Apache Druid на собеседовании Data Engineer

Что такое Druid

Open source real-time analytics DB. Sub-second queries на billions rows. Time-series oriented.

Use cases: real-time dashboards, click-stream analytics, network performance monitoring.

Архитектура

Multi-tier:

• Brokers. Query routing, merge results.
• Historicals. Serve data из segments.
• MiddleManagers. Ingest data, create segments.
• Coordinator. Manages segments distribution.
• Overlord. Manages ingestion tasks.

Metadata store — Postgres / MySQL.

Deep storage — S3 / HDFS.

Segments

Druid stores data в segments — partitioned by time.

2026-05-07T00:00:00Z to 2026-05-07T01:00:00Z (hour segment)

Каждый segment — column-oriented, compressed, indexed.

Distributed across historical nodes для parallel query.

Ingestion

Real-time. Kafka indexer reads streams → segments в memory → flushed disk периодически.

Batch. From files (Parquet, CSV), JDBC sources.

Native ingestion spec. JSON config defining schema, time column, dimensions, metrics.

{
  "dataSource": "events",
  "timestampSpec": {"column": "event_time"},
  "dimensionsSpec": {"dimensions": ["user_id", "event_type"]},
  "metricsSpec": [{"type": "count", "name": "count"}, {"type": "sum", "fieldName": "amount", "name": "revenue"}]
}

Queries

Druid SQL (since 0.10).

SELECT event_type, SUM(revenue) FROM events
WHERE __time >= TIMESTAMP '2026-05-01'
GROUP BY 1;

Native JSON query для advanced.

Sub-second на billions rows для time-bounded queries.

Druid vs ClickHouse

В РФ: ClickHouse dominates over Druid. Druid — niche real-time analytics.

Связанные темы

• ClickHouse MergeTree для DE
• TSDB для DE
• DWH ClickHouse для DE
• Snowflake vs BigQuery для DE
• Подготовка к собесу Data Engineer

FAQ

Это официальная информация?

Нет. Статья основана на документации Apache Druid.

Тренируйте Data Engineering — откройте тренажёр с 1500+ вопросами для собесов.