Aggregating Metrics In-Flight: Challenges and Opportunities
summary
Claims that pre-aggregation at scrape or client side—called Stream Aggregation—lets collectors reduce Prometheus load by aggregating in memory and sending fewer series; a VictoriaMetrics agent (VMAgent) prototype handles ~130k samples/s, a histogram query drops from ~11s to ~1s after aggregation, a hotspot with ~5,000 unique series/day (7% of storage) shrinks, and a custom label compressor yields about 5× memory reduction, with HA dup suppression and remote_write compatibility.
AI-written summary · about
chapters (12)
- 0:00In-flight metric aggregation intro
- 3:01High cardinality in metrics
- 7:38Traditional recording rules problems
- 10:11VictoriaMetrics agent for aggregation
- 13:04Streaming aggregation mechanics
- 18:29Streaming aggregation opportunities
- 24:16VMAgent processing pipeline
- 26:47Aggregation implementation challenges
- 30:01Label compression for memory
- 33:42Horizontal scaling with sharding
- 34:47Summary of streaming aggregation
- 36:12Q&A on intervals
related talks
20mAggregating Metrics In-Flight: Challenges, Opportunities
40mScale Your Monitoring Solution With the VictoriaMetrics Ecosystem
28mHow to reduce expenses on monitoring with VictoriaMetrics
11mAggregating, Alerting and Graphing on Millions of Prometheus T...
19mA History of Automatic Aggregations
36mVictoriaMetrics: scaling to 100 million metrics per second (EN)
31mObservability: Dust Off Prometheus with VictoriaMetrics
23mDesigning for High-cardinality Metrics