Version 9.0.x (cloud native 3.0)

Data lakehouse

Readable and writable foreign tables

Readable and writable foreign tables are supported, allowing for read and write on a single foreign table.

-

Foreign table query performance optimization

Predicate pushdown for in and not in enhances partition pruning.

-

Complex data type support

Complex data types such as map, struct, array, and bytea/blob are supported.

-

Parallel foreign table export

Parallel exporting of foreign tables is supported with enable_insert_ft_dop (default: disabled).

-

INSERT OVERWRITE

The insert overwrite operation can be performed on either the entire table or individual partitions.

INSERT

HiveMetaStore statistics accessibility

HiveMetaStore statistics of external schema tables can be read.

-

Decoupled storage and compute

Asynchronous read optimization

Asynchronous read is optimized.

-

Data can be imported in parallel to tables with decoupled storage and compute.

Data can be imported to in parallel to OBS tables with decoupled storage and compute (enable_insert_dop enabled, which is disabled by default). Exporting data in parallel consumes more CPU and memory resources.

-

INSERT OVERWRITE for OBS tables with decoupled storage and compute

OBS tables with decoupled storage and compute support INSERT OVERWRITE.

INSERT

Hot and cold local disk cache

The local disk cache supports hot and cold settings. Different cache policies can be configured for different partitions.

-

Enhanced engine

jsonb/json for column-storage

Column storage supports the jsonb/json type.

-

Data lakehouse

One-click lakehouse

create external schema is used to directly connect to HiveMetaStore metadata, avoiding complex create foreign table operations.

-

Parallel foreign table ANALYZE

This function is controlled by the enable_parallel_analyze parameter. By default, this function is enabled, which significantly improves the speed of collecting ANALYZE statistics.

Note: Foreign tables do not have the AUTO ANALYZE capability. You are advised to manually perform ANALYZE to collect statistics after data import.

-

parquet/orc read and write for foreign tables

Overwriting, appending, and multi-level partition read/write are supported.

CREATE FOREIGN TABLE (SQL on OBS or Hadoop)

Foreign table read acceleration by disk cache

Foreign tables can use the local disk cache for read acceleration.

-

Foreign table data masking and row-level access control.

Foreign tables support datamasking and row-level access control.

-

Decoupled storage and compute

Decoupled storage and compute

You can specify COLVERSION=3.0 to create a column-store table as an OBS table with decoupled storage and compute. In this way, user data is stored in OBS and is charged on demand.

CREATE TABLE

Asynchronous read and write for OBS tables with decoupled storage

OBS tables that use decoupled storage can perform asynchronous reads and writes.

-

Parallel ANALYZE for OBS tables with decoupled storage

OBS tables with decoupled storage support parallel ANALYZE, which is controlled by the GUC parameter enable_parallel_analyze. This function is enabled by default.

-

Backup and restoration

OBS tables with separated storage and compute support only full backup and restoration.

-

Elastic VW

Elastic Virtual Warehouse (VW) supports read/write acceleration. It takes 12 minutes to start an elastic VW, which is irrelevant to the current node scale, number of objects, and data volume. Integrating additional elastic VWs can bolster the system's read/write throughput, concurrent processing, and resource isolation.

Periodically Adding or Deleting DWS 3.0 Logical Clusters

Elastic scale-out and ordinary scale-out.

Elastic scale-out does not redistribute data on OBS. Classic scaling redistributes all data. The system automatically determines the scaling mode based on whether the value of table_buckets divided by the total number of DNs exceeds the [min_table_buckets_per_dn, max_table_buckets_per_dn] range.

Impact on the System

OBS monitoring view

Added the OBS monitoring view PGXC_OBS_IO_SCHEDULER_PERIODIC_STATS.

-

Disk cache view

Added the disk cache views PGXC_DISK_CACHE_STATS and PGXC_DISK_CACHE_ALL_STATS.

-

Clearing residual files

Added functions pgxc_scan_residualfiles and pgxc_rm_scan_residualfiles_archive to detect and clear residual files.

-

Performance optimization

Memory sorting performance optimization

Optimized the memory sorting performance.

-