Home › Blog › Digital Infrastructure
Digital Infrastructure

Website and Checkout Load Testing Before 11.11: A Two-Week Plan for Singapore SMEs

Website and Checkout Load Testing Before 11.11: A Two-Week Plan for Singapore SMEs

Checkout load testing before 11.11 means driving your own website and payment flow at the traffic you actually expect on the day, then fixing what breaks while there is still time to fix it. For most Singapore SMEs running a Shopify, WooCommerce or custom store alongside the marketplaces, the honest result is the same every year: the product pages hold up and the checkout does not. With two weeks left, do one thing properly - script the full checkout journey, run it at three to four times your busiest hour of the past year, and fix in order of what fails first.

Why does the checkout fail before the rest of the site?

A product page can be cached. Once the first visitor loads it, every visitor after that can be served the same stored copy, which is why your homepage can take a surprising amount of traffic on cheap hosting.

The checkout cannot work that way. Every single step is unique to one customer and has to be calculated fresh: check the stock, validate the promo code, fetch a live shipping rate, call the payment gateway, redirect to the bank for two-factor authorisation, write the order, send the confirmation email, fire the webhook to your accounting or inventory system. That is seven or eight dependencies, several of them owned by somebody else, all triggered at once by the same crowd. The queue forms at the narrowest of them, and the customer sees a spinner, a timeout, or worse - a charged card with no order.

How much traffic should you test for?

Forget pageviews. The number that matters is concurrent checkouts - how many people are inside the payment flow in the same minute.

Pull your busiest hour from the last twelve months. For most SMEs that is last year's 11.11, a flash sale, or a Deepavali promotion. Take the orders from that hour, assume this year's peak is three to four times heavier because your audience has grown and the sale is more aggressive, and assume half of those orders arrive in a twenty-minute burst when the discount goes live or the email lands.

A worked example: 300 orders on your best day last year, 120 of them in the peak hour, means you should be testing for roughly 400 to 500 orders in an hour, with 60 to 80 customers sitting in the checkout at the same time. That is a modest number. It is also, routinely, four times what an untuned SME setup can handle.

What should the test actually do?

Hitting the homepage with a thousand requests tells you nothing useful. The test has to be a scripted shopper: land on a product, add to cart, apply a discount code, enter an address, pull a shipping rate, and submit a payment using your gateway's test mode.

Run it against a staging copy with production-sized data - a catalogue of fifty test products will pass tests that eight thousand real SKUs fail. Tools like k6 or Locust are free and good enough; your developer or agency can have a usable script running in a day.

Measure four things: how long the cart and checkout pages take to respond, the percentage of attempts that error, how long the payment step takes end to end, and whether your stock counts are still correct afterwards. That last one catches a problem no speed graph will show you - two customers buying the last unit at the same instant because the stock check and the order write are not locked together.

What usually breaks, and what can still be fixed this week?

The common culprits, roughly in the order they appear: too few PHP worker processes, so requests queue before they even reach your code; a database connection limit lower than the number of workers; a shipping or tax API call with no timeout, so one slow partner freezes your whole checkout; confirmation emails and accounting webhooks sent while the customer waits; a promo code table that locks under heavy use; and a shared hosting plan with a hard ceiling you cannot buy your way past in November.

The fixes that genuinely fit a two-week window, in order of value for effort:

  1. Put a strict timeout and a fallback on every third-party call in the checkout. If the courier rate service is slow, show a flat rate rather than nothing.
  2. Move emails, webhooks and syncs out of the customer's wait and into a background queue.
  3. Raise worker and database connection limits together, then re-test. Raising one without the other just moves the queue.
  4. Cache everything that is not the cart or checkout, aggressively.
  5. Prepare a plain static holding page and a simple queue page, so a bad moment looks deliberate instead of broken.

What does not fit the window: re-platforming, rewriting the cart, or changing payment providers. Park those for January.

How do you test without breaking the live site or upsetting your providers?

Three rules. Test against staging, not production, and use the payment gateway's sandbox keys so no real card is touched. Tell your hosting provider and your gateway what you are doing and when - unannounced load generation looks exactly like an attack, and getting your own IP blocked on 8 November is an avoidable own goal. And disable outbound email and SMS in the test environment first, or your customers will receive a few hundred confirmations for orders that do not exist.

One more: do not load test Shopee, Lazada, TikTok Shop or your courier's API. Those are rate limited, the limits are theirs, and tripping them can suspend the integration you need most that week.

What is the decision if the test fails?

You have three levers, and you should pick before the day rather than during it.

Buy capacity - move to a bigger plan or a managed host for November and December. For most SMEs this is a few hundred dollars a month, which is cheap against a failed peak day.

Reduce the load - strip the live chat widget, review trackers and heavy scripts from the checkout, and release the sale in waves by email segment instead of one simultaneous blast. Staggering the announcement by thirty minutes across three segments cuts your peak concurrency by roughly two thirds at no cost.

Shift the volume - lean on the marketplaces for the first hours and treat your own site as the margin channel for later in the day.

Whichever you choose, name one person who watches the error rate on the day and holds the authority to switch the holding page on. A group chat is not an owner.

FAQ

Is this worth doing if most of our sales come from Shopee and Lazada?

Yes, but scale the effort. Your own site is where your best margin sits and where your brand is judged, and it is the one channel whose uptime you control. If it carries under 10 percent of peak volume, spend a day on timeouts, caching and a holding page rather than a full test programme.

How long does a useful load test take to set up?

One day to script the checkout journey and get a staging copy with real data, half a day to run and read the results, then two to five days of fixes and a re-test. Starting the week of 27 October still leaves room to fix what you find.

Our site was fine last 11.11. Do we need to test again?

Test again if anything changed - a new theme, new apps or plugins, a new payment method, a bigger catalogue, or more traffic. Each addition puts another dependency inside the checkout, and last year's result does not cover this year's configuration.

Next step: find your busiest hour of the past twelve months, multiply the orders by four, and ask whoever maintains your site one question - can the checkout take that, and how do we know? If the answer is an opinion rather than a test result, you have your task for this week.

Ready to Transform Your Business?

Let Digital Perpetual help you automate, streamline, and grow.

Get Started with Digital Perpetual →
load testing 11.11 e-commerce checkout peak trading website performance payment gateway