📈Adobe's Prometheus Proxy Solves Multi-Tenant Visibility
Teams get self-service access to Prometheus metrics
TL;DR
Adobe engineers have created an open-source proxy for multi-tenant Kubernetes, giving teams self-service access to Prometheus metrics. The proxy ensures data isolation and performance, reducing storage overhead by up to 99.7%.
Adobe has released an open-source proxy for multi-tenant Kubernetes environments, allowing teams to access Prometheus metrics without exposing other teams' data. This is crucial for GPU-heavy workloads, where visibility into power consumption and utilisation is key. The proxy restricts queries to a tenant's namespace, reducing storage overhead by up to 99.7% in some cases. Teams can now build custom queries to optimize GPU usage and reduce costs.
Key Points
Adobe's proxy identifies tenant, discovers Prometheus backends, and restricts queries to tenant's namespace.
Teams use MetricAccess Kubernetes custom resource to declare required metrics, using exact names, regex, or PromQL selectors.
Proxy enables periodic remote-writing of curated metrics into tenant-specific Prometheus instances.
MetricIsolation feature reduces stored series from over 10,000 to roughly 300 in one configuration.
Proxy keeps shared Prometheus for infrastructure-wide collection, while tenant-specific instances handle team queries.
Why It Matters
If you're managing GPU-heavy workloads in Kubernetes, Adobe's Prometheus proxy can help optimize GPU usage and reduce costs. Teams can build custom queries to identify idle GPUs and underutilized workloads, improving overall efficiency. The proxy's tenant-aware architecture ensures data isolation and performance, making it a game-changer for multi-tenant environments.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,556 builders reading daily.