Complete Analysis of New Features in Apache Doris 3.0 (2026)

Apache Doris 3.0 officially went GA in April 2026. As the fastest-growing project in the OLAP space, this major release brings several key changes.

1. Storage-Compute Separation Officially GA

Doris's traditional architecture is a shared-nothing, storage-compute integrated design:

┌──────────────────────────────────────┐
│  FE (Frontend)  ← Metadata + Query Plan │
│  BE (Backend)   ← Compute + Storage Together │
└──────────────────────────────────────┘

3.0 supports decoupled storage and compute deployment:

┌────────────┐    ┌──────────────────┐
│  BE (Compute) │───▶│ Shared Storage (S3/HDFS) │
└────────────┘    └──────────────────┘

Benefits: - Elastic Scaling: Compute nodes can scale independently without affecting data - Cost Optimization: Cold data on object storage, hot data on local SSD - Multi-Cluster Sharing: Multiple compute clusters share the same data

2. Arrow Flight SQL: A Leap in Query Performance

Previously, Doris returned results via the MySQL protocol, incurring significant serialization overhead. 3.0 natively supports Arrow Flight SQL:

# Connect directly using Arrow Flight SQL
mysql -h fe_host -P 9040 -u root  # MySQL protocol
# Arrow Flight: port 8060, columnar transfer

Benchmark comparison (10 million row query):

6x speedup + 60% bandwidth savings, because columnar data doesn't need to be deserialized into row format.

3. Semi-Structured Data Support

The VARIANT type is now GA, allowing you to store JSON in Doris and query it efficiently:

CREATE TABLE events (id BIGINT, payload VARIANT);
SELECT payload:user.name FROM events WHERE payload:event_type = 'click';

Under the hood, JSON is stored in a columnar format, and queries only read the required fields.

4. Inverted Index Enhancements

Full-text search capabilities have been significantly improved. You can now perform Elasticsearch-like text searches directly in Doris:

CREATE INDEX idx_content ON articles(content) USING INVERTED;
SELECT * FROM articles WHERE content MATCH 'Doris|OLAP';

Summary

The biggest significance of Doris 3.0: it has evolved from an OLAP engine into a unified real-time data platform. Storage-compute separation + Arrow Flight + semi-structured data enable it to cover a wider range of scenarios.

If you work in data backend engineering, I recommend paying attention to these three areas: storage-compute separation architecture, columnar transfer protocols, and the convergence of inverted indexes with OLAP.

References:

• Apache Doris Official Documentation