Prometheus is an open-source time-series database widely used for monitoring and alerting. Prometheus scrapes metrics at regular intervals from endpoints called exporters, which expose data in a Prometheus format. The Prometheus server collects these metrics with HTTP requests and stores them, timestamped, in a searchable database.
Prometheus easily gathers metrics like CPU, memory, and network utilization from native Kubernetes exporters within an AI cluster. However, sophisticated AI workloads often involve shared infrastructure beyond compute nodes—especially storage platforms like Pure Storage FlashBlade™.
FlashBlade is ideal for scaling AI pipelines, enabling:
- Seamless access to high-volume shared datasets.
- Collaboration among data scientists.
- Efficient, scalable performance for both training and inference.
Modern Monitoring with OpenMetrics Exporter
Pure Storage has deprecated the legacy Pure Exporter and now strongly recommends adopting the Pure Storage OpenMetrics Exporter for FlashBlade. The OpenMetrics Exporter delivers:
- OpenMetrics standardization for broad ecosystem support.
- Stronger security: API tokens are handled via secure configuration files (not GET parameters), supporting modern authentication.
- Modular endpoints for granular metric selection and easier management.
Deployment Overview
- Install via Docker: The OpenMetrics Exporter for FlashBlade is best deployed as a Docker container. Build or pull the latest image from Pure Storage’s repositories.
docker build -t pure-fb-ome:<VERSION> .
# Replace <VERSION> with your desired version tag
Or pull the published image from Pure’s container registry.
- Configuration (Secure by Default): Supply the FlashBlade address and API token through a YAML config file:
flashblade_arrays:
fb-main:
address: <FLASHBLADE_IP_OR_HOSTNAME>
api_token: <YOUR_API_TOKEN>
Run the docker image with a bind-mounted config file for robust API credential management.
Prometheus Integration Example
Replace deprecated config parameters and use the OpenMetrics Exporter’s secure endpoints:
– job_name: ‘pure_flashblade’
scrape_interval: 30s
metrics_path: /metrics/flashblade
static_configs:
– targets: [‘<openmetrics_exporter_host>:<port>’]
- <openmetrics_exporter_host> and <port> are set according to your Docker/container network.
Grafana Visualization
Take advantage of detailed FlashBlade metrics to build advanced Grafana dashboards, leveraging improved metric categories and visualizations for capacity, performance, and reliability.
Key Reasons to Update Now
- Legacy Pure Exporter Is Deprecated: Switch to OpenMetrics Exporter for an officially supported, secure, and future-proof solution.
- Enterprise AI-Ready: The new exporter and FlashBlade platform together deliver unified observability for entire AI and data pipelines, ready for scale and cloud integration.
- Security and Compliance: YAM-based API credential management supports security best practices (least privilege, timely credential rotation).
FlashBlade continues to be the best place to store massive AI and monitoring datasets—offering performance, simplicity, and resilience. The 2025 Pure Storage vision champions enterprise data cloud and AI-ready infrastructure: all monitoring, security, and data management needs, unified for your organization’s success.
References:
See Also:
- Providing Data Science Environments with Kubernetes and FlashBlade
- Storing a Private Docker Registry on FlashBlade S3
Pure Storage Business White Paper
What Do AI Projects Really Demand from IT?
An AI primer for business leaders
