Design a caching strategy for an inventory system where data changes frequently. Which caching approach would you choose and why?

How I frame it

Frequently changing inventory is a poor fit for long-TTL read cache as source of truth. I split browse path (approximate OK) from purchase path (must be exact).

Recommended strategy

• Purchase path — no stale cache: reserve() uses DB row lock or atomic Redis DECR synced to DB. Authoritative at commit time.
• Browse path — short TTL + event invalidation: "In stock" badge cached 5–15s; bust cache on StockChanged Kafka event.
• Approximate display: Show "Only few left" buckets (In stock / Low / Out) not exact count — reduces cache sensitivity.
• Read replicas for reporting only: Analytics on replica; never for checkout.
• Write-through for slow movers: SKUs that change once a day can cache longer with invalidation on shipment.

PLP browse:  cache in_stock=true  TTL 10s  (stale ok)
Add to cart:  no stock hold yet
Checkout:     atomic reserve() → DB/Redis truth
Why:          oversell cost >> stale badge 10s

Walkthrough — flash sale SKU

iPhone launch: 500 units, 50k RPM on product page. Browse cache shows "Low stock" bucket refreshed every 10s via invalidation on each sale batch. Checkout never reads that cache — reserve() decrements authoritative counter. User who saw stale "In stock" at browse gets honest sold-out at pay — acceptable; oversell is not.

What I would avoid

Caching exact quantity=47 for 5 minutes on a flash sale SKU — guaranteed angry users at checkout and oversell incidents.

Weak answer

"Don't cache inventory at all." DB melts on browse traffic; cache with correct boundary (browse vs reserve) is required. "Cache everything with 1 hour TTL" — wrong for fast-moving SKUs.

Follow-up questions

Redis counter vs DB for flash sale?

Redis for speed; async reconcile to DB with unique reservation ids. DB is legal record for shipment.

Multi-warehouse inventory?

Cache per (sku, warehouse) or aggregate "available somewhere" for browse; reserve picks warehouse in transaction.

Link to shopping cart question?

Cart does not hold stock; cache strategy applies at browse. Reserve at checkout uses non-cached path.