18.20. Release 302

General Changes

  • Fix cluster starvation when wait for minimum number of workers is enabled. (#155)
  • Fix backup of queries blocked waiting for minimum number of workers. (#155)
  • Fix failure when preparing statements that contain a quoted reserved word as a table name. (#80)
  • Fix query failure when spilling is triggered during certain phases of query execution. (#164)
  • Fix SHOW CREATE VIEW output to preserve table name quoting. (#80)
  • Add Elasticsearch Connector. (#118)
  • Add support for boolean type to approx_distinct(). (#82)
  • Add support for boolean columns to EXPLAIN with type IO. (#157)
  • Add SphericalGeography type and related geospatial functions. (#166)
  • Remove deprecated system memory pool. (#168)
  • Improve query performance for certain queries involving ROLLUP. (#105)

CLI Changes

  • Add --trace-token option to set the trace token. (#117)
  • Display spilled data size as part of debug information. (#161)

Web UI Changes

  • Add spilled data size to query details page. (#161)

Security Changes

  • Add http.server.authentication.krb5.principal-hostname configuration option to set the hostname for the Kerberos service principal. (#146, #153)
  • Add support for client-provided extra credentials that can be utilized by connectors. (#124)

Hive Connector Changes

  • Fix Parquet predicate pushdown for smallint, tinyint types. (#131)
  • Add support for Google Cloud Storage (GCS). Credentials can be provided globally using the hive.gcs.json-key-file-path configuration property, or as a client-provided extra credential named hive.gcs.oauth if the hive.gcs.use-access-token configuration property is enabled. (#124)
  • Allow creating tables with the external_location property pointing to an empty S3 directory. (#75)
  • Reduce GC pressure from Parquet reader by constraining the maximum column read size. (#58)
  • Reduce network utilization and latency for S3 when reading ORC or Parquet. (#142)

Kafka Connector Changes

  • Fix query failure when reading information_schema.columns without an equality condition on table_name. (#120)

Redis Connector Changes

  • Fix query failure when reading information_schema.columns without an equality condition on table_name. (#120)

SPI Changes

  • Include query peak task user memory in QueryCreatedEvent and QueryCompletedEvent. (#163)
  • Include plan node cost and statistics estimates in QueryCompletedEvent. (#134)
  • Include physical and internal network input data size in QueryCompletedEvent. (#133)