EventBridge:EventHouse overview

Core components

EventHouse consists of three core components. Each works independently but collaborates closely:

Core value

Zero-ETL (seamless data integration)

Map external data sources (such as RDS and OSS) directly. Run federated queries without moving data into EventHouse. This reduces data latency and storage cost.

Unified governance

Use Open Catalog to manage metadata and track lineage for “dark data” from message queues that lack schema definitions. Break down data silos.

Agentic analytics

Natively integrate MCP (Model Context Protocol). AI agents understand event data structures and perform analytics using natural language questions.

Unified metadata management

• Multi-source mapping: The catalog automatically discovers and registers metadata from data sources such as Kafka, RocketMQ, and RDS.

• Schema evolution: Automatically infer and manage schema versions for event data. When upstream fields change, maintain compatibility to prevent downstream analysis tasks from breaking.

• Data lineage tracking: Track events across their full lifecycle—from production (producer) and storage (eventstore) to analysis—to support troubleshooting and impact assessment.

Open ecosystem compatibility

Open Catalog supports open table formats such as Iceberg, Hudi, and Delta Lake. Avoid vendor lock-in and choose your compute engine freely.

Permissions and security

Provide fine-grained access control (ACL) at the database, table, and column levels.

Scenario: Unified data view

In E-commerce, order data may be split between RocketMQ (real-time stream) and MySQL (persistence). Use the catalog to create a unified view that logically joins the real-time order stream in MQ with user information tables in the database. Query this view directly—no need to know where data is physically stored.

Intelligent query engine

• Multimodal querying: Support three query modes—SQL (structured), NoSQL (document-based), and External (external data source queries).

• Unified stream and batch: Use the same SQL syntax to query historical archived data (batch) and real-time event streams (streaming).

• Materialized view: Precompute and cache frequent query results in materialized views for millisecond-level response times.

Federated query

• Cross-source joint analysis: Without data migration, use SQL JOIN to directly link internal tables in EventHouse with external sources such as OSS log files or RDS dimension tables.

• Predicate pushdown: Push filter conditions down to the source. Pull only necessary data to improve query efficiency.

Real-time anomaly detection

• Use built-in time window functions (Tumble, Hop, Session) to compute real-time metrics such as transaction success rate and latency distribution.

• Combine with a rules engine. When analysis results cross a threshold (for example, “more than 100 failed orders in one minute”), trigger an alert event automatically.

Technical advantages

DataAgent

Luma includes a built-in DataAgent that autonomously runs a “sense–plan–act” loop:

1. Sense: Detect an unusual drop in transaction volume.

2. Plan: Decide to query payment gateway logs and database connection pool status.

3. Act: Generate SQL for correlation analysis and output a root cause report.

Scenario: E-commerce risk control

1. An operator asks, “Were there any abnormal brushing behaviors in the last 30 minutes?”

2. The Luma agent uses MCP to fetch catalog info and identifies Transaction_Table and User_Behavior_Log.

3. The agent auto-generates a correlated SQL query (with time windows, IP aggregation, and device fingerprint analysis) and runs it in the EventHouse analysis engine.

4. Return a list of suspected brusher UserIDs and generate a risk report using the knowledge base.