700+ API services moved to a new gateway platform with zero downtime

Context

Years of digital growth leave an insurer with a dense web of APIs: mobile apps, web channels, agencies, partner systems and internal platforms, all talking through one gateway. Here that gateway carried hundreds of services and high traffic around the clock. Customer-facing channels and back-office processes both depended on it.

As part of its technology roadmap, the client decided to move to a new API management platform. The question was not whether the new platform could do the job, but whether the move could be made without anyone outside noticing.

The constraints

More than 700 API services, each with its own configuration and routing rules
Hundreds of client applications integrated over OAuth 2.0
Client IDs and secrets could not be changed on the client side
Existing API addresses had to stay exactly as they were
No tolerance for downtime: the services run 24 hours a day
Any migration error would carry operational and financial risk

Our approach

Inventory before anything moved. Every endpoint, OAuth configuration, rate limit and routing rule on the existing gateway was read out and written down, including the behaviour nobody had documented.
One-to-one mapping. Each policy was mapped to its equivalent on the new platform. Identity integration was configured so that existing tokens, client IDs and secrets kept working unchanged.
Addresses preserved. Routing rules on the new platform reproduced the existing URL scheme exactly, so no client had to learn a new address.
Both platforms side by side. The new environment ran in parallel with the old one. Every service was validated in test environments before it carried production traffic.
Cut-over at the load balancer. Traffic was switched blue-green at the load balancer. Switching back was one step away for as long as it was needed.
Low-risk services first. Services with little traffic went first and were watched closely; the busy ones followed once their behaviour matched.

The result

More than 700 production APIs moved to the new platform with no downtime. OAuth credentials, token structures and authorisation flows stayed as they were, and so did every address. No client application had to change.

Because the cut-over happened at the load balancer, the platform under the client’s critical services was modernised without a maintenance window, and the organisation went on running its business on a platform that is easier to manage and to scale.

Lessons from this project you will find in the product

Measure before you move

The migration was decided on numbers: latency per route, error rates and who answered each error. In G8 those numbers are in the console from the first request, in Traffic Explorer and Analytics.

Many clusters, one place to manage them

The client ran several gateway environments, each with its own console and its own habits. G8 manages every environment from one installation, with access decided per environment.

Every change on the record

During the project, being able to say who changed which route and when was as important as the change itself. In G8 every write leaves an audit record before it reaches the gateway.