In the world of data analytics, the OLAP engine serves as the foundation of the platform and the driving force behind overall product performance. It must deliver exceptional data ingestion capabilities to support real-time data value while also providing high-performance query processing for demanding OLAP workloads.

Traditionally, higher performance often comes at the cost of greater infrastructure consumption. Reducing resource requirements while maintaining the same data processing capability creates significant value for both customers and product scalability. To achieve this, Bonree ONE replaced its legacy Druid architecture with the more lightweight and efficient ClickHouse engine.

As an integrated AI-native observability platform, Bonree ONE combines RUM, APM, data analytics, ETL, alerting, adaptive AI, and many other microservices. In its original architecture, these capabilities required 38 separate components, resulting in high resource consumption, complex maintenance, and heavy deployment requirements. A standard POC environment required a single server with 64 GB of memory while still offering limited performance.

To address these challenges, Bonree comprehensively redesigned the Bonree ONE architecture and introduced a lightweight architecture strategy.

Four Unified Layers, One Integrated Architecture

The Bonree ONE architecture follows the principle of simplifying complexity through Four Unified Layers and One Integrated Platform.

Unified Objects

Bonree ONE provides unified data collection through OneAgent, enabling consistent collection across multiple endpoints, data sources, and object types.

SmartGate further enhances data collection by providing intelligent aggregation, routing, compression, buffering, batch transmission, load balancing, and security management, ensuring reliable and efficient data acquisition.

Unified Data

OneData serves as the unified observability data platform, integrating metrics, traces, topology, events, logs, and other observability data types.

OneService, the unified data service layer, introduces Bonree's proprietary BPL (Bonree Processing Language), allowing computation and querying across heterogeneous data sources. This enables true data correlation and eliminates the fragmentation common in traditional observability platforms.

Unified Models

Bonree ONE standardises data modelling and combines it with AI-powered multimodal lakehouse capabilities, providing both out-of-the-box models and adaptive AI models for online analysis.

Unified Applications

All applications follow a unified OneAPI standard.

Powered by OneFramework, microservices communicate seamlessly while remaining fully modular. Applications can run independently or be deployed together as plugins. The platform also supports open integration, making Bonree ONE both extensible and easy to integrate with external systems.

One Integrated Platform

At the core of the architecture is complete integration.

Unified data, services, and models work together as a single system, enabling lightweight support for complex business scenarios, advanced data modelling, and high-performance analytics.

A Lightweight OneData Foundation

The redesigned OneData architecture significantly reduces infrastructure complexity while improving performance.

Key Advantages

• The data platform contains far fewer components. ClickHouse serves as the only OLAP engine, accompanied by only two lightweight services for data ingestion and metadata coordination, both supporting high availability.

• In POC environments, a streamlined ClickHouse deployment fully satisfies business write-throughput requirements.

• Small-scale clusters require only a single-shard ClickHouse deployment while still delivering high availability and excellent performance.

• ClickHouse natively supports partitioning, allowing late-arriving data to be processed with the same efficiency as real-time data.

• Materialised Views and multi-granularity tables enable millisecond-level query performance across large time ranges with consistently stable response times.

Ongoing Feature Development

Continuous enhancements include:

• Balanced write-resource scheduling

• Batch ingestion control at table granularity

• Multi-tenant support

• Automated data governance

• Improved query performance for complex data types

• Continuous optimisation of read/write performance and resource efficiency

• Intelligent data redistribution during cluster expansion

• Online DDL enhancements supporting dynamic addition and removal of columns for Materialised Views and Projections

• Lightweight operations and maintenance capabilities

Optimisation Results

Compared with the previous architecture, Bonree ONE delivers substantial improvements.

1. Component Reduction

• Total platform components reduced from 38 to 19, significantly simplifying deployment and maintenance.

2. Memory Optimisation

• POC deployment requirements decreased from 64 GB to 32 GB of memory.

• A single-node deployment now supports 100 agents while consuming approximately 23 GB of runtime memory.

• Heavyweight components including Hadoop, Kafka, BRFS, and ZooKeeper have been removed from the POC edition.

• Migrating the metrics storage engine from Druid to ClickHouse reduces memory consumption by 60% in both POC and clustered deployments.

• SwiftAI has been rearchitected, consolidating all AI services into a single process.

• OneService replaces the traditional Presto-based data service stack with a lightweight federated data service.

• OneAPI and OneView support both plugin consolidation and standalone deployment, replacing legacy SSO and Cloud components.

3. Storage Optimisation

APM storage requirements have been reduced to one-fifth of previous levels, with daily storage growth decreasing from 100 GB to 20 GB.

Contributing improvements include:

• Eliminating Kafka from POC deployments

• Trace write-compression optimisation

• Removal of Druid's secondary Hadoop storage layer

• ClickHouse compression ratios ranging from 8× to 20× depending on workload

• AI data governance combined with an Iceberg lakehouse architecture, requiring only 40 GB of storage for two months of default observability data

4. Performance Improvements

Based on production public cloud performance statistics:

• The ClickHouse-powered OneData platform delivers over six times higher P99 query performance than the previous Druid-based architecture.

• Average query latency has been reduced from seconds to milliseconds.

• Late-arriving data processing is significantly more stable and efficient.

• Ongoing maintenance costs for this capability have effectively been reduced to zero.

5. Expanded Functionality

Compared with previous product releases, Bonree ONE now enables AI capabilities and session analytics by default, without requiring additional infrastructure resources in either POC or clustered deployments.

6. Modern Application Framework

Bonree ONE introduces OneFramework, a powerful plugin container framework that provides a unified runtime environment and shared infrastructure capabilities, including:

• Dependency management

• Gateway services

• Connection pool abstraction

• Common utility libraries

• Shared session management

With OneFramework, development teams can focus exclusively on business logic while independently developing, deploying, and upgrading microservices. A single framework supports both public cloud and private deployment scenarios, significantly accelerating development, release cycles, and operational efficiency.