US 11,755,725 B2
Machine learning anomaly detection mechanism
Amey Ruikar, San Francisco, CA (US); Carl Meister, Seattle, WA (US); Tony Wong, San Francisco, CA (US); Charles Kuo, Saratoga, CA (US); Aishwarya Kumar, Fremont, CA (US); Wayne Rantala, Aurora (CA); and Shailesh Govande, Milpitas, CA (US)
Assigned to Salesforce, Inc., San Francisco, CA (US)
Filed by Salesforce, Inc., San Francisco, CA (US)
Filed on Jan. 30, 2019, as Appl. No. 16/261,753.
Prior Publication US 2020/0242240 A1, Jul. 30, 2020
Int. Cl. G06F 21/55 (2013.01); G06N 20/00 (2019.01); G06Q 30/01 (2023.01)
CPC G06F 21/552 (2013.01) [G06F 21/554 (2013.01); G06N 20/00 (2019.01); G06F 2221/034 (2013.01); G06Q 30/01 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method, comprising:
obtaining a first plurality of performance metric messages at a database system;
extracting a first plurality of anomaly detection data from the first plurality of performance metric messages;
obtaining a second plurality of performance metric messages at the database system;
extracting a second plurality of anomaly detection data from the second plurality of performance metric messages;
performing maintenance of a monitoring component;
storing the first plurality of anomaly detection data and the second plurality of anomaly detection data by a queueing system in memory separate from a first execution engine and a second execution engine;
distributing, by a load balancer, the first plurality of anomaly detection data to the first execution engine from the memory;
distributing, by the load balancer, the second plurality of anomaly detection data to the second execution engine from the memory;
determining, by the first execution engine, whether one or more data points in the first plurality of anomaly detection data is anomalous by applying a first machine learning model to the first plurality of anomaly detection data;
determining, by the second execution engine, whether one or more data points in the second plurality of anomaly detection data is anomalous by applying a second machine learning model to the second plurality of anomaly detection data, wherein the determining of the one or more data points in the first plurality of anomaly detection data and the second plurality of anomaly detection data is performed while the monitoring module undergoes maintenance without the loss of the first plurality of anomaly detection data and the second plurality of anomaly detection data at the first and second execution engines; and
generating, by a monitoring component, one or more alerts according to a result of the determining whether one or more datapoints in the first and second plurality of anomaly detection data are anomalous, the monitoring component operating independently from the first and second execution engines.