Version 4.3.0

Released on 2020-10-16.

Note

If you are upgrading a cluster, you must be running CrateDB 4.0.2 or higher before you upgrade to 4.3.0.

We recommend that you upgrade to the latest 4.2 release before moving to 4.3.0.

A rolling upgrade from 4.2.6+ to 4.3.0 is supported.

Before upgrading, you should back up your data.

Table of Contents

Breaking Changes

• Added support for the g flag to function regexp_matches and changed its type from scalar to table. It now returns a table where each row contains a single column groups of type array(text).

• Changed values of information_schema.columns.ordinal_position and pg_catalog.pg_attribute.attnum for object data type sub-columns from NULL to an incremented ordinal. Values of top-level columns may change if their position is after an object type column.

Deprecations

• Deprecated the *.overhead setting for all circuit breakers. It now defaults to 1.0 for all of them and changing it has no effect.

• Deprecated the indices.breaker.fielddata.limit and indices.breaker.fielddata.overhead settings. These no longer have any effect as there is no fielddata cache anymore.

Changes

Performance improvements

• Improved the performance of aggregations in various use cases. In some scenarios we saw performance improvements by up to 70%.

• Changed the default for soft_deletes.enabled. Soft deletes are now enabled by default for new tables. This should improve the speed of recovery operations when replica shard copies were unavailable for a short period.

SQL Standard and PostgreSQL compatibility improvements

• Added scalar function translate(string, from, to).

• Added support for SET AND RESET SESSION AUTHORIZATION SQL statements.

• Added scalar function pg_get_function_result.

• Added scalar function pg_function_is_visible.

• Added table function generate_subscripts

• Added the pg_catalog.pg_roles table

• Added full support for quoted subscript expressions like "myObj['x']". This allows to use tools like PowerBI with tables that contain object columns.

Administration

• Added a new cluster.max_shards_per_node cluster setting that limits the amount of shards that can be created per node. Once the limit is reached, operations that would create new shards will be rejected.

• Added the read_only_allow_delete setting to the settings['blocks'] column of the information_schema.tables and information_schema.table_partitions tables.

• Changed OPTIMIZE to no longer implicitly refresh a table.

• Changed the privileges for KILL, all users are now allowed to kill their own statements.

• Removed the Twitter tutorial from the Admin Console.

Error handling improvements

• Added detailed information on the error when a column with an undefined type is used to GROUP BY.

• Added detailed information to possible errors on repository creation to give better insights on the root cause of the error.

• Changed the error code for the PostgreSQL wire protocol from XX000 internal_error when:

  • a user defined function is missing to 42883 undefined_function

  • a column alias is ambiguous to 42P09 ambiguous_alias

  • a schema name is invalid to 3F000 invalid_schema_name

  • a column reference is ambiguous to 42702 ambiguous_column

  • a relation exists already to 42P07 duplicate_table

  • a column does not exist to 42703 undefined_column

  • a relation does not exist to 42P01 undefined_table

  • a document exists already to 23505 unique_violation

• Changed the error code for dropping a missing view from the undefined 4040 to 4041.

• Changed the error handling so it returns the error message and the related exception without being wrapped in a SqlActionException. Error codes remain the same.