A method comprises collecting parameters corresponding to at least one microservice operation processed by a first instance of a microservice, and analyzing the parameters using one or more machine learning algorithms. Based at least in part on the analyzing, a prediction is made whether the at least one microservice operation is anomalous. In the method, the first instance of the microservice is designated as being in an anomalous state responsive to predicting that the at least one microservice operation is anomalous. One or more microservice requests for the microservice are routed to a second instance of the microservice responsive to the anomalous state designation.