Cloud SPE — Update #3
Period: March 1, 2026 – March 31, 2026
Status: On track
───
Summary
March marked the most intensive phase of the project — transitioning from alpha infrastructure into a production-ready analytics platform. We completed 6 of 7 total iterations, finalizing the data pipeline, hardening infrastructure, expanding the API surface well beyond the original scope, and delivering the first draft of SLA scoring logic.
The single largest effort this period was data analysis and quality assurance across the analytics pipeline. This work, spanning over 5 weeks across M1 and M2, fell squarely into what we call “hidden costs” — unplanned but essential work that surfaces in any data-intensive project. Getting the data right before building on top of it was non-negotiable.
With M2 now complete, the platform is live, usable, and ready for community consumption. Our final milestone (M3) is targeted for April 10, 2026.
───
Completed Deliverables
Milestone Progress: Production Pipeline, Hardened Infrastructure & Expanded API
Evolved the analytics platform from an alpha environment into a production-grade system with hardened infrastructure, a finalized data pipeline, and an API surface that significantly exceeds the original scope.
• Iterations Completed: 6 of 7 total iterations
• Finalized Analytics Pipeline:
End-to-end data processing pipeline fully implemented and validated for production use.
• Daydream Data Loading via Kafka (MirrorMaker 2):
Ingested 35+ million historical records into the analytics platform using Apache Kafka with MirrorMaker 2, enabling comprehensive network-wide analysis.
• Secured & Hardened Infrastructure:
Multiple Cloud SPE Ubuntu servers running Kafka, ClickHouse, API Server, Prometheus, Grafana, and the Analytics Pipeline — all hardened for production workloads.
• Expanded API Server (46 Endpoints):
Delivered the 3 promised APIs — GPU Metrics, SLA Compliance, and Demand — plus an additional 43 endpoints exposing a wide range of network data slices. We initially put the APIs inside of the legacy leaderboard-serverless api, but decided that service was not the ideal place for the suite of new apis going forward. In fact we will probably end up deprecating it. The new API is live at: NAAP Analytics API (NAAP Analytics API)
• Improved AI Job Tester:
Significant enhancements to live video-to-video workload testing:
• High, Medium, and Low prompt complexity testing
• Orchestrator cap detection
• Improved error handling
• Individual orchestrator tests plus full-network tests covering orchestrator selection, swap behavior, and failure rates
• SLA Scoring Logic (First Draft):
Implemented the initial SLA scoring model. The logic is usable and live today — designed to be adjusted and refined as users consume and provide feedback on the data.
• Data Analysis & Quality Assurance:
Extensive validation and quality checks across the analytics pipeline. This effort spanned 5+ weeks across M1 and M2 and represented the largest single unplanned cost in the project — a necessary investment to ensure data integrity before building production APIs on top of it.
───
Key Resource
Same repositories as prior updates:
• Naap Analytics Pipeline & API (GitHub - Cloud-SPE/livepeer-naap-analytics · GitHub)
• Project Plans NaaP MVP: Make Network Data More Observable · GitHub
• Grafana Dashboard https://grafana.livepeer.cloud/
• API Endpoints NAAP Analytics API
───
What’s Next — M3 (Final Milestone)
ETA: April 10, 2026
• Final forum post with full project summary
• All final documentation published
• Project wrap-up and handoff
• Analytics Dashboard & API Demo and “Peek” behind the scenes of the analytics pipeline.
───
Looking Ahead
This API project represents the beginning of network analytics infrastructure for Livepeer — not the end. To continue and expand this work beyond M3, it will require community involvement or additional treasury funding. We will continue working with the Livepeer Foundation and the NaaP platform team to scope and adjust the API and Analytics Pipeline to suit the network’s evolving needs.