Apache Doris Lakehouse Guide

Lakehouse model

Query open data without building a second silo.

Doris can use its MPP, vectorized query engine across warehouse tables and external lake data exposed through catalogs.

01
SQL and optimizer
Plan joins and push computation toward connected sources.
02
Catalog and metadata
Resolve databases, tables, partitions, schemas, and snapshots.
03
Remote storage and systems
Keep authoritative data in open tables, files, or databases.

Catalogs

Give every external source a familiar SQL address.

A Catalog defines a connection and exposes its databases and tables through the same three-level namespace used by SQL.

CatalogSource + connector
DatabaseExternal namespace
TableSchema + files
PartitionPrunable metadata

Hive integration

Connect metadata and storage as one path.

A Hive Catalog needs both metastore access and permission to read the files described by that metadata.

Metadata plane

Resolve databases, tables, schemas, and partitions.

Hive Metastore endpoint
Catalog properties
Metadata events and cache

Data plane

Read or write the files behind those table definitions.

HDFS or object storage
Network and credentials
Format and snapshot support

Federated SQL

Join sources, but move the minimum data.

Doris can join external lake tables, external databases, and internal tables in one query while applying connector-aware pruning and pushdown.

ResolveCatalog metadata
PrunePartitions + columns
Push downSupported predicates
ScanParallel remote I/O
JoinDoris execution

Cache and I/O

Turn repeated remote reads into local work.

Metadata caches reduce catalog round trips, while data caching keeps frequently read remote blocks on local storage.

METAMetadata cache

Reuse table, partition, file, and snapshot metadata.

DATAData cache

Keep hot remote blocks on local disks near compute.

PUSHPushdown

Reduce columns, partitions, and rows before transfer.

Write-back and DDL

Treat external writes as a shared-system change.

Supported catalogs can write data or manage external objects, but the target format, metastore, permissions, and other engines must agree on the result.

CAPCapability

Confirm connector, format, and version support.

AUTHAuthorization

Grant only required metastore and storage permissions.

COMMITCommit behavior

Understand atomicity, snapshots, and failure cleanup.

READCross-engine check

Verify other consumers observe the intended result.

Lakehouse recap

Operate the connector boundary as a product.

A production lakehouse path combines correct metadata, secure storage access, efficient pruning, measurable caching, and tested write semantics.

ConnectCatalog + auth
QueryPrune + push down
AccelerateCache + materialize
GovernWrites + schema
OperateObserve end to end