Weave Cortex: Horizontally Scalable, Multitenant Prometheus as a Service
summary
Weave Cortex implements a multitenant Prometheus as a service that splits Prometheus into retrieval, distributor and ingester microservices, stores 1KB compressed chunks in DynamoDB and S3, and hashes series into a Cassandra‑style token ring for replication. A cluster of six ingesters handles ~100k samples/s (≈300 samples/chunk, ~1.3 bytes/sample after encoding); distro latency 99th ≈120ms, query latencies ~100–500ms. The system runs in production and serves multiple tenants.
AI-written summary · about
chapters (11)
- 0:00Introduction to Cortex
- 1:32Design requirements
- 5:02System architecture
- 10:01Retrievers and configuration
- 11:45Distributor and scaling
- 15:48In-gester functionality
- 19:35Storage backend
- 22:53Demo and dashboard
- 26:48Performance evaluation
- 29:42Cortex vs Prometheus
- 32:42Future work and Q&A
related talks
37mIntro: Cortex
45mScaling Prometheus with Cortex, one time series at a time
25mCortex: Ingesting and Querying 100s of Billions of Samples Per Day
39mDeep Dive: Cortex
20mIntro to Scaling Prometheus with Cortex
20mCortex: Intro and Production Tips
30mCortex: Insights, U...
39mHow To Take Prometheus Planet Scale: Massively Large Scale Metrics Installations