https://cwiki.apache.org/confluence/download/attachments/103092375/Pinot.pdf
Summary
- Column Store
- Characteristics
- Fast, interactive-level performance. E.g low latency queries
- p99 query latency 70ms
- p99 ingestion lag 30s
- 99,99 HA
- Exactly-once-ingestion
- Scalability
- Cost-effectiveness
- Low data-ingestion latency
- Flexibility → Ability to drill down arbitrarily without being bound by pre-aggregated results. Druid does not support This
- Optimized to handle high throughput of simple analytical queries and more complex queries at lower rates.
- Fault Tolerant
- Queries supported
- PQL, subset of SQL
- Supports selection, projection, aggregations, top-n queries
- Inverted Indexes
- Queries not supported:
- Joins or nested queries.
- No updates or deletions
- Record level creation, DDL
- Dependencies
- Zookeeper
- Helix (Cluster management)
Architecture
Comparison
Use cases
Practical details