The bug, which began at 1600 GMT, affected Azure Front Door, the company's content delivery network service that is used by enterprise customers to optimize application performance.

In an update at 2230 GMT, Microsoft said it had finished deploying its "last known good" configuration and that some users could encounter "intermittent failures" as the system recovers.

Microsoft said it was seeing strong signs of improvement across affected regions and that it expected Azure operations to be back to normal by 2320 GMT.

The crowdsourced error reporting site DownDetector showed problems across a wide spectrum of customer-facing websites, including Xbox, Alaska Airlines and retailer Costco.

Configuration changes are routine in technology operations -- companies make them constantly to improve services, add features, or fix problems.

However, even a small error in configuration can cascade through highly interconnected systems and spread almost instantly to cloud customers worldwide.

"We are currently recovering nodes and re-routing traffic through healthy nodes across our fleet," Microsoft said in the update.

"This recovery effort involves reloading configurations and rebalancing traffic across a large number of nodes to restore full operational scale."

Last week, a different outage in Amazon's crucial cloud network, AWS, saw popular internet services ranging from streaming platforms to messaging services to banking taken offline for hours.

AWS leads the cloud computing market, followed closely by Microsoft's Azure, with Google Cloud in third place.

Businesses, governments and consumers worldwide rely on their infrastructure for online activities.