AI Model Monitor Agent 0.2.0 is only compatible with AI Inference Server 1.3.0 and newer, as the interface between these applications has changed. Monitoring AI Inference Server 1.2.0 is only possible with AI Model Monitor 0.1.0.
Each endpoint has a recommended minimum scraping time, the configuration file is shipped with them by default. It is not recommended to deviate from the minimal scraping times, because the application might show unexpected behaviour. By default, the application will not duplicate or cancel scrapes, so if a higher scrape frequency is configured to an endpoint than its response time, the application will produce a batch of errors referring to a 429 HTTP status code. In case this is experienced in the logs, a lower frequency time should be chosen.
The scrape time of the application is dependent on the response time of the scraped services and on the resources of the edge device. In case the scraped applications can not be reached, an error_metric is created.