Million-SKU catalogs need tiered search indexes, CDN edge cache, and async warm-up — not full-table LIKE queries on every keystroke.
Engineering notes
- Separate hot MPN exact match from fuzzy alternates
- List endpoints return lean payloads; details load on demand
- Zero-result analytics feed procurement and content teams
Performance SLAs belong in the project scope. Inquire with your SKU count.
Deep dive
Million-SKU pain usually comes from monolithic LIKE scans, not CPU size alone. Split exact MPN, prefix, and fuzzy paths; edge-cache hot fragments. Typing through to source DB every keystroke will fail INP on EU 4G no matter how big the VM.
Contract SLAs (e.g., search P95 <800ms) and warm indexes from real query logs — not just homepage load tests. Stagger stock sync away from full reindex windows; monitor beside Core Web Vitals.
Related reading
Plans & conversion
Declare SKU scale at kickoff — Online Trade and Source Code support distributed search; Mall RFQ fits sub-10M common cases. Inquire with counts and buyer regions.
Extended FAQ
Q: Will bigger servers fix it? Vertical scale helps marginally; indexing tiers and cache policy fix root cause.
Q: Does caching hurt SEO? Not when crawlers get stable HTML URLs and proper TTL on dynamic stock.
