Performance & reliability · High-traffic retail
Cost Price Supplements
Performance and reliability work on a high-traffic supplements retailer sustaining 20–30 orders a minute at peak, plus warehouse synchronisation and search relevance.
- Peak throughput
- 20–30 orders / minute
The problem
The platform is not built for 20–30 orders a minute. At peak the store was contending on the database, the admin became unusable exactly when the business needed it most, and stock levels drifted from what the warehouse actually held — which turns a good traffic day into a refund queue.
What I built
Profiled the checkout and order-write path to find where lock contention actually was, rather than adding caching in front of the symptom.
Reduced and batched what each order write touched, which is where the real ceiling was.
Introduced Redis for the read paths and session state that were hitting the database hardest at peak.
Built continuous, idempotent warehouse synchronisation so stock levels reconciled against the real source of truth instead of drifting.
Replaced default search with Luigi's Box, which handles the intent-heavy queries supplement shoppers actually type.
Added Cypress coverage on the purchase path so peak-traffic regressions were caught before a sale, not during one.
Architecture
A conventional commerce stack, hardened at each layer that peak traffic exposed.
Edge
Application
Integrations
Data
What was hard
Peak load hit the write path, not the read path
The obvious fix — cache the catalogue — did nothing for the real bottleneck. Order writes were the contended resource, so the work was in reducing and batching what each order touched. Profiling first is not process; it is the difference between fixing it and not.
Stock accuracy under fast selling
At 20–30 orders a minute, even a short sync delay oversells. Reconciliation had to be continuous and idempotent, tolerating retries and partial failures without double-counting.
Business impact
- Sustained 20–30 orders per minute at peak without checkout degradation.
- Stock accuracy held during fast-selling periods, cutting oversell and refunds.
- Search relevance improved on the long-tail queries that drive discovery.
- Purchase-path regressions caught in CI rather than in production.
Technologies
Working on something like this?
I'm open to senior Shopify and full stack roles. Tell me what you're building.