Apache Druid: Why Speed In Analytics Matters
summary
Claims concrete numbers and architecture: Apache Druid targets millisecond latency for queries, ingests millions per second from Kafka/Kinesis and backloads from S3/GCS/Azure, and uses 700 MB segments with 10–15x compression via bitmap and dictionary encoding. Segments memory map off‑heap, queries vectorize and use late materialization, brokers route subqueries, and tiering isolates SLAs. The demo-level evidence is architecture plus reported cluster scale (up to ~2000 nodes) and practical ingestion/query strategies.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 2:02Datadriven culture
- 4:00Realtime analytics
- 7:50Smart homes
- 8:40Industrial 40 revolution
- 10:36Antimoney laundering
- 12:32Digital marketing
- 12:58What is Druid
- 14:49Key Features
- 16:41The Heart of Druid
- 19:10Segments
- 21:14Optimal segment size
- 22:08Segment structure
- 23:52Query example
- 24:14Late materialization
- 27:11Druid Architecture
- 30:36Ingest Architecture
- 32:09Query Flow
related talks
27mTales at Scale: Analytics at 1000 QPS and Beyond
28mThe Heat is On: Architecting for Hot Analytics
06mInterview
23mAdobe + Aerospike: Real-time Streaming Segmentation
32mDiscover the Modern Way to Build Real-Time Analytics Solutions at Any Scale
01h 03mExtreme Performance Architecture
38mCharacterizing the performance of Big Data processing platforms
45mAccelerating OpenSearch With Streaming: Apache Arrow, Flight, Data...