Version 5.4.0

Released on 2023-07-11.

Note

If you are upgrading a cluster, you must be running CrateDB 4.0.2 or higher before you upgrade to 5.4.0.

We recommend that you upgrade to the latest 5.3 release before moving to 5.4.0.

A rolling upgrade from 5.3.x to 5.4.0 is supported. Before upgrading, you should back up your data.

Warning

Tables that were created before CrateDB 4.x will not function with 5.x and must be recreated before moving to 5.x.x.

You can recreate tables using COPY TO and COPY FROM or by inserting the data into a new table.

Table of Contents

Known Issues

  • Version 5.0.0 introduced a regression which can cause some JOIN queries to return no results when the query optimizer re-orders the joined tables. As a workaround, users should apply the following session settings before running such queries to prevent the query optimizer from re-ordering them and therefore produce the correct results:

    On CrateDB >= 5.4.0:

    SET optimizer_reorder_hash_join = false
    SET optimizer_reorder_nested_loop_join = false
    

    On CrateDB < 5.4.0 (this will disable the hash-join algorithm which may lead to poor performance):

    SET enable_hashjoin = false
    

Breaking Changes

  • Implemented several changes to the pg_catalog tables, involving the addition and removal of columns, and modification of column data types, to align with PostgreSQL version 14:

    • pg_attribute

      • Added: atthasmissing, attmissingval

      • Type Changed: spcacl from OBJECT[] to STRING[]

    • pg_class

      • Added: relrewrite

      • Removed: relhasoids, relhaspkey

      • Type Changed: relacl from OBJECT[] to STRING[]

    • pg_constraint

      • Added: conparentid

      • Removed: consrc

      • Type Changed: conbin from OBJECT to STRING

    • pg_index

      • Added: indnkeyatts

    • pg_namespace

      • Type Changed: nspacl from OBJECT[] to STRING[]

    • pg_proc

      • Added: prokind, prosqlbody, prosupport

      • Removed: proisagg, proiswindow, protransform

      • Type Changed: proargdefaults from OBJECT[] to STRING

    • pg_type

      • Added: typacl, typalign, typanalyze, typdefaultbin, typmodin, typmodout, typstorage, typsubscript

  • Raise an exception if duplicate columns are detected on named index column definition instead of silently ignoring them.

  • Adjusted allowed array index range to be from Integer.MIN_VALUE to Integer.MAX_VALUE. The behavior is now also consistent between subscripts on array literals and on columns, and between index literals and index expressions. That means something like tags[-1] will now return NULL just like ARRAY['AUT', 'GER'][-1] or ARRAY['AUT', 'GER'][1 - 5] did.

Deprecations

None

Changes

SQL Statements

  • Extended the EXPLAIN statement output to include the estimated row count in the output of the execution plan. The statement also has now options for ANALYZE and COSTS to have better control on the generated output plan.

SQL Standard and PostgreSQL Compatibility

  • Bumped the version of PostgreSQL wire protocol to 14 since 10 has been deprecated.

  • Added any_value as an alias to the arbitrary aggregation function, for compliance with the SQL2023 standard. Extended the aggregations to support any type.

  • Changed literal INTERVAL data type to do normalization up to day units, and comply with PostgreSQL behavior, e.g.:

    cr> SELECT INTERVAL '1 month 42 days 126 hours 512 mins 7123 secs';
    +------------------------------+
    | 'P1M47DT16H30M43S'::interval |
    +------------------------------+
    | 1 mon 47 days 16:30:43       |
    +------------------------------+
    
  • Added attgenerated column to pg_catalog.pg_attribute table which returns '' (empty string) for normal columns and 's' for generated columns.

  • Added the pg_catalog.pg_cursors table to expose open cursors.

  • Added the standard_conforming_strings read-only session setting for improved compatibility with PostgreSQL clients.

  • Allow casts in both forms: CAST(<literal or parameter> AS <datatype>) and <literal or parameter>::<datatype> for LIMIT and OFFSET clauses,

    e.g.:

    SELECT * FROM test OFFSET CAST(? AS long) LIMIT '20'::int
    
  • Added support for ORDER BY, MAX, MIN and comparison operators on expressions of type INTERVAL.

  • Added support for setting session settings via a "options" property in the startup message for PostgreSQL wire protocol clients.

    An example for JDBC:

    Properties props = new Properties();
    props.setProperty("options", "-c statement_timeout=90000");
    Connection conn = DriverManager.getConnection(url, props);
    
  • Added support for underscores in numeric literals. Example:

    SELECT 1_000_000;
    
  • Added support for updating arrays by elements, e.g.:

    UPDATE t SET a[1] = 2 WHERE id = 1;
    
  • Array comparisons like = ANY will now automatically unnest the array argument to the required dimensions.

    An example:

    cr> SELECT 1 = ANY([ [1, 2], [3, 4] ]);   -- automatic unnesting
    True
    
    cr> SELECT [1] = ANY([ [1, 2], [3, 4] ]); -- no unnesting
    False
    

Scalar and Aggregation Functions

Performance and Resilience Improvements

  • Improved the partition filtering logic to also narrow partitions if the partition is based on a generated column using the date_bin scalar.

  • Improved COPY FROM retry logic to retry with a delay which increases exponentially on temporary network timeout and general network errors.

Data Types

Administration and Operations

  • Added optimizer rules for reordering of joins for hash and nested-loop joins. This allows now to control the join-reordering and disable it, if desired, with session settings:

    SET optimizer_reorder_hash_join = false
    SET optimizer_reorder_nested_loop_join = false
    

    Note that these settings are experimental, and may change in the future.

  • Added a statement_timeout session setting and cluster setting that allows to set a timeout for queries.

  • The severity of the node checks on the metadata gateway recovery settings has been lowered from HIGH to MEDIUM as leaving these to default or suboptimal values does not translate into data corruption or loss.

  • Added the ability to set a storage_class for S3 repositories.