HomeResourcesCase study
Case study

Razorpay Throttled Its Own Merchants — Then Rebuilt

By The SDL team·3 min read·Updated Sep 16, 2026

The famous microservices migration started with a payments monolith buckling under IPL-season traffic. The lesson is isolation, not speed.

Razorpay broke its monolith into microservices, and everyone remembers it as a "modernization." The actual trigger was an IPL season where a flood of traffic forced them to throttle their own paying merchants.

In 2019, gaming and food-delivery surges during the Indian Premier League hit Razorpay's payment monolith hard enough that the only lever left was to slow down merchants — the worst thing a payments company can do during the moment merchants are making money.

Plain English

A monolith is one big application where every feature shares the same process and the same fate. Convenient to start with. The problem: when one feature gets slammed — say, payment creation during a flash event — it starves every other feature of resources, because they all live together. You can't add capacity to just the part under pressure; you have to scale (and risk) the whole thing.

Splitting into microservices means each capability runs as its own service that can be scaled, deployed, and fail independently. The goal isn't speed. It's isolation.

Now the engineering

Razorpay decomposed the monolith in phases — carving out payments, refunds, settlements, and more into independent services — rather than attempting a risky big-bang rewrite. Around the new architecture they layered the controls that actually tame spikes: autoscaling plus pre-provisioning ahead of known events (you can't autoscale fast enough for an instantaneous IPL surge, so you pre-warm), cache TTL with jitter to prevent synchronized expiry stampedes, and event-driven queues to absorb bursts asynchronously instead of forcing everything through synchronous paths.

IPL traffic met the monolith — and the monolith throttled merchants monolith payments + refunds + settlements all share fate spike in ONE overwhelms ALL payments svc refunds svc settlement svc scale + isolate independently autoscale + pre-provision cache TTL + jitter event queues absorb bursts Microservices didn’t make it faster. They let the team scale the part under fire without scaling — or risking — everything else.
Isolation, not speed. Splitting the monolith let Razorpay scale the service under fire without scaling or endangering the rest.

Worth knowing

The phased approach is the quiet wisdom here. A big-bang monolith-to-microservices rewrite is one of the most reliable ways to destroy a company's quarter. Razorpay extracted services incrementally, keeping the system shippable throughout — the same caution Figma and Notion showed in their migrations. Evolution beats revolution when the system is load-bearing.

The gap it reveals

"Microservices scale better" is a half-truth that hurts juniors. The real insight is that microservices buy fault and resource isolation — the ability to scale and protect the hot component independently — at the cost of new distributed-systems complexity. Knowing that the IPL fix was pre-provisioning and queues (not merely 'splitting the monolith'), and that the migration must be phased, is the operator's view.

In the interview room

If you propose microservices in a design round, the sharp follow-up is "what does that actually buy you here, and what does it cost?" Strong answer: independent scaling and blast-radius isolation for the hot path, paid for with network calls, distributed transactions, and operational overhead — plus pre-provisioning for known spikes since autoscaling lags. That balance signals you've run this, not just read about it.

The reframe

Architecture decisions are rarely about elegance; they're about which failure you're trying to prevent. Razorpay didn't want microservices — they wanted to never again throttle a merchant because an unrelated feature was on fire. Name the failure first, and the architecture chooses itself.

Microservices aren't a speed upgrade. They're a blast-radius decision.

Primary source →
Nirakar Asha on Medium — How Razorpay Rebuilt Its Payments Platform: From Monolith to Microservices

Want feedback on your design?

The weekly teardown

One real-world architecture, every week.

How real companies actually built it: the design, the trade-offs, and what to say about it in an interview. Free, and one click to unsubscribe.

Related articles