Architecture & Data Flow

Separation of Compute and Storage: Metadata, Compute, and Object Storage in Three Layers

Sources

TiDBReal-time synchronization with TiCDC
Aurora / MySQLData Integration
Kafka / LogsS3 Stage
S3 FilesCSV · Parquet · NDJSON
IcebergExternal Catalog

TiDB Cloud Lake

Metadata Service · RaftSchema · Clusters · RBAC · Security Policies

COMPUTE · Multiple warehouses, each independently scalable, suspendable, and billed separately

Ingestion
Batch Processing
BI
Ad Hoc Queries
Suspended$0

Storage · FuseEngineS3BYOBParquet

~8× columnar compression · MinMax / Bloom / inverted / vector / aggregate indexes

Consumers

BI ToolsTableau · Superset · etc.
ApplicationsJDBC · MySQL Protocol · Go
Data Engineeringdbt · SQLAlchemy · Jupyter
AI AgentsMCP Server
Write-BackTiDB · Redis · Services