Abstract: Various embodiments of the disclosure disclosed herein provide techniques for detecting anomalies across one or more components within a distributed computing system, according to various embodiments of the present disclosure. An anomaly detection system retrieves event data associated with a real-time stream of events generated by one or more components within a distributed computing system. The anomaly detection system computes a failure metric based on the event data. The anomaly detection system determines that the failure metric exceeds a dynamically adjustable trigger condition. The anomaly detection system generates an alert associated with the failure metric.

Abstract: In one embodiment of the present invention, a sticky canary router routes each request associated with a service to either a canary cluster of servers that implement a modification to the service or a baseline cluster of servers that do not implement the modification. The sticky canary router implements a mapping algorithm that determines the routing of each request based on a current time, a time window for the routing, and a characteristic of the request. Notably, the mapping algorithm may be implemented such that, for time segments with duration equal to the time window, the sticky canary router routes all requests received from a particular device in a consistent fashion—either to the canary cluster or to a baseline cluster. Configured thusly, the sticky canary router enables the analysis of approximately full sections of client interactions with the canary servers, thereby facilitating identification of client-side effects of the changes.

Abstract: Various embodiments of the disclosure disclosed herein provide techniques for detecting anomalies across one or more components within a distributed computing system, according to various embodiments of the present disclosure. An anomaly detection system retrieves event data associated with a real-time stream of events generated by one or more components within a distributed computing system. The anomaly detection system computes a failure metric based on the event data. The anomaly detection system determines that the failure metric exceeds a dynamically adjustable trigger condition. The anomaly detection system generates an alert associated with the failure metric.

Abstract: In one embodiment of the present invention, a sticky canary router routes each request associated with a service to either a canary cluster of servers that implement a modification to the service or a baseline cluster of servers that do not implement the modification. The sticky canary router implements a mapping algorithm that determines the routing of each request based on a current time, a time window for the routing, and a characteristic of the request. Notably, the mapping algorithm may be implemented such that, for time segments with duration equal to the time window, the sticky canary router routes all requests received from a particular device in a consistent fashion—either to the canary cluster or to a baseline cluster. Configured thusly, the sticky canary router enables the analysis of approximately full sections of client interactions with the canary servers, thereby facilitating identification of client-side effects of the changes.