Release 1.3.0 (30 Jun 2021)

Key Features

AreaFeaturePR #s
Task RecoveryFixed a few important bugs and further improved the stability of this function. It now can work with spill-to-disk feature.812,813,837,
CTE (Common Table Expressions)Additional optimization on top of 1.2.0 CTE optimization. Added cost based decision to decide whether to enable CTE or not. Added support for pushdown of dynamic filters and predicates into CTE nodes.722,811,815,
DM (Data Management)Further improved the performance of Data Management Operations. Exposed performance tuning parameters as:
- metastore -client-service-threads: Parallelize operations to Hive metastore by using multiple clients to send/receive requests
- metastore -write-bach-size: Reduce round trip to hive metastore by packing multiple operation objects per call
Star Tree Index1. Star Tree Cube now supports up to 10 billion cardinality.
2. openLooKeng CLI updates to improve cube management experience. User now has to issue a single sql statement to both create and populate data in the cube instead of multiple sql statements. The CLI changes help avoid query exceeded cluster memory limit issue.
3. Bug fixes
a. Merge continuous ranges into single range so cube can be utilized
b. Count distinct issue: Filter source data during cube insertion
CBOSupport Sorted Source Aggregator

Added support for sort based aggregator in cases where input source is pre-ordered. This greatly reduces the amount of memory used for hashes and can finalize the majority of the results at the partial aggregation stage itself, thereby reducing the final aggregation load at a higher plan stage.

The openLooKeng optimizer makes choices between Sort Aggregator and Hash Aggregator based on the estimated cost of operation for the given memory.
Hudi ConnectorSnapshot queries for Hudi COW table is supported; snapshot queries and read optimized queries for HUDI MOR table are supported.881,900
GreenPlum ConnectorSupport read and write operations on the GreenPlum datasource. But update and delete operations are not yet supported.689
Oracle ConnectorAdd new capability to support update and delete operation within Oracle datasource.897
ClickHouse ConnectorSupport read and write operations to the ClickHouse datasource.

Also add support for SQL query pushdown, and registration & pushdown of external functions.
JDBC ConnectorEnhance JDBC to support the multiple splits so that it can improve the performance of high concurrency scenarios.939
Hive ConnectorUpgrade the Hive dependency from 3.0.0 to 3.1.2 and fixed the compatibility issue of timestamp caused by the upgrade.903
Memory ConnectorMemory Connector Optimizations

- HetuMetaStore integration to persist table info
- New data formation (LogicalParts) to support sorting and indexing
- Predicate pushdown
- Automatic spill-to-disk management
ResourceEnhanced resource group to throttle scheduling or kill query based on resource usage and user configurations.779,821,822,

Known Issues

CategoryDescriptionGitee issue
Task RecoveryAn error message: “Unsuccessful query retry”, is shown when CTA creates a transaction table and inserts data.I3YF45
A query can hang when there is insufficient memory for a node.I3YF4O
When an exception is thrown during stage 1, the value is doubled.I3YF4V

Obtaining the Document

For details, see