Materialize v0.45

v0.45.0

Sources and sinks

• Expose source progress metadata as a subsource that can be used to monitor ingestion progress. The name of the progress subsource can be specified using the EXPOSE PROGRESS AS clause in CREATE SOURCE; otherwise, it will be named <src_name>_progress by default.

  Example

  -- Given a "purchases" Kafka source, a "purchases_progress"
  -- subsource is automatically created
  SELECT partition, "offset"
  FROM (
        SELECT upper(partition)::uint8 AS partition, "offset"
        FROM purchases_progress
  )
  WHERE partition IS NOT NULL;
  
   partition |  offset
  -----------+----------
   0         | 13645902
   1         | 13659722
   2         | 13656787
  

  For Kafka sources, the progress subsource returns the next possible offset to consume from the identified partitions, and for PostgreSQL sources it returns the last Log Sequence Number (LSN) consumed from the upstream replication stream.

SQL

• Improve the behavior of the search_path configuration parameter to match that of PostgreSQL. You can now specify multiple schemas and Materialize will correctly resolve unqualified names by following the search path, as well as create objects in the first schema named (i.e. the current schema).

• Support options settings on connection startup. As an example, you can now specify the cluster to connect to in the psql connection string:

  psql "postgres://user%40domain.com@host:6875/materialize?options=--cluster%3Dfoo"
  

• Add support for the \du meta-command, which lists all roles/users of the database.

• Add support for new SQL functions:

  Function	Description
  ceiling	Works as an alias of the ceil function.

• Remove the CREATE USER command, as well as the LOGIN and SUPERUSER attributes from the CREATE ROLE command. This is part of the work to enable Role-based access control (RBAC) in a future release (#11579).

Bug fixes and other improvements

• Improve the error message for naming collisions, specifying the catalog item type.

  Example

  CREATE VIEW foo AS SELECT 'bar';
  
  ERROR:  view "materialize.public.foo" already exists
  

• Fix a bug that would sporadically prevent clusters from coming online (#17774).

• Improve SUBSCRIBE error handling. Prior to this release, subscriptions ignored errors in their input, which could lead to correctness issues.

• Return an error rather than crashing if source data contains invalid retractions, which might happen in the presence of e.g. incomplete or invalid data (#17709).

• Fix a bug that could cause Materialize to crash when expressions in CREATE TABLE ... DEFAULT clauses or INSERT ... RETURNING clauses contained nested parentheses (#17723).

• Avoid panicking when attempting to parse a range from strings containing multibyte characters (#17803).