Turing routers (as well as enrichers and ensemblers) are deployed using the Knative framework, on top of the Istio Service mesh. Both tools provide some out-of-the box Prometheus metrics to track common stats such as error rate, request latency, etc. and can be configured by following the official guides:
The following custom Prometheus metrics are published by the Turing routers too.
mlp_turing_exp_engine_request_duration_ms
The duration for fetching a treatment from the experiment engine
Histogram
status, engine
Milliseconds
mlp_route_request_duration_ms
The duration for the call to a route
Histogram
status, route
Milliseconds
mlp_turing_comp_request_duration_ms
The duration for a custom operation in the code, useful for debugging
Histogram
status, component
Users are also free to publish their own custom metrics from the Enricher / Ensembler. All custom metrics (from the router, enricher or ensembler) should be scraped from the user-container pods for use.
Once the required dashboard has been created using the Prometheus metrics and other data, the template string RouterDefaults.MonitoringURLFormat can be used to configure the monitoring URL, when deploying the Turing application (please refer to the sample Helm values file for an example). This URL will be used by Turing when creating / editing a router and subsequently, in the Turing UI as a navigation link to the dashboard.
TODO
Last updated
