Monitor TensorFlow I/O with Prometheus and Grafana Cloud
The open source project TensorFlow I/O from TensorFlow provides a Prometheus exporter so that you can scrape metrics from your server at any one given point in time. To get scalable, long-term metrics storage, configure your Prometheus instance to push scraped samples to compatible remote storage endpoints such as Grafana Cloud’s serverless Mimir database using using
remote_write. View the example below for how to configure
remote_write in the
prometheus.yml config file for TensorFlow I/O.
Prometheus Configuration for TensorFlow I/O
prometheus.yml config file for TensorFlow I/O can be found at the following location:
Push metrics from TensorFlow I/O to Grafana Cloud using
Below is an example
remote_write code snippet you will need to add to your
prometheus.yml config file in order to forward your metrics from your local Prometheus instance to remote storage with Grafana Cloud:
remote_write: - url: <Your Metrics instance remote_write endpoint> basic_auth: username: <Your Metrics instance ID> password: <Your Grafana.com API Key>
You can find the
https://...grafana.net/api/prom/push URL, username, and password for your metrics endpoint in the Cloud Portal. Click ‘Connections’ and select ‘Hosted Prometheus metrics’. The details display on the right. Need an account? Create a new Grafana Cloud account for free.
Learn more about the available
remote_write configuration parameters from the Prometheus.io Configuration Docs.
Control the number of metric data points pushed to Grafana Cloud:
Every Grafana Cloud account includes 10,000 active series metric storage, free-forever.
Grafana Cloud provides additional 1 data point per minute (DPM) resolution for $8 per 1,000 series. You can adjust the total DPM you push to Grafana Cloud by adjusting the
scrape_interval settings in your
prometheus.yml config file. Learn more about how to adjust your Prometheus scrape interval.