Aggregating Metrics In-Flight: Challenges, Opportunities
summary
Claims that a stream-aggregation collector (VictoriaMetrics agent) can precompute and drop raw series to prevent cardinality explosions, reduce query cost and cut ingestion overhead. Experiment: scraping 130k samples/sec; Prometheus with recording rules uses ~22 GB RAM versus the agent ~5 GB, and a recorded query drops from ~11s to ~1.66s (~8x faster). Agent adds ~1 CPU and ~1 GB when aggregating, supports downsampling, sharding, and prometheus remote_write, but requires careful per-metric design and correct scrape topology.
AI-written summary · about
chapters (8)
- 0:00Introduction to Aggregating Metrics
- 0:23Understanding High Cardinality
- 1:27Dealing with High Cardinality
- 2:24Relabeling and Recording Rules
- 6:32Stream Aggregation Concept
- 8:58Stream Aggregation in Practice
- 17:32Challenges and Considerations
- 18:54Conclusion and Q&A
related talks
38mAggregating Metrics In-Flight: Challenges and Opportunities
28mHow to reduce expenses on monitoring with VictoriaMetrics
11mAggregating, Alerting and Graphing on Millions of Prometheus T...
40mScale Your Monitoring Solution With the VictoriaMetrics Ecosystem
19mA History of Automatic Aggregations
31mObservability: Dust Off Prometheus with VictoriaMetrics
29mStreaming Recording Rules for Prometheus, Thanos, and Cortex Using...
23mAutomatically Making Dashboards Load 100X Faster