Slow API calls cost real money. They delay orders, stall inventory syncs, and frustrate users right when traffic peaks.
Here's the thing most teams miss: when a Shopify integration crawls under load, the bottleneck is almost never Shopify. It's the client layer you wrote around it.
This is how I think about building a client that stays fast whether you push 100 orders a day or 100,000.
The client is your performance ceiling
Shopify gives you the API. How you call it sets your speed.
Most clients start as a thin wrapper around fetch. Fine at small scale. Then volume grows and the cracks show up: 429s, timeouts, duplicate calls.
A solid client juggles three things at once:
- Request rate (don't get throttled)
- Data reuse (don't fetch what you already have)
- Failure handling (don't crash the whole flow on one bad response)
Nail those three and optimization stops being a fire drill. It becomes a property of the system.
REST vs GraphQL
Your API choice shapes everything downstream.
REST is predictable but forces multiple round trips and over-fetches. GraphQL lets you grab exactly the fields you want in one call, at the cost of a query-cost budget you have to respect.
| Factor | REST | GraphQL |
|---|---|---|
| Data shape | Fixed per endpoint | Custom per query |
| Over-fetching | Common | Rare |
| Round trips | Often multiple | Usually one |
| Rate limiting | Request-based (leaky bucket) | Cost-based (points) |
| Best for | Simple, stable reads | Complex, nested data |
For most modern builds, GraphQL wins. Just budget your query cost carefully, a sloppy nested query drains your points fast.
Respect rate limits by design
Rate limits aren't obstacles. They're the rules of the road.
REST uses a leaky bucket. GraphQL uses a calculated cost model. Either way, hit the ceiling and you eat a 429.
A good client never blindly retries. It reads the rate-limit headers on every response and paces itself before the wall, not after. The cleanest version is a token bucket on your side that mirrors Shopify's:
class TokenBucket {
constructor(capacity, refillPerSec) {
this.capacity = capacity;
this.tokens = capacity;
this.refillPerSec = refillPerSec;
this.last = Date.now();
}
async take(cost = 1) {
this.refill();
while (this.tokens < cost) {
await new Promise(r => setTimeout(r, 100));
this.refill();
}
this.tokens -= cost;
}
refill() {
const now = Date.now();
this.tokens = Math.min(
this.capacity,
this.tokens + ((now - this.last) / 1000) * this.refillPerSec
);
this.last = now;
}
}
Batch to cut round trips
Every network call carries overhead. Fewer calls = faster results.
GraphQL batches naturally with aliases:
query {
a: product(id: "gid://shopify/Product/1") { title }
b: product(id: "gid://shopify/Product/2") { title }
c: product(id: "gid://shopify/Product/3") { title }
}
Ten products in one call instead of ten. Latency drops, rate budget stretches.
For big datasets, reach for the Bulk Operations API. It runs server-side and hands you a downloadable file when ready.
| Method | Best for | Speed profile |
|---|---|---|
| Single request | Real-time single record | Lowest latency |
| Batched query | A few related records | Fast, fewer calls |
| Bulk operation | Thousands of records | Highest throughput |
Cache hard, invalidate smart
The fastest API call is the one you never make.
Product data, collections, and shop settings rarely change. Cache them. Set a TTL per data type, long for static, short for volatile stuff like inventory.
The hard part is invalidation. Stale data means wrong prices and overselling. Use webhooks to bust cache entries the moment upstream data changes.
A layered cache wins: in-memory for hot keys, Redis for distributed access.
Concurrency without self-throttling
Parallel requests speed you up. Too many at once get you throttled.
Use a concurrency limiter that caps parallel requests and queues the rest. Set the cap based on your rate budget, not your CPU count, your machine can fire hundreds, but Shopify will reject most.
Pair the concurrency pool with the token bucket: the pool controls parallel slots, the bucket controls overall pace.
Retry logic that heals itself
Failures happen. A naive client gives up or retries instantly (making it worse). A smart one backs off.
async function withRetry(fn, max = 5) {
for (let attempt = 0; attempt < max; attempt++) {
try {
return await fn();
} catch (err) {
if (![429, 500, 502, 503, 504].includes(err.status)) throw err;
const base = 2 ** attempt * 200;
const jitter = Math.random() * 200;
await new Promise(r => setTimeout(r, base + jitter));
}
}
throw new Error("Max retries exceeded");
}
Retry only transient errors (429, 5xx, timeouts). Never retry a 400 or 422, the request itself is broken. Always cap retries, and pair them with idempotency keys so repeats never create duplicate orders.
Paginate the right way
Large result sets arrive in pages via cursors. Never loop through a huge dataset synchronously, push it into a background job.
Fetch a page, process it, request the next with the returned cursor, stop when hasNextPage is false.
Go async
Not every task needs an instant answer. When an order lands, accept it fast and queue the heavy lifting, inventory sync, CRM updates, notifications.
This keeps the client responsive during spikes and smooths your API usage instead of bursting.
Architect for resilience
A fast client on a fragile system still fails. Spread load across services. Add circuit breakers so one failing dependency doesn't drag everything down. Monitor latency, error rates, and throttle counts, you can't fix what you can't see.
Design for failure as the normal case. Degrade gracefully, don't collapse.
Quick audit checklist
| Area | Action | Impact |
|---|---|---|
| API choice | Prefer GraphQL for nested data | Fewer round trips |
| Rate limits | Track headers, self-throttle | No 429 storms |
| Batching | Group reads, use bulk ops | Higher throughput |
| Caching | Layer memory + Redis | Skip redundant calls |
| Concurrency | Cap parallel requests | Stable under load |
| Retries | Backoff with jitter | Self-healing |
| Async | Queue heavy work | Responsive client |
| Monitoring | Track latency + errors | Fast diagnosis |
Wrapping up
A high-performance client isn't one trick. It's a stack of good decisions: right API, respected limits, batched reads, smart caching, capped concurrency, self-healing retries, async processing, all wrapped in monitoring.
I wrote a fuller version of this on our blog with deeper links into each topic: Building High-Performance Shopify API Clients.
What does your retry/backoff setup look like? Curious how others handle the cost-budget side of GraphQL.











