Microsoft Technology Licensing, LLC

Min Shao

Shobana Krishnamoorthy

Ankur Jauhari

Nidhi Verma

Randy David Thomson

Divyachapan Sridharan Padur

Ilker Celikyilmaz

Jun Zhang

Rahul Nigam

Patrick Simek

Brian G O'Connor

Roberta Cannerozzi

Roshane Silva

Zohar Raz

Gaia Carini

Prashant Gaurav

Sudharsan Ganesan

Morten Haavaldsen

Reliable feature deployment is provided. Features are evaluated for their readiness to be graduated and features that are ready to be graduated are added to a graduation list. When features are added to the graduation list, evaluation of the features is discontinued, thereby reducing the performance and runtime costs associated with deploying features. Furthermore, engineers can consult the graduation list to identify features that are ready for graduation, thereby reducing the risk of premature flight graduation. Data associated with features is analyzed to map the features to corresponding flights. In this way, when an engineer identifies a feature for graduation, the engineer is able to identify the corresponding flight (or flights) from the mapping. Even when a feature is selected for graduation, the present systems provide additional safeguards to ensure that corresponding flights are not improperly or prematurely graduated, thereby preventing a subpar customer experience.

Reliable feature graduation

The techniques disclosed herein enable systems to deploy upgrade events in a controlled manner to different resource units that provide a service utilizing predefined rollout policies. To deploy an upgrade event, a system determines a risk factor for the upgrade event and presents predefined rollout policies to a feature group for selection based on the risk factor. Upon selection of a rollout policy, the system can deploy the upgrade event according to parameters defined by the selected rollout policy. The system is further configured to analyze telemetry data received from the resource units to determine an updated risk factor and determine whether the updated risk factor crosses a risk factor threshold. If the updated risk factor crosses the threshold, the system can identify another rollout policy to replace the selected policy. In addition, the system can receive an override workflow request to expedite upgrade event deployment.

Reliable deployment of upgrade events for a service based on predefined rollout policies

Issues may arise when customers that are highly engaged with application services do not receive reliable feature updates on a predictable schedule. For instance, releasing and updating features may burden cloud infrastructure and/or may not perform properly. Not only are unpredictable feature updates associated with inconsistent customer messaging and potential service disruptions, such unpredictable updates may cause a random end user experience and customer dissatisfaction with the product. The methods and systems address these issues by dynamically ranking tenants and distributing updates and releases to tenants with lower feature usage before distributing to tenants with higher feature usage. The ranking of the tenants may dynamically change based on current usage. In this way, higher ranked tenants receive access to new and updated features after additional regression testing. This quality-driven, phased release of features enables load level management of computing resources as well as a predictable feature experience for end users.

Predictable distribution of program instructions

Various systems and methods for flight delivery are disclosed herein. In one example, a system includes a processor to process a flight configuration delivery service instruction to schedule the flight by sending the flight to a service bus based on an identification of a temperature of the farm. The system may also include the service bus to alert the farm of the flight for the farm to consume, wherein the service bus returns a notification to the flight configuration deliver service for a farm that does not consume the flight. In an example, a memory stores, based on instructions from the flight configuration delivery service, an identification of the farm that did not consume the flight in the memory.

Flight delivery architecture

A hosted service may include multiple applications and have a large number of active features, each in varying stages of deployment, that are gradually rolled out in one or more deployment environments. The hosted service may include a flight state machine, which may provide policy based flight management for feature rollout. For example, in response to receipt of a flight request associated with a feature of the hosted service, the flight state machine may be configured to retrieve a policy associated with rollout of the feature, determine a flight state, and determine a deployment state of the feature. The flight state machine may then be configured to initiate or continue deployment of the feature in a current deployment environment or initiate deployment of the feature in a next environment based on the policy, the flight state, and the deployment state of the feature.

Policy driven flight management

Technologies are described for updating a live system with static changes. According to some examples, changes such as installing and/or removing features may be applied to a cloud based service in live environment without disrupting the service. New software may be patched and installed on an upgraded machine (i.e., server). New features, diagnostics, and language packs may be added. Existing features and certificates may be removed. To accomplish these, new software to be installed and existing software to be removed may be detected. Removal and additions may be performed without preparation and activation of replacement farms.

Updating live system with static changes

A cyclical patching process associated with a cloud application may be defined to ensure high availability (HA) of the cloud application in order to prevent impacting an availability to end users. A list of server identities corresponding to one or more servers of a datacenter hosting the cloud application may be accepted. HA metric values for each of the server identities may be determined in order to compute an overall HA metric value for the cloud application. A subset of the servers may be removed from a rotation framework of the cloud application based on the determined HA metric values, where the removal does not affect the overall HA metric value of the cloud application. One or more patches may be applied to each server within the subset of servers in parallel, and the subset of servers may be reinstated in the rotation framework of the cloud application.

Growing community of inventors

Bellevue, WA, United States of America

Min Shao