• turn off column cache when building an index, vacuum, and analyze (#250)
• fix memory leak when executing vectorized quals (#242)
• do not release the chunk group state if there is data (#245)
• fix case sensitivity bug in columnar.alter_table_set_access_method (#237)
• fix a crash when chunk_group_row_limit is larger than 110000 (#235)
• add extensions: wrappers, pgsodium, and supabase_vault (PG >=14) (#236)

Thanks to @japinli for multiple bug fixes in this release!

• fix uncapped memory growth when importing data via logical replication (#226)
• fix handling of large data columns (e.g. text or json) causing overflow of 1GB
  chunk size (#220)
• restrict the maximum singular datum size to 256MB to ensure that the maximum
  datum size of 1GB is not exceeded in columnar metadata (#222)
• do not insert vectorization node into the plan if columnar scan isn't being used
  (#228)

The corresponding docker images are:

• PG 16: docker pull ghcr.io/hydradatabase/hydra:16-dd0ef07209dd23b9fca213a152ee3ee13249bb34
• PG 15: docker pull ghcr.io/hydradatabase/hydra:15-dd0ef07209dd23b9fca213a152ee3ee13249bb34
• PG 14: docker pull ghcr.io/hydradatabase/hydra:14-dd0ef07209dd23b9fca213a152ee3ee13249bb34
• PG 13: docker pull ghcr.io/hydradatabase/hydra:13-dd0ef07209dd23b9fca213a152ee3ee13249bb34

• added support for upserts and other ON CONFLICT clauses for INSERT
  queries (#174)
• added a custom index-backed scan (#205). This scan is disabled by default as it
  may adversely affect performance. To enable this scan, use:
  SET columnar.enable_columnar_index_scan = true;
• added Postgres 16 support (#200)
  • parquet_s3_fdw and multicorn-based FDWs are not yet supported in PG 16. These
    extensions are included only on PG 13-15 builds.
• update pg_ivm to 1.7.0 (#200)
• update pgsql-http to 1.6.0 (#200)

The corresponding docker images are:

• PG 16: docker pull ghcr.io/hydradatabase/hydra:16-e8107709cc62316da9b2cb8fe0d3cb22ba6e470f
• PG 15: docker pull ghcr.io/hydradatabase/hydra:15-e8107709cc62316da9b2cb8fe0d3cb22ba6e470f
• PG 14: docker pull ghcr.io/hydradatabase/hydra:14-e8107709cc62316da9b2cb8fe0d3cb22ba6e470f
• PG 13: docker pull ghcr.io/hydradatabase/hydra:13-e8107709cc62316da9b2cb8fe0d3cb22ba6e470f

This version of Hydra Columnar requires ALTER EXTENSION columnar UPDATE after installation.

• fix incorrect results when using an aggregate FILTER by not vectorizing these aggregates (#181)
• fix possible bypass of table constraints by forcing constraints to be checked during multi-insert (#182)
• fix possible table corruption when running columnar.vacuum (#190)
• support chunk filtering for types that have a family comparator, namely varchar (#184)
• allow higher values (up to 10M) for chunk_group_row_limit and stripe_row_limit (#186)
• fix txid wraparound (#190)

The corresponding Docker tags are:

• PG 15: 15-d0087c3d78b336990f1d2ecd8e53ffe8561587e5
• PG 14: 14-d0087c3d78b336990f1d2ecd8e53ffe8561587e5
• PG 13: 13-d0087c3d78b336990f1d2ecd8e53ffe8561587e5

• updated pgvector to 0.5.1, allowing use of HNSW indexes.
  Users of pgvector should run ALTER EXTENSION vector UPDATE in any
  database where the extension is installed. (#171)
• stop resetting to default columnar on Hydra Cloud (spilo). This will allow
  the user to change the default. (#173)
• improve handling of stripe assignment to workers to reduce spinlock
  contention, which can cause a crash (#170)
• change default qual_pushdown_correlation_threshold from 0.9 to 0.4. This
  addresses a performance regression we observed in some clickbench queries.
  (#159)

The corresponding Docker tags are:

• PG 15: 15-22e7434a7e34d12a57a346e5d6c1d74da75f5046
• PG 14: 14-22e7434a7e34d12a57a346e5d6c1d74da75f5046
• PG 13: 13-22e7434a7e34d12a57a346e5d6c1d74da75f5046

Today we are pleased to announce Hydra 1.0 is GA. 🎉 You can learn more about the release and everything included in 1.0.0 on our web site.

There are no changes since v1.0.0-rc2.

The corresponding docker images are:

• PG 15: ghcr.io/hydradatabase/hydra:15-1.0.0
• PG 14: ghcr.io/hydradatabase/hydra:14-1.0.0
• PG 13: ghcr.io/hydradatabase/hydra:13-1.0.0

This is a bugfix release for a bug we found in RC1:

• bugfix: disable custom vectorization on non-column arguments (#147)

In addition to bug fixes, this release includes significant vectorization optimization for aggregates (eg.g COUNT and SUM), enabled new index types, and added pg_hint_plan.

• vectorization of direct aggregates (#143)
• fix an inefficiency with vacuum if there is only one stripe (#143)
• enable O3 compilation optimizations (#143)
• disable parallelism for CREATE TABLE ... AS queries (#138)
• allow gin, gist, spgist, and rum indexes on columnar tables (#133)
• add pg_hint_plan extension (#134)
• extensions are now installed during the build process using pgxman (#137)
• bugfix: columnar.alter_table_set_access_method should correctly disallow conversion of heap tables that have foreign keys (#136)
• bugfix: crash when using explain when cache is enabled (#125)
• bugfix: cache could evict a chunk currently in use (#142)

The corresponding Docker tags are:

• PG 15: 15-d099b1ce2517966f1344536f400f8e74bc79ba0a
• PG 14: 14-d099b1ce2517966f1344536f400f8e74bc79ba0a
• PG 13: 13-d099b1ce2517966f1344536f400f8e74bc79ba0a

Columnar-optimized vacuuming allows columnar tables to be compacted after updates and deletes without full rewrite, which will also improve peformance after vacuum. Auto-vacuum support combines recent inserts into a single stripe. A new column cache makes JOIN queries more efficient. Vector similarity search and data types are now available via pgvector.

• default table access method is now columnar (#107)
• add pgvector extension (#106)
• add vacuum_full UDF (#93)
• bugfix: vacuum udf could get into a look and overwrite stripes (#92)
• add columnar decompressed chunk cache (#86)
• bugfix: vacuum hanging indefinitely in some cases (#80)
• bugfix: release memory during long sequential scans (#78)
• add user function to allow for incremental vacuum and space reclamation (#71)
• add support for postgres 15 (#75)
• add columnar vacuum to combine latest stripe into 1 stripe (#51, #74)

The corresponding docker tags for this release are:

• Postgres 13: 13-c108761ca80fffb97f477d8c5aa36a000b2c3fe9
• Postgres 14: 14-c108761ca80fffb97f477d8c5aa36a000b2c3fe9
• Postgres 15: 15-c108761ca80fffb97f477d8c5aa36a000b2c3fe9