Case Study: Scaling a Next.js Site to 10K+ Daily Users with Minimal Ops
In twelve weeks, we migrated a content-heavy B2B catalog from a legacy CMS to Next.js and grew to 10,000+ daily users without adding an SRE team. The constraints were strict: two full-stack engineers, a modest budget, and zero appetite for nightly pager duty. This deep dive unpacks the exact architecture, settings, and operational playbook that kept costs low while performance stayed high.
Architecture that prefers static, tolerates dynamic
We adopted the Next.js App Router with a bias toward pre-rendering. Roughly 92% of routes shipped as static HTML, and the rest used ISR or Edge Functions for personalization. A slim runtime meant fewer moving parts and near-zero cold starts.
- Rendering strategy: generate product and category pages at build; use ISR (revalidate: 600s) for pricing blocks.
- Edge middleware: header-based country routing and feature flags, no server session storage.
- Media: next/image with AVIF, 768/1280/1920 breakpoints; remote patterns locked to one CDN domain.
- Fonts: self-hosted variable fonts; preloaded CSS; CLS dropped below 0.02.
Data and caching: deterministic, layered, boring
We centralized writes in PostgreSQL (managed) via Prisma and cached hot reads in Redis with a simple key schema: entity:version:id. Next.js Route Handlers fetched from Redis first, then DB, then background revalidation published to a lightweight queue. Cache invalidation piggybacked on webhooks from the headless CMS.
- Redis TTL: 15 minutes for product detail, 2 minutes for inventory, 24 hours for category trees.
- Stale-while-revalidate: expose max-age=60, stale-while-revalidate=600 on CDN; users never waited on origin.
- Build-time data: a 3-minute step emitted a compact JSON index (200 KB gzip) for client search.
- Zero server sessions: JWTs signed short-lived (15m) with refresh via httpOnly cookie; decreases memory pressure.
Delivery and ops: push-button, budget-friendly
We deployed to Vercel with regional datacenters close to buyers, and placed Cloudflare in front for WAF, bot rules, and global caching. CI ran on GitHub Actions with a single concurrency group to prevent stampede builds. Every feature shipped behind a flag with instant rollback via config, not redeploys.
- Compression: Brotli level 5 static, gzip fallback; target TTFB under 150ms from edge cache.
- Next.js middleware trimmed cookies, shrinking request headers under 1KB to unlock CDN caching.
- Logging: structured JSON shipped to OpenSearch; 7-day hot, 30-day warm retention.
- Cost guardrail: nightly report combining CDN egress, function invocations, and DB compute; redlines triggered PR checks.
Observability and the Mobile analytics and crash monitoring setup
Minimal ops only works with maximal visibility. We instrumented server, edge, and client with Datadog RUM and Sentry. For parity with the React Native app, our Mobile analytics and crash monitoring setup mirrored web events, letting growth and product compare funnels across surfaces. Release health pinned regressions to a commit within minutes.
- Key dashboards: LCP, CLS, TTFB percentiles; route-level error rate; cache hit ratio across CDN and Redis.
- Alerts: p95 API latency over 400ms for 5 minutes; error rate over 1%; cold path invocation spikes.
- Tracing: W3C tracecontext headers flowed from edge to DB; one-click flamegraphs for slow queries.
SEO gains that marketers can bank
Static-first rendering compounded SEO. We emitted dynamic sitemaps per locale, canonical tags, and JSON-LD for products. A hybrid ISR approach kept "fresh" pages within 10 minutes while preserving cacheability. Organic sessions rose 38% in six weeks; paid CAC fell as quality scores climbed with faster LCP.
Team, sourcing, and the right partners
Speed came from focus and the right people. We pulled a senior Next.js engineer and a platform-savvy generalist via a talent marketplace for developers, and we paired them with an Enterprise digital transformation partner for governance and security reviews. When we needed extra hands, slashdev.io supplied vetted remote engineers who slotted into our rituals within days and acted like an extension of the core team.
- Onboarding: day 1 access via SSO and least-privilege roles; reproducible dev with pnpm + Docker Compose.
- Security: SOC 2-friendly logs, secret rotation with cloud KMS, and dependency review gates.
- Cadence: two-week sprints; performance budget reviewed in sprint review alongside feature demos.
What we would repeat-and what we'd skip
- Repeat: heavy ISR, lean server code, Redis-first reads, and CDN-stable headers.
- Repeat: typed analytics events with contract tests to keep growth dashboards trustworthy.
- Skip: monolithic GraphQL gateway; REST with resource caching proved simpler under load.
- Skip: per-user server personalization; edge experiments via cookies performed better and cached.
Action checklist to hit 10K daily users with minimal ops
- Classify routes: static, ISR (revalidate windows), or truly dynamic; keep dynamic under 10%.
- Enforce cacheability at headers: small cookies, long max-age, and clear SWR.
