on one instance. A memory spike from an indexer could stall user-facing requests. A misconfigured module update could break GraphQL responses without a single error appearing in the PHP logs. Nothing was catastrophically broken, but the platform was misaligned with the scale Veppy was heading toward, and the gap was widening.

For Veppy, selecting the right infrastructure partner for a live eCommerce migration required confidence not just in technical capability, but in platform-specific expertise. Magento combined with ScandPWA, headless routing, GraphQL schema dependencies, and PWA-specific Varnish VCL is a combination that very few teams are equipped to operate correctly.

After a thorough evaluation, our combined depth in Magento architecture, ScandPWA compatibility, and AWS cloud engineering gave Veppy the assurance to proceed.

Rather than a like-for-like migration to a new host, we designed a purpose-built multi-node architecture that separated every service concern and eliminated single points of failure. Six dedicated nodes were deployed within a private VPC across three AWS availability zones in the UAE region: a Bastion for secure access, a Varnish node for full-page caching, a Web node running Nginx and PHP-FPM behind an Auto Scaling Group, a dedicated Admin node isolated from all traffic, a Services node housing OpenSearch, MariaDB, Redis, and RabbitMQ with zero public exposure, and a Prerender node for self-hosted SEO bot rendering.

Shared media and application files were moved to Amazon EFS, mounted across nodes for consistent access without replication overhead. The cutover itself was executed with zero downtime. DNS TTLs were reduced days in advance, the new environment was fully validated in parallel before any traffic was shifted, and Cloudflare was used as a traffic control layer to incrementally route live requests to the new infrastructure while the rollback procedure remained on standby throughout.

The most technically demanding part of the engagement was not the migration itself but the deep compatibility remediation required to ensure Magento and ScandPWA worked correctly on the new stack. ScandPWA's dependence on Magento's GraphQL API means that module conflicts often produce silent failures, breaking product listings, cart behavior, or checkout flows without generating any PHP error logs.

Our team audited every installed Magento module for compatibility with Magento 2.4.6-p6, PHP 8.1, and OpenSearch 2.12.0. Where vendor patches were unavailable, code-level fixes were applied directly. PHP-FPM pool settings, OPcache configuration, Nginx FastCGI parameters, Varnish VCL rules, and RabbitMQ consumer supervision were all tuned specifically for ScandPWA's API-heavy request patterns, not just for standard Magento.

For a headless PWA like Veppy, prerendering is not optional it is the mechanism that makes the storefront indexable by search engines and properly shareable on social platforms. Veppy had been relying on prerender.io's cloud service, but as traffic scaled, per-render costs and limited cache control became constraints that did not fit the growth trajectory.

We deployed a self-hosted prerender service on a dedicated m7i.large node within the private VPC. Nginx on the web tier was updated to route bot traffic to the internal prerender service, and a cache warming strategy was established for high-priority pages. The self-hosted model eliminated third-party latency, removed per-render cost scaling, and gave Veppy complete control over how and when pages are rendered, a capability that directly supports SEO performance at scale.