Statistically, September CommitFests feature the fewest commits. Apparently, the version 18 CommitFest is an outlier. There are many accepted patches and many interesting new features to talk about.

If you missed the July CommitFest, get up to speed here: 2024-07.

• Conflict detection and logging in logical replication
• Planner: no more 10000000000
• Planner: memory management and usage monitoring for temporary tuple storage
• Reading buffers during index backwards scan
• pg_upgrade: asynchronous operations in multiple databases
• Hash Join speed-up
• Faster text value processing in JSON
• Optimized numeric-type multiplication
• Optimized numeric-type division
• ANALYZE ONLY and VACUUM ONLY
• Improved checkpointer statistics
• pg_stat_statements: normalization of SET commands
• postgres_fdw_get_connections and remote connection status
• file_fdw: ignore format conversion errors
• New function has_largeobject_privilege
• Functions crc32 and crc32c
• Client-server protocol: report search_path changes to the client
• psql: support for named prepared statements
• pg_verifybackup: integrity verification for tar backups

• Planner: Hash Right Semi Join support
• Planner: materializing an internal row set for parallel nested loop join
• Planner support functions for generate_series
• EXPLAIN (analyze): statistics for Parallel Bitmap Heap Scan node workers
• Functions min and max for composite types
• Parameter names for regexp* functions
• Debug mode in pgbench
• pg_get_backend_memory_contexts: column path instead of parent, new column type
• Function pg_get_acl
• pg_upgrade: pg_dump optimization
• Predefined role pg_signal_autovacuum_worker

Since the PostgreSQL 17 RC1 came out, we are on a home run towards the official PostgreSQL release, scheduled for September 26, 2024.

Letʼs take a look at the patches that came in during the March CommitFest. Previous articles about PostgreSQL 17 CommitFests: 2023-07, 2023-09, 2023-11, 2024-01.

Together, these give an idea of what the new PostgreSQL will look like.

Reverts after code freeze

Unfortunately, some previously accepted patches didn't make it in after all. Some of the notable ones:

• ALTER TABLE... MERGE/SPLIT PARTITION (revert: 3890d90c)
• Temporal primary, unique and foreign keys (revert: 8aee330a)
• Planner: remove redundant self-joins (revert: d1d286d8)
• pg_constraint: NOT NULL constraints (revert: 16f8bb7c1)

Now, letʼs get to the new stuff.

SQL commands

• New features of the MERGE command
• COPY ... FROM: messages about discarded rows
• The SQL/JSON standard support

Performance

• SLRU cache configuration
• Planner: Merge Append for the UNION implementation
• Planner: materialized CTE statistics (continued)
• Optimization of parallel plans with DISTINCT
• Optimizing B-tree scans for sets of values
• VACUUM: new dead tuples storage
• VACUUM: combine WAL records for vacuuming and freezing
• Functions with subtransactions in parallel processes

Monitoring and management

• EXPLAIN (analyze, serialize): data conversion costs
• EXPLAIN: improved display of SubPlan and InitPlan nodes
• pg_buffercache: eviction from cache

Server

• random: a random number in the specified range
• transaction_timeout: session termination when the transaction timeout is reached
• Prohibit the use of ALTER SYSTEM
• The MAINTAIN privilege and the pg_maintain predefined role
• Built-in locale provider for C.UTF8
• pg_column_toast_chunk_id: ID of the TOAST value
• pg_basetype function: basic domain type
• pg_constraint: NOT NULL restrictions for domains
• New function to_regtypemod
• Hash indexes for ltree

Replication

• pg_createsubscriber: quickly create a logical replica from a physical one
• Logical slots: tracking the causes of replication conflicts
• pg_basebackup -R: dbname in primary_conninfo
• Synchronization of logical replication slots between the primary server and replicas
• Logical decoding optimization for subtransactions

Client applications

• libpq: non-locking query cancellation
• libpq: direct connection via TLS
• vvacuumdb, clusterdb, reindexdb: processing individual objects in multiple databases
• reindexdb: --jobs and --index at the same time
• psql: new implementation of FETCH_COUNT
• pg_dump --exclude-extension
• Backup and restore large objects