Apache Doris — Modern MPP Analytical Database for Real-Time Reporting

What it is

Apache Doris is a high-performance, real-time analytical database built for online analytical processing (OLAP). It provides MySQL-compatible SQL, sub-second query latency on large datasets, and federated queries across data lakes including Hive, Iceberg, and Hudi.

Apache Doris targets data engineers and analysts who need fast dashboards, ad-hoc reporting, and real-time analytics without the complexity of a separate ETL pipeline.

How it saves time or tokens

Doris ingests data in real time and serves analytical queries immediately, eliminating the batch ETL window. You can query fresh data within seconds of ingestion. The MySQL protocol compatibility means existing BI tools (Grafana, Superset, Metabase) connect without custom drivers.

The built-in materialized views and rollup tables pre-aggregate common queries, reducing compute time for repeated dashboard requests.

How to use

1. Deploy Doris: download the binary or use Docker
2. Start the Frontend (FE) and Backend (BE) nodes
3. Connect with any MySQL client: mysql -h 127.0.0.1 -P 9030 -u root
4. Create tables and load data via Stream Load or Routine Load from Kafka

Example

-- Create an aggregate table for web analytics
CREATE TABLE page_views (
    event_date DATE,
    page_url VARCHAR(512),
    user_id BIGINT,
    view_count BIGINT SUM DEFAULT '0'
)
AGGREGATE KEY(event_date, page_url, user_id)
DISTRIBUTED BY HASH(user_id) BUCKETS 16
PROPERTIES ('replication_num' = '1');

-- Load data via Stream Load
-- curl -u root: -H 'format: json' -T data.json http://fe:8030/api/db/page_views/_stream_load

-- Query with standard SQL
SELECT page_url, SUM(view_count) as total_views
FROM page_views
WHERE event_date >= '2026-04-01'
GROUP BY page_url
ORDER BY total_views DESC
LIMIT 10;

Related on TokRepo

• Database tools -- Database management and analytics
• Monitoring tools -- Dashboards and real-time monitoring

Common pitfalls

• Doris requires separate FE and BE processes; minimum production setup is 3 FE nodes and 3 BE nodes for high availability
• Choosing the wrong data model (Aggregate, Unique, or Duplicate) affects query performance significantly; read the model guide before designing tables
• Stream Load has a 10 GB default limit per request; batch large imports into smaller chunks