Microsoft Technology Licensing, LLC

Daniel Oh

Rahul Nigam

Nidhi Verma

Hans Christian Andersen

Pritvinath Obla

Rohan Khanna

Amber Litteken

The techniques disclosed herein enable systems to safely deploy a plurality of upgrade variants to different resource units that provide a service by utilizing small-scale deployment and validation. To deploy upgrade variants, a system receives a selection of upgrade variants from a feature group and automatically selects an appropriate set of resource units at which to deploy the upgrade variants. The system is further configured to collect and analyze telemetry data from the set of resource units to determine if any problems have occurred as a result of the deployed upgrade variants. By analyzing the telemetry data, the system can also identify one or more upgrade variants that are causing the problems. In response, the system can remove the identified variants and proceed with deployment of the remaining upgrade variants.

Dark deployment of infrastructure cloud service components for improved safety

Disclosed herein is a system for improving the user experience in the face of a regression by returning resources that offer a service to a 'last known good' upgrade. In other words, the state of the resources is reconfigured to scale back from recent upgrade(s), the deployments of which likely caused the regression, to a previous upgrade that is known to have little or no effect on the user experience. To identify a problem, the system collects performance data from different resource units that make up a cloud-based platform. The performance data is collected for each upgrade event in a sequence of upgrade events that are currently deployed or being deployed. The system continually tracks and analyzes qualification data collected for each of the deployed upgrade events. The system can tag an upgrade event as the last known good upgrade event when the collected qualification data satisfies predefined qualifications.

Tagging a last known good upgrade event for automatic rollback based on detected regression

Growing community of inventors

Redmond, WA, United States of America

Daniel Oh