What's new in the YugabyteDB v2025.2 LTS release series

Improvements

YSQL

• Ensures YSQL index backfill operations resume seamlessly after a master leader failover. #6218
• Disallows interleaving SAVEPOINT and DDLs in a transaction block if transactional DDL is enabled, ensuring the safe execution of DDLs in Read Committed (RC) isolation. #3109
• Fixes a use-after-free bug in postponed DDL operations to prevent potential issues in transaction commits. #3109, #26707
• Allows setting and modifying options for foreign tables using ALTER FOREIGN TABLE ...OPTIONS (...). #1124
• Allows execution of multiple TRUNCATE statements within the same sub-transaction without causing errors. #3109, #27203
• Simplifies handling of temporary tables with on-commit actions by committing the DocDB transaction after executing these actions, preventing errors related to unnecessary transaction states. #3109, #27307
• Fixes the ANALYZE command error occurring when run outside of a transaction block, ensuring smoother operations. #3109, #27125
• Resolves a bug in DDL atomicity workflow that incorrectly committed transactions involving table creation and deletion in the same block, preventing PG and DocDB schema divergence. #3109, #27359
• Enables configuration of transactional DDL blocks via the yb_ddl_transaction_block_enabled flag. #3109, #27715
• Resets the DDL state for read-only transactions to avoid state persistence issues. #3109, #27650
• Introduces the yb_fk_references_cache_limit parameter to cap FK references cache entries at 65535. #27739
• Allows specifying colocation option for parent partition tables and splits parent partition tables into a single tablet. #16209, #27314
• Enables transactional Data Definition Language (DDL) as a regular flag for interleaving DDL and Data Manipulation Language (DML) statements within a transaction block, enhancing system catalog read/write operations. #3109
• Enables transactional DDL in YSQL, allowing DDLs to run within a transaction block and interleave with DML operations, while ensuring DDL atomicity. #3109, #27616
• Enables speculative execution for certain PL statements to avoid unnecessary flushes, enhancing performance. #27464
• Simplifies control flags for the auto analyze service, merging two existing flags into one, easing the use of the auto analyze feature. #24548
• Reduces the frequency of auto-analyze on small, frequently-updated tables. #27627
• Supports executing DDL statements in READ COMMITTED isolation with options to disable retries. #27716
• Adds extra locality info from PostgreSQL to TServer for better transaction categorization and performance. #28270
• Add a new YSQL configuration parameter that allows all DDL commands to increase the catalog version. #28253
• Reduces catalog cache memory spikes during new connections by introducing a TriggerRelcacheInitConnection RPC that allows only superuser postgres to rebuild the relcache init file. #28758
• Limits relcache init file rebuilding to internal connections only, reducing the memory spike associated with new connections. #28758
• Adds debug logs to help diagnose errors when rebuilding the relcache init file. #28758
• Reduces potential catalog cache memory spikes by enabling internal connections to perform minimal catalog preload. #28758
• Adds a lint rule to ensure the pairing of Form_* and FormData_* lines in yb_typedefs.list, increases code readability, and updates yb_typedefs.list and yb_thirdparty_typedefs.list. #27319
• Adds support for deferred mode in yb_read_after_commit_visibility for writes. #24940
• Adds estimated roundtrip counts to EXPLAIN output for scans. #25063
• Adds support for LSM index creation on multirange columns. #27297
• Displays write RPC metrics in EXPLAIN (ANALYZE, DIST, DEBUG) output. #26086
• Logs now detail the cause and context of read restart errors for better troubleshooting. #24431
• Adds a new YSQL parameter yb_enable_cbo for safer query optimization without statistics. #26868
• Enables integration of YSQL with PG's snapshot management for Repeatable Read isolation level. #28333
• Introduces a per-backend in-memory cache for the hints table in the pg_hint_plan extension, reducing query planning time by avoiding inter-node lookups for each query execution. #26828
• Adds YB-specific counters to pg_stat_statements for detailed stats tracking. #26791
• Re-imports latest commit files from yb-pg15 branch, cherry-picks YB commits, removes unneeded commits, updates ysql_bench metrics, and fixes other minor issues. #24822
• Simplifies code management by separating cancelable and non-cancelable scopes. #26751
• Refactors the PgGate fetch control, introduces new classes for combining results from multiple PgsqlOp objects, optimizes index scan read requests for copartitioned vector indexes, and removes unnecessary index requests. #22826
• Limits output buffer to 8kB to ensure compatibility with certain clients. #26744
• Enhances tserver by adding support for garbage collection of invalidation messages, reducing memory usage. #26940
• Enhances accuracy of cost modeling for backward index scans based on the use_fast_backward_scan flag. #26991
• Updates the pg_stat_statements header to reset stats on upgrade from 2.25 to 2.27. #27180
• Enhances Postgres logging integration by utilizing the ErrorData.backtrace field, adjusting the control of backtrace addition through the backtrace_functions parameter, and introducing a yb_debug_original_stacktrace_format flag to switch stacktrace formats. #26447
• Increases the timeout for detecting pg_yb_catalog_version mode from 10 seconds to 20 seconds. #26990
• Streamlines UPDATE/DELETE operations and standardizes function signatures in YSQL. #27179
• Allows DDL statements to update the shared catalog version, reducing unnecessary master RPC calls and improving performance. #27028
• Prevents auto-analyze DDLs from blocking or preempting user DDLs. #27036
• Removes unnecessary YB_TODO comment in pl_exec.c related to memory context lifespan. #27060
• Optimizes catalog version updates on DDL statements to avoid unnecessary refreshes. #27080
• Adds support for cube and earthdistance extensions in YSQL upgrades. #27141
• Adds a new YSQL configuration parameter to disable the hint table cache. #27151
• Allows cleaner and when-needed building of dummy_seclabel in build_postgres.py instead of adding it as a default subdir. #25892
• Reduces unnecessary full catalog cache refreshes by calling update functions atomically. #27239
• Streamlines the database by removing obsolete functions related to drop/remove object by OID. #27352
• Removes the YB_BITMAP_DISCOURAGE_MODIFIER to better reflect actual bitmap scan costs. #23565
• Enhances the function to find the first catalog version more quickly using std::lower_bound instead of a linear scan. #27316
• Updates the yb_pgindent run and fixes some related issues for better code style and format. #27341
• Removes unused macro CHECK_REL_PROCEDURE2 and cleans up obsolete YB_TODO in heapam/indexam. #27328
• Automates directory switch and indent option computation in yb_pgindent to simplify usage. #27384
• Removes a redundant YB_TODO comment in finalize_partialaggregate that was introduced during the PG 15.2 initial merge. #27419
• Adjusts the bad_variable_declaration_spacing lint rule to ignore macros and exclude certain paths, enhancing code consistency. #27418
• Removes obsolete tasks in pg_dump.c, enhancing initial privileges retrieval for extension objects during backup/restore. #27423
• Skip lint rules for certain files, reducing warnings/errors and simplifying exclusion logic. #27446
• Removes a duplicate and outdated recordDependencyOnTablespace, enhancing the initial merge process. #27467
• Adds pss_mem_bytes field to yb_active_session_history for tracking process memory usage. #27474
• Enables sending buffered write operations and multiple read operations in a single RPC for optimized performance. #27432
• Allows setting the flag ysql_yb_enable_invalidation_messages to disable incremental catalog cache refresh. #27286
• Enables setting ysql_yb_enable_cbo flag on tserver for query optimization. #27495
• Switches to __thread in regex code for thread-local variables, ensuring correctness and safety in multi-threaded environments. #26847
• Allows YbctidReaderProvider to work with Slice instead of std::string for optimized PgFKReferenceCache memory usage. #28175
• Enables parallel scanning in all safe cases, thereby enhancing speed and performance. #23149
• Reduces unexpected full catalog cache refreshes in new PG backends. #27533
• Allows turning a stored generated column into a normal one with ALTER TABLE ALTER COLUMN DROP EXPRESSION. #27456
• Adds verbose logging for Perform RPCs between Pg and Tserver via shared memory. #27571
• Modifies the bad_comment_style rule to bypass lint errors for nested comments in d2s_intrinsics.h and d2s.c files. #27623
• Adjusts lint rule to ignore false positives for pg_hint_plan comments inside string. #27655
• Rectifies PostgreSQL lint rule bad_spacing_after_comment violations and upgrades it from warning to error. #27693
• Enhances brace lint rules to ignore false positives from code comments and macros, and upgrades them from warnings to errors. #27740
• Enhances logging for troubleshooting unexpected full catalog refreshes by adding CatalogInvalMessagesDataDebugString. #27387
• Displays storage flush reasons under debug logs for better query performance analysis. #27588
• Simplifies conditional cleanup actions with MakeOptionalScopeExit function. #27786
• Replaces the rss_mem_bytes column in pg_stat_activity with pss_mem_bytes for more accurate memory measurements. #27768
• Removes the beta label for various Foreign Data Wrapper commands in YSQL. #9979
• Streamlines request pagination in scans for consistent behavior across all directions. #27738
• Allows upgrade_ysql to exit early if YSQL is not enabled. #27167
• Adds GUC_EXPLAIN flag to YB-specific GUC parameters for enhanced EXPLAIN output. #26864
• Simplifies code for Perfrom requests, enhancing readability and usability. #27965
• Optimizes table cache by using invalidation messages to clear only relevant entries, reducing OpenTable RPC count. #28049
• Displays TCMalloc and PGGate memory stats in pg_log_backend_memory_contexts for enhanced debugging. #27879
• Introduces hnsw.ef_search as a new YSQL configuration parameter, mirroring ybhnsw.ef_search. #26629
• Displays read operations in EXPLAIN for better insight into the number of operations, unaffected by request parallelism. #28028
• Reduces memory allocations by using a generator instead of non-reused vectors in PgDmlRead::SetHoldingRequestedYbctids method. #27991
• Enhances row boundary settings for hash-partitioned tables using encoded DocKeys. #28219
• Adds the ability to display commit stats with the EXPLAIN command, helping users understand latency in multi-region workloads. #26964
• Introduces a new yb_tablet_metadata view that allows users to fetch tablet distribution and leadership information across the entire cluster. #28107
• Simplifies the PgTableCache interface reducing boilerplate code by accepting a single shared pointer of type PgTablesQueryTableReceiver instead of two independent pointers. #28069
• Simplifies adding types to yb_typedefs.list and running yb_pgindent. #28063
• Enables the now function to be evaluated once, replaced with a constant, and pushed down to DocDB, enhancing query execution. #26497
• Ensures all COPY FROM commands in a transaction block use one transaction unit. #27061
• Adds a new column to the yb_wait_event_desc view for more context about wait events. #26571
• Reduces master load by allowing auth backends to utilize the tserver cache during authorization checks. #28144
• Includes transaction isolation level in serialization error messages. #28114
• Simplifies query handling via shared exchange by substituting callback in PgClientData with std::promise, saving memory allocation. #28205
• Set YSQL configuration parameter yb_read_after_commit_visibility to relaxed to avoid restart read required errors during yb_index_check. #27288
• Enables bounded staleness for PostgreSQL authentication backend cache using pg_cache_response_trust_auth_lifetime_limit_ms. #28261
• Reduces master RPCs for PG auth backends by using tserver cache. #28144
• Allows dynamic updates to ysql_hba_conf_csv without server restarts. #28361
• Reduces chances of multiple notifications to listeners when opening multiple tables concurrently by refining the Waiter helper class. #28069
• Adds functionality to the three extra columns in pg_stat_statements for tracking query retry counts. #28446
• Changes default password encryption from md5 to scram-sha-256 for new clusters without requiring extra steps. #5399
• Allows altering table constraints using the new ALTER TABLE syntax. #3945
• Reduces memory usage by limiting the collection of intents in PgFKReferenceCache to the max batch size. #28175
• Adds new RPC stats to pg_stat_statements and introduces a new flag yb_enable_pg_stat_statements_rpc_stats for capturing RPC wait times. #28474
• Allows VERBOSE output in YSQL configuration parameter yb_explain_hide_non_deterministic_fields mode. #28515
• Shifts documentdb build logic to a dedicated Makefile for more reliable checks during building process. #28559
• Add a new counter to track new connections performing catalog preloads, aiding in debugging. #28538
• Introduces DEBUG_ONLY macros to keep debug injection code compact and improve readability. #28554
• Refactors YbcFlushDebugContext to reduce copying and fix pgindent errors by passing it as a pointer. #28605
• Updates copyright format to match the most current version, "...YugabyteDB, Inc.", across all files. #28591
• Refactor YbRunWithPrefetcher code to avoid errors when running yb_pgindent and improve code formatting. #28706
• Avoids unnecessary memory spikes and ensures freshest catalog data for relationship cache initialization. #28144
• Adds new YSQL configuration parameter yb_enable_cbo=legacy_ignore_stats_bnl_mode. #28703
• Adds support for docdb_rows_returned column in pg_stat_statements to count rows returned by the storage layer. #28505
• Expands EXPLAIN (ANALYZE, DIST) output to include transaction type, enabling a more formal way to deduce transaction type. #14804
• Allows pushdown of stable and parameterless SQL Value Functions as storage filters. #28997
• Increases timeout for tserver catalog version wait to 30 seconds, reducing connection time outs. #28978
• Enables Early Access to Query Diagnostics from 2025.2. #28959
• Allows preloading of RANGEMULTIRANGE to reduce cache misses when pg_range is preloaded. #28991
• Fixes pg_total_relation_size to include index sizes, ensuring accurate and consistent size reporting across YugabyteDB's distributed storage. #23181
• Adds a new ysql_userid column to Active Session History to track user activity and query performance. #28565
• Reduces peak memory usage in yb_active_session_history by filtering out rows more quickly. #29089
• Adds variable bloom filter support for queries with multiple keys, enhancing efficiency during scan operations. #28439
• Cleans up YbBindSearchArray, enhancing code efficiency and preparing for future bucketized index work. #28834
• Auto analyze flag now defaults based on user-set cost-based optimizer (CBO) flags, reducing bad combinations. #28944
• Adds a check to prevent autonomous DDL transactions during an active regular transaction block. #29239
• Simplifies type conversions in index backfill code to enhance consistency and prevent potential overflow. #29038
• Adds four transaction-specific wait events to Active Session History for better operation tracking. #28792
• Disables Memoize under BNL to reduce CPU overhead without sacrificing performance. #25251
• Reduces race conditions in DROP DATABASE operations by waiting for stats update in Connection Manager. #27644
• Adds a flag ysql_enable_scram_channel_binding to enable SCRAM with channel binding. #28108
• Reduces query execution latency by optimizing GUC handling in Connection Manager. #20939
• Introduces a flag to enable or disable the import of stats during pg_upgrade, reducing memory usage and enhancing performance. #26240
• Adds a flag to simulate DDL operations during YSQL major upgrade. #27185

YCQL

• Removes unused code that allows changing the table namespace, simplifying the source code. #27315

DocDB

• Tracks ByteBuffer memory usage with MemTracker. #26875
• Adds data block consistency check to sst_dump tool. #27233
• Adds flag to allow tablet writes after a compaction failure. #27269
• Adds remove_corrupt_data_blocks_unsafe flag for yb-ts-cli compact_tablet command. #27381
• Ensures only voter peers are considered for leadership during a leader stepdown. #27568
• Removes outdated fields from RaftGroupReplicaSuperBlockPB to avoid confusion, ensuring safe upgrade and rollback. #25582
• Fixes overflow in FormatBytesAsStr to prevent large memory allocations. #27924
• Adjusts the wal_size metric to include the size of the current active, writable segment, ensuring accurate reflection of on-disk size. #15274
• Changes default value of ysql_num_tablets to 1 for initial number of tablets in hash partitioned tables. #28060
• Adds logging to detect and inform if scheduled tasks in the reactor run slower than expected. #24801
• Optimizes index scans by adaptively deciding to use next or re-seek, reducing query latency. #28616
• Reduces load balancer's hold on Catalocustomeranager::mutex_ to prevent potential master leader unavailability. #29183
• Prevents data loss during early compaction of replicated colocated tables. #25926
• Introduces asynchronous lock functionality in ObjectLockManager to avoid holding up rpc threads during conflicts, and includes deadlock detection. #24669
• Allows DDL release operations to withstand master restarts or failovers and ensures successful execution of release requests even when network issues occur. #25680, #26498
• Enhances table locking by propagating lock calls to the tserver and expanding lock acquisition to all objects. #26803
• Displays RBS progress details on the master's Cluster Balancer UI page. #26853
• Introduces a debug tool to release global/exclusive table locks via yb-admin command, assisting in managing unexpected lock leaks. #26526
• Support now correctly captures table IDs and colocation IDs during DDL replication, ensuring accurate replication for unique constraints, partitioned indexes, and table rewrites. #27081, #26453
• Allows outstanding transactions to continue during DDL operations when table locks are enabled. #26869
• Allows tracking of table-level lock states across all nodes, with accurate reporting in the pg_locks view. #24670
• Triggers automatic background compaction for vector index based on size amplification, using the vector_index_max_size_amplification_percent flag. #27089
• Sets xcluster_backfill_hybrid_time for all indexes to ensure no missed rows during backfill after data insert. #27445
• Upgrades OpenSSL to version 3.0.16 and enhances aarch64 support. #27491
• Disables statement retries when object locking feature is on, due to lack of re-acquisition of object locks during retries in extended query mode. #27520
• Ensures consistent xCluster safe time during failover, including executed DDLs, by pausing all pollers and updating the safe time accordingly. Avoids issues during normal scans and inconsistencies during failover. #23957
• Disables transparent statement retries in all paths when the object locking feature is activated. #27520
• Switches YSQL object lock calls to use shared memory, reducing overhead to 5-15%. #27510
• Allows seamless failover to the consumer cluster by using a variant of PITR that only restores data tablets, not pg catalog tables or yb master tablet. #27595
• Allows transaction status tablets to be grouped by tablespace for future categorization of transactions. #28269
• Ensures reliable xCluster history cutoff by backfilling namespace_id when necessary. #28124
• Adds support for tablespace-local localities, allowing transactions to use status tablets corresponding to a specific tablespace, improving data replication handling for geographically close regions. #28273
• Limits TServer crashes by capping log reader memory usage in xCluster replication. #28124, #22992
• Enables safe activation and deactivation of the skip prefix locks feature and allows for safe upgrades. #28823
• Allows control over the number of tasks running in parallel via the introduction of PriorityThreadPoolToken. #29151
• Control the number of vector index compactions per tserver with new flags, offering control over chunks taken for compaction and limiting running compactions per tserver. #27601
• Promotes object locking preview flag to a regular flag, allowing users to control the feature through the enable_object_locking_for_table_locks flag. #29317
• Adjust leader lease to be larger than the heartbeat interval to avoid unnecessary retries and latency. #1348
• Optimizes colocated table read performance by caching tombstone check results, making bulk updates 6x faster. #17843
• Adds a metric to monitor the wait time of write and remove threads in joining the write batch group. #26548
• Adds new wait events for ASH instrumentation to identify bottlenecks during bootstrap. #23198
• Integrates RWQueue for enhanced concurrent queue performance and introduces TrivialSpinlock for improved spinlock functionality. #26818
• Enables ysql_yb_enable_advisory_locks by default and makes it a runtime flag. #25293
• Enables cloning of vector indexes in databases. #26872
• Allow tablet server to wait for YSQL lease before starting postmaster, improving connection handling. #25880
• Adds paginated querying for vector index operations. #26884
• Allows setting ybhnsw.ef_search to customize search expansion factor. #26883
• Allows partial processing of DDLs during failover by incrementing xCluster safe time after each DDL replication. #25999
• Removes libcds dependency, simplifying threading model and reducing runtime costs. #26926
• Enforces shared memory access restrictions at compile time. #26924
• Cancels vector index compaction during VectorLSM shutdown. #26489
• Enables saving of in-memory HNSW graph data to file system, enhancing usability and efficiency. #26937
• Enables consistent backup and restore for vector indexes. #26965
• Replaces most instances of LOG(ERROR) with either LOG(WARNING) or LOG(DFATAL) to align with coding style. #27042
• Enables priority thread pool usage for vector index compactions. #26971
• Adds detailed logging for failed DDL executions in xCluster replication. #23702
• Enables faster table lock acquisition through shared memory in YugabyteDB. #25518
• Stores object lock requests in shared memory to enhance crash safety. #25987
• Enhances lock management by disabling fastpath during lock conflicts. #27813
• Adds ASH instrumentation to snapshots for better tracking of backup and snapshot bottlenecks. #23200
• Adds base structure for background compaction in vector index. #27227
• Allows async functionality in TSLocalLockManager lock calls for enhanced database operations. #27238
• Defers processing of records with commit times exceeding the apply safe time to the next batch. #27318
• Corrects spelling error in DEPRICATED_pgschema_name to Deprecated_pgschema_name safely. #27332
• Reduces wait time for deadlocked transactions to abort by timing out waiters after FLAGS_refresh_waiter_timeout_ms. #26903
• Adds a flag to control the use of YbHnsw for vector indexing, requiring a drop and recreate for changes. #27417
• Allows detection of deadlocks involving fast path transactions when object locking is enabled. #24630
• Refactors shared memory management for better session control between tserver and pggate. #27484
• Enables chunks picking for background compaction based on size ratio. #27377
• Enables conditional tracing of ASH events to conserve memory. #27455
• Allows exporting snapshots with tables undergoing DDL operations. #27425
• Reverts commit disabling outstanding transactions during DDL until table locks feature is fully enabled. #27472
• Changes the default value for the ysql_allow_duplicating_repeatable_read_queries flag to true. #27652
• Increases limits for concurrent replica removals and over-replicated tablets to speed up cluster operations. #27488
• Updates the Tablet Server UI page for better organization and readability. #27821
• Removes -ftime-trace from Clang builds to enhance compatibility with ccache. #27499
• Adds memory tracking to vector indexes for better resource management. #27548
• Enhances check_data_blocks command output and logging to include data block number, index key, and distinguish corruption. #27594
• Automatically adds ASH metadata to selected RPCs, maintaining a list of exceptions where metadata shouldn't be added. #26305
• Speeds up cluster load balancing by considering tablet sizes for parallelism. #17561
• Handles out-of-order messages to prevent lock leakage by cleaning up after specified ignore_after_hybrid_time. #27705, #26393
• Reduces risks by not updating master_ysql_operation_lease_ttl_ms flag unnecessarily. #27681
• Replaces yb::ThreadPool with YBThreadPool to enhance concurrent access and reduce latency by 5-8%. #4624
• Converts xcluster_enable_ddl_replication from a preview flag to an kExternal flag. #27831
• Enables TRUNCATE TABLE support in xCluster automatic mode with new flags and triggers. #23958
• Streamlines operations by removing unnecessary flag checks on the master for table locks. #27521
• Ensures xCluster DDL replication tables are included in backups. #27959
• Reduces latency of YSQL write statements and lock-grabbing reads in distributed transactions by enabling async writes. #28382
• Reduces unnecessary transaction conflict errors by auto-retrying status requests. #27973
• Removes unnecessary log line that occurs whenever a tablet is deleted. #27980
• Enhances on-disk size metrics to include snapshots, improving cluster balancer accuracy. #28010
• Adds metrics for tracking fastpath object lock acquires. #27344
• Turn on enable_object_lock_fastpath by default in Linux release builds, not affecting debug builds. #28065
• Allows table servers to attempt to release their YSQL lease during shutdown, enabling other servers to process DDLs and DMLs without delay. #27832
• Enables core dumps on FATAL errors with new flag disable_core_dumps_on_fatal. #28007
• Disables object locking for PgTabletShutdownTest to allow DDLs like DROP/TRUNCATE to abort in progress DMLs. #28161
• Enables enable_object_lock_fastpath and pg_client_use_shared_memory in all Linux builds. #28165
• Adds hnswlib as an additional backend for hnsw vector index, enhancing search performance options. #28278
• Boosts search performance by skipping deletion checks on tables that only perform inserts. #28229
• Adds metadata and wait events to xCluster poller, enhancing visibility into bottlenecks. #23202
• Enables dependency on the transactional DDL flag for the object locking feature to prevent self-deadlocks. #28212
• Enables automatic selection of optimal distance function based on user's CPU capabilities, enhancing USearch's efficiency. #28329
• Enhances reliability by retrying exclusive lock requests during YSQL lease membership changes. #27192, #28198
• Allows execution of ANALYZE command on both primary and standby clusters independently. #28328
• Allows checking multiple replication infos and tablet numbers for creating multiple tables simultaneously. #28331
• Refactors boost::optional to std::optional for consistency and removes unused files, enhancing error handling in arclint. #28353
• Enables AVX implementation in HNSWlib for better vector index performance using advanced CPU capabilities. #28364
• Optimizes asan/tsan builds to use -O3 level and allows optional debug checks with new asan_release/tsan_release builds. #28553
• Allows parallel merging of vector index chunks using the insert priority thread pool to reduce compaction time. #28397
• Raises the minimum GCC version to GCC 12, compatible with all supported build platforms. #28540
• Adjust assert logic in PgCatalogVersionTest to handle changes when object locking and transactional DDL are enabled. #28413
• Implements custom DDL-based timeouts for DropTable and DropDatabase operations. #28394
• Splits client's FlushTables into two methods, CompactTables and FlushTables, for enhanced extensibility. #28467
• Allow vector indexes compaction with compact_table by using the ADD_VECTOR_INDEXES parameter. #27090
• Simplifies DocDB iteration by removing unnecessary split between IntentAwareIteratorIf and DocRowwiseIteratorBase for new feature implementation. #28475
• Allows toggling the colocated table tombstone cache optimization using the flag FLAGS_enable_colocated_table_tombstone_cache. #28470
• Adds CRC checksum to RPC responses to prevent data corruption. #28665
• Optimizes row scanning in HybridScanChoices by introducing ScanTarget class to eliminate the need for repeated key decoding. #28608
• Allows updating remote_bootstrap_rate_limit_bytes_per_sec flag in real-time for each file or session. #28600
• Adds vector index reverse mapping decoding support to the sst-dump tool. #28653
• Switch to clang built with PGO in llvm-installer version for faster C/C++ compile and link times, reducing overall build times by 9-11%. #28692
• Updates compaction logic to correctly handle and clean tombstoned reverse-mapping vector index keys. #27095
• Adds server-level metrics to track the number of transactions started as global, region-local, and tablespace-local. #28921
• Allows tablespace-local transactions to write to tables in any contained tablespace without global promotion. #28272
• Displays xCluster safe time on yb-master UI page for enhanced debugging. #28910
• Introduces two YSQL configuration parameters to control use of tablespace-local locality, offering workaround for suboptimal performance in certain cases. #28604
• Calculates region locality on the tserver side for improved region local transactions handling. #28828
• Allows reads and writes to proceed without waiting for a full quorum acknowledgment, reducing YSQL write and read latencies. #28320
• Introduces disk write statistics for vector compaction, which offers insights into IO across vector index libraries. #28842
• Configure RocksDB keys delta-encoding separately for YCQL and YSQL for faster read path and space efficiency. #28862
• Enables skip prefix locks by default for new installation clusters. #28894
• Enables vector index background compactions by default to control parallel execution and limit compactions per tserver. #29212
• Refactors code to address contention issues on Catalocustomeranager::mutex_ in larger clusters, enhancing stability and performance. #29186
• Enables dynamic sorting by columns in heap snapshot UI and reduces risk of database overload. #26693
• Adds BSON and BID libraries to cmake, making the DocumentDB extension part of the build process, except for aarch64 due to upstream restrictions. #26695
• Adds DocumentDB extension to non-aarch64/arm64 builds, with new flag ysql_enable_documentdb for protection. #26947
• Supports major YSQL upgrades with xCluster by relaxing checks, allowing soft delete and recreate of databases. #27532
• Adds a yb-admin command to check if nodes can be taken down without losing quorum. #28020
• Standardizes long-form flag names in documents to use double dashes for clarity. #28240
• Adds periodic cleanup for DDL replication tables to prevent unbounded growth. #19193
• Simplifies ObjectLockManagerImpl code for easier maintenance and future updates. #26931
• Ensures AuthOK messages are sent immediately to clients, simplifying future changes. #27825
• Allows users to specify SSL ciphers and minimum TLS version on webserver ports. #27375
• Renames "Network" wait_event_type to "RPCWait" for clarity. #26986

CDC

• Eliminates usage of pgschema_name in CreateTable pipeline and introduces a new field internal_table_type to replace direct reading of PG Schema Name during table creation. #12770
• Adds a yb_restart_time column to pg_replication_slots. #27764
• Prevents data loss during restarts by not sending checkpoints if tablet safe time exceeds slot restart time. #28014
• Adds instrumentation for RPCs sent to CDC Service, and new wait events enabling detection of bottlenecks in CDC by ASH. #23199
• Displays replica identity map in list_change_data_streams output. #28646
• Enhances error propagation from GetChanges to clients for more transparency. #28033
• Updates descriptions for CDC retention flags and their dependencies, and clarifies validator error messages. #28019

yugabyted

• Activates read committed isolation and cost-based optimization by default for newly started clusters. #29169
• Adds an environment variable to show accurate CPU details and corrects VCPU display in yugabyted-ui in Kubernetes deployments. #26624
• Upgrades yugabyted-ui TypeScript version to 5.0.4 for better package compatibility. #26958
• Enables yb_extension role to create access methods during extension setup. #26857
• Aligns the styling of assessment voyager page with the provided ficustomera design without disrupting schema migration page operation. #28214
• Updates the voyager analysis page in yugabyted according to the Ficustomera provided. #28220

Bug fixes

YSQL

• Reduces sporadic "duplicate key value violates unique constraint" errors in the Long-running System Test framework. #12449
• Allows preservation of read time history to prevent "Bad read time serial" error during COPY operations in middle-commit scenarios. #26365
• Allows prompt response to Postgres interrupts, enhancing handling of statement timeouts and cancel requests. #26284
• Exempts walsender processes managed by the connection manager from YSQL backend checks. #27072
• Allows setting YSQL configuration parameters to a whitespace empty value without causing error. #27106
• Reverts a previous change that caused performance issues, ensuring timely execution of statements by honoring the YSQL configuration parameter statement_timeout during index scans with specific predicates. #26284
• Fixes errors and crashes when creating temporary tables with ON COMMIT DROP. #27339
• Resolves GCC 13 release build failures on Linux. #27916
• Allows sharing of perform operation serial number counter among multiple Postgres processes to ensure unique serial numbers and better operation execution. #23648
• Allows users to preload pg_enum in ysql_catalog_preload_additional_table_list. #28757
• Prevents YSQL major upgrade failure caused by inconsistent namespace mapping during a failed upgrade attempt. #28815
• Allows yb-admin ysql_catalog_version command to return per-DB Catalog Version value for a provided database. #28690
• Adds missing primary keys and unique constraints to system catalogs, enabling them during initdb and YSQL upgrade. #27261
• Disables neg hit optimization for inherited cache when minimal preload is on. #27543
• Enhances upgrades for indexes with hash column groups to correctly handle aliases and metadata. #27629
• Allows group and others read permission on initdb log files by hardcoding the permission. #28435
• Ensures joins are only pruned when a Leading hint exists, avoiding errors. #26670
• Allows on-demand reading from non-prefetched tables by downgrading log level from DFATAL to WARNING, eliminating Snapshot too old and cache inconsistency issues. #26515
• Allows the use of PostgreSQL's snapshot management for foreign key triggers in YSQL, eliminating potential foreign key violations and enhancing data integrity. #26163
• Ensures TimeZone is not lowercased in connection manager for correct pgJDBC interpretation. #28537
• Keeps historical query diagnostics data persistent, even after a server restart or crash. #25471
• Allows placement of table replicas in any specified region or zone using * wildcard. #26671
• Reduces sequence-related errors when using ysql_conn_mgr_sequence_support_mode=pooled_with_currval_lastval. #27024
• Corrects issues with foreign keys referencing partitioned tables, preventing TupleDesc reference leak warnings and crashes during insert queries. #26772
• Corrects a queue ordering logic by using std::lower_bound to prevent tserver crashes. #27170
• Preserves column order during backup restore of inherited tables, matching DocDB and PG orders. #27324
• Fixes ANALYZE failures on wide tables by adjusting fetch size limits. #27202
• Ensures auto analyze runs after table rewrites like TRUNCATE. #26109
• Allows dumping of profiles and role mappings in ysql_dumpall. #25731
• Enhances session stickiness when acquiring session-scoped advisory locks. #27249
• Fixes DROP CASCADE to remove generated columns from DocDB. #27604
• Ensures consistency in secondary index scans during concurrent DELETE_INSERT operations. #17159
• Allows flushing or dropping of buffered operations before executing commands that affect the session state. #27673
• Resets user ID and security context between ANALYZE retries for stability. #27188
• Ensures DDL statements initiate DocDB transactions correctly to prevent errors. #27859
• Disables automatic retries for COPY FROM/TO commands. #28093
• Disables writes to role profile tables during YSQL upgrades to prevent errors. #28016
• Enables accurate tablespace dependency tracking for tables moved via ALTER TABLE SET TABLESPACE. #26995
• Avoids segmentation faults in ASH queries by initializing wait states to kIdle instead of kUnused. #28303
• Fixes a bug that caused yb-master to crash when executing the CREATE INDEX command outside of a transaction block with savepoint for DDL support enabled via TEST_ysql_yb_enable_ddl_savepoint_support. #28955
• Supports multiple analyze commits and concurrent DDL, ensuring low priority lock on pg_yb_catalog_version is retaken during subsequent transactions. #28731
• Prevents Postgres crash by avoiding invalidation messages during catalog refresh when prefetcher starts. #28490
• Forces loading of index expressions during index creation to prevent missed updates and potential inconsistencies. #28267
• Fixes the issue of transaction self-abort with rollback to savepoint, allowing users to rollback DDL operations without facing transaction expired/conflict error. #28956
• Disables query layer retry of ANALYZE command to prevent potential crashes caused by dirty states. #29025
• Eliminates an issue causing stuck rollback operations to sub-transactions with alter table. #28957
• Automatically sets default values for dependent flags when the YSQL Cost-Based Optimizer (CBO) is enabled or disabled. #29081
• Ensures successful backup restore of colocated databases with unique constraints by correctly preserving implicit tablegroup. #28781
• Disallows disabling DDL atomicity with YSQL configuration parameters if transactional DDL is enabled, preventing potential issues. #29358, #28284
• Adds multi-snapshot execution mode to yb_index_check operation, reducing risk of Snapshot too old error on larger indexes. #26283
• Allows large values to be inserted into temporary tables by enabling toast tables for temporary relations. #26007
• Prevents errors by detoasting values before insertion into DocDB tables. #26561
• Enhances fast-path transaction accuracy by adjusting read time settings. #26204
• Re-enables password redaction for specific SQL statements in YugabyteDB. #22421
• Restores the call to ScheduleCheckObjectIdAllocators inadvertently removed. #26902
• Fixes a use-after-free bug in ysql_dump by copying tablegroup_name. #26547
• Allows yb_binary_restore to be set by ybdbadmin for vector extension creation. #26941
• Enables postgres_fdw to use ybctid for identifying tuples, ensuring compatibility and functionality. #11419
• Ensures ybctid is correctly populated in RETURNING clauses for DML queries. #26951
• Ensures auto analyze service no longer persistently analyzes deleted tables that met analyze threshold before deletion. #27033
• Allows backfill reads to succeed by ignoring the isolation level when the default_transaction_isolation option is set to SERIALIZABLE on the cluster level. #26974
• Ensures YSQL statistics tables are saved and persist after a database restart. #26758
• Enables inheritance support by default, fixing issues with dropping constraints on inheritance parents. #26216, #5956
• Adds missing migration script for pg_hint_plan extension update. #27025
• Ensures accurate reset of stats before every EXPLAIN (ANALYZE) query. #27093
• Ensures metadata consistency in upgraded indexes with renamed columns during YSQL major upgrades. #26928
• Allows index column names to be emitted as aliases during a YSQL major version upgrade when the flag yb_format_funcs_include_yb_metadata is set to true. #26928
• Increases the default ysql_output_buffer_size to 1MiB to enhance query retries and reduce errors. #22245
• Fixes a bug to prevent NULL insertion in pg_yb_catalog_version. #27103
• Adds COMMIT as a DDL in YSQL upgrade scripts to handle cache invalidation correctly. #27148
• Fixed the issue causing rewrites on empty temporary tables to fail, allowing successful alteration of temp table columns. #26367
• Adjusts the unique_key_recheck to fetch ybctid from the correct slot, even though deferrable unique constraints aren't supported yet. #27163
• Ensures CREATE FUNCTION also updates the catalog version. #27166
• Fixes secondary index scan cost estimation on colocated tables. #27214
• Supports accurate cost estimates for hypothetical indexes in both old and new models. #22472, #22473
• Restores single-RPC optimization for certain index scans. #27294
• Fixes the assertion failure for ALTER DATABASE OWNER when using SELECT FOR UPDATE. #27280
• Enables successful execution of comma-separated ALTER commands, even when one requires DocDB schema changes and the other doesn't. #27263
• Fixes double output issue during a schema version mismatch retry by resetting the buffer position. #23825
• Fixes crashes when locking rows in temporary tables by setting the correct transaction state. #27149
• Ensures consistent query IDs for plan hints across settings and uses. #26621
• Removes duplicate YSQL configuration parameter entry. #27427
• Prevents ASAN failures during certain SQL operations. #27447
• Converts process-local variables to thread-local ones in lc_collate_is_c, eliminating potential data races. #27500
• Removes deprecated masters argument from ysql_dump/ysql_dumpall tools to reduce future merge conflicts. #26915
• Fixes TServer crashes during concurrent metadata changes and tablet destruction and a segfault when rewriting tables with vector indexes. #26879
• Fixes a memory leak in yb_cancel_transaction by freeing YbcStatus. #27554
• Prevents assertion failure by checking and skipping the sending of zero requests in PgGate's parallel requests. #27535
• Assigns accurate query IDs to RPCs executed outside of executor hooks. #24693
• Fixes a compilation issue by converting an assert to an error. #27338
• Prevent server crashes caused by large regular expressions with a new flag yb_max_recursion_depth. #27009
• Corrects EXPLAIN output estimates for Bitmap Index Scans, enhancing accuracy in performance analysis. #27602
• Ensures unique event codes in yb_wait_event_desc to improve data analysis accuracy. #27260
• Ensures accurate EXPLAIN output for forward scans with aggregate pushdown. #27642
• Fixes DROP COLUMN handling for tables with child partitions requiring rewrite. #27636
• Calculates scan bounds during PgDocReadOp initialization, stores them with the request, and eliminates unnecessary recalculations and requests to partitions without matches, thus saving resources. #27695
• Enables TOAST table creation for temporary tables in REFRESH MATERIALIZED VIEW CONCURRENTLY. #26007
• Disables yb_force_early_ddl_serialization by default to avoid conflict errors. #27761
• Fixes stale reads in pg_yb_invalidation_messages to prevent full catalog refreshes. #27776
• Ensures PostgreSQL shuts down before deleting upgrade data to prevent crashes. #27726
• Reduces unnecessary full catalog cache refreshes in certain scenarios. #27822
• Corrects the NoYbBatchedNL hint behavior by properly configuring yb_prefer_bnl flag and defining clear semantics for NL and BNL hints. #27728
• Prevents server crashes when multiple backend processes terminate while one holds a spinlock. #27874
• Fixes crash by initializing CacheMemoryContext during backend startup. #27864
• Fixes crash when using planner hints with hypothetical indexes. #27927
• Ensures accurate catalog read counts for the first query in a session. #28011
• Prevents crashes on transaction rollback with ALTER TABLE when transactional DDL is enabled. #28044, #28040, #28000
• Separates commands in database upgrades to prevent read restart errors. #26282
• Adds pgaudit 1.7.1 to enhance audit accuracy and resolve specific bugs. #28087
• Displays the path of the initdb log file in stdout, aiding in debugging when global initdb fails. #26316
• Enables dump_role_checks flag for new role-related statements in ysql_dump and ysql_dumpall. #27795
• Enables yb_db_admin roles to set PL speculative execution parameters. #28101
• Allows metadata to pass with AcquireObjectLocks and ReleaseObjectLocks RPCs for better thread communication. #28129
• Ensures yb_enable_cbo correctly activates when legacy YSQL configuration parameters are ON. #28156
• Update prevents auto-analyze from interrupting DDL statements by adjusting lock priorities. #28201
• Reverts recent changes to handle postmaster PANIC, improving stability during error scenarios. #27874
• Ensures the correct renew strategy for IsRenewRequired works as expected. #28226
• Adds options to enable pre-2024.1 BNL cost behavior in non-CBO mode. #28222
• Enables dump_role_checks in ysql_dumpall for role-related statements after creating profiles. #28235
• Ensures consistent row filtering and paging in PgsqlVectorFilter. #28411
• Fixes the issue where database fetch attempts could enter an endless loop, improving scan operations. #28462, #28476
• Reduces unnecessary catalog version increments for renames on temporary relations. #28477
• Fixes deletion of leftover invalidation messages when databases are dropped. #28348
• Adjusts CFLAGS to ignore pg_parquet build warning, resolving clean build failure on some MacOS builds. #28494
• Resets catalog read time after updating catalog version to prevent catalog inconsistency for concurrent DDL + DML under table locks. #28493
• Sets the ybIsHinted flag correctly for Projection paths to prevent cheaper paths from losing based on cost. #28072
• Enables transactional Data Definition Language (DDL) in the database, allowing DDL to run as part of the transaction block like Data Manipulation Language (DML), enhancing transaction consistency. #28000
• Prevents failure of ALTER TABLE operations on UNIQUE columns in colocated databases by excluding colocation ID emission. #28177
• Prevents a rare tserver crash by fixing a bug in YbNeedTupleRangeCheck and removing an unused variable from YbIsTupleInRange. #28708
• Deprecates the YSQL configuration parameter yb_make_next_ddl_statement_nonincrementing to enable incremental catalog cache refresh by default. #28765
• Resolves issues in TestPgRegressPlpgsql by enabling transactional DDL, reducing the chance of crashes due to many retries when transaction rollback is performed, especially with ALTER TABLE statements. #28000
• Fixes a bug with transactional DDL not considering the yb_force_catalog_update_on_next_ddl flag, preventing crashes during transaction rollback containing an ALTER TABLE statement. #28000
• Reduces RPCs by using Index Only Scan for INSERT ...ON CONFLICT ...DO NOTHING statements. #28324
• Adds logic to accurately merge catalog modification aspects in nested ddls when transactional ddl is enabled. #28000, #28791
• Introduces a new flag ysql_auth_method to select the password authentication method for YSQL, ensuring backward compatibility and avoiding user lockouts during upgrades. #28809
• Fixes a regression bug that caused an OBJECT_NOT_FOUND error during an index alteration, ensuring smoother table modifications. #28849
• Restores distinct error messages for kAbort and kConflict errors for clearer troubleshooting. #28157
• Global DDL operations won't be aborted by concurrent (auto)ANALYZE processes in separate databases. #28877
• Enables execution of DELETE and (re-)INSERT operations in a single flush for indexed JSONB columns, making updates less expensive. #27948
• Resets transaction read point for DML statements following a DDL execution in transactional DDL mode. #28871
• Fixes an error where a join condition was incorrectly pushed down in a batched nested loop, preventing crashes during planning. #28112
• Displays accurate Storage Flush Requests and Execution Time for non-buffered write operations in the EXPLAIN ANALYZE output. #26963
• Prevents dangling pointer issue in YbcFlushDebugContext by properly storing temporary strings. #28913
• Adds a preflight check to fail upgrades if user-defined roles with the yb_ prefix exist, except yb_superuser. #28977
• Prevents regular DDL within a transaction block from being aborted by auto analyze. #28926
• Enhances error message clarity when altering the primary key index owner. #29001
• Fixes error in REINDEX INDEX statement execution on partitioned tables with transactional DDL enabled. #29058
• Corrects index-only scans in unbatched INSERT...ON CONFLICT...DO NOTHING queries for multi-column indexes. #29109
• Resolves a bug that reset the query id of the walsender process to 7 from 11. #29147
• Increases default values for parallel_setup/tuple_cost to better balance YB scan node and PG node costs. #28939
• Refactors YbBindSearchArray's handling to eliminate potential bugs related to mutating is_column_bound and setting ybScan->all_ordinary_keys_bound, improving result accuracy. #29042
• Prevents crashes during scalar array operations with arrays of dimension greater than two. #29188
• Adds a backend type and libpq option to prevent analyze executions from hanging and ensure smooth CREATE INDEX operations. #29115
• Renames various terms for rows read during backfill to num_rows_read_from_table_for_backfill for clarity, and directly increments the backfill job field, eliminating unnecessary atomic field accumulation. #29052
• Updates reltuples for the new index and base table when creating an index, reducing the need to run ANALYZE and providing more accurate index scan cost estimates. #25394
• Enables yb_enable_update_reltuples_after_create_index when the Cost-Based Optimizer (CBO) is turned on using the ysql_yb_enable_cbo flag or yb_enable_cbo YSQL configuration parameter. #25394
• Enhances log message clarity for AlterTable CHECK failures in PgSchemaChecker to aid debugging in PITR restore scenarios. #28535
• Reduces transient memory spikes during long running write queries by fixing intra-query memory leaks. #29056, #29057
• Adjusts reltuples from 0 to -1 for dumps of older versions and enables ysql_upgrade_import_stats by default for improved database upgrade support. #27057
• Fixes crashes during ysqlsh connections with empty passwords under load. #28174
• Fixes connection manager crash during authentication due to double freeing of msg object. #29195
• Enables defining the maximum percentage of total YSQL connections for the connection manager. #27701
• Enhances visibility of ALTER ROLE ... SET configurations across multiple databases. #27945
• Fixes authentication errors during consecutive cluster connections. #28149
• Renames conflicting random number functions to prevent data race issues. #28248
• Avoids crashes and data races during shutdown by safely managing thread resources and memory clean-up. #28415
• Ensures accurate logging in PgAudit by using strdup to prevent garbage values. #27108
• Allows the Connection Manager to use the highest possible integer as a bit-mask value for custom YSQL configuration parameters' stickiness. #27374
• Prevents cascading timeouts during authentication lags in the connection manager. #28080
• Prevents crashes by ensuring thread-safe rule de-referencing during route cleanup. #28102
• Prevents crashes by holding route locks until the server closes in multi-route pooling. #27894
• Prevents indefinite stalls in multi-route pooling by ensuring signals are broadcast across all routes. #28283
• Enhances stability for YSQL connection manager during COPY operations from Go drivers. #28148

DocDB

• Ensures consistent data reads by setting propagated safe time only after successful operation replication. #23696
• Extends cloning timeout for multi-region deployments to prevent premature timeouts. #26626
• Ensures yb-admin commands respect user-specified timeouts for table flushes and compactions. #19957
• Prevents deadlocks and unresponsiveness during high-load schema changes by fetching table info asynchronously. #26909
• Reduces follower lag by ensuring consistent Raft config during tablet splits. #26644
• Allows successful response to duplicate write requests to prevent unnecessary query failures. #7251
• Ensures accurate replication for transactional writes on xCluster range partitioned tables. #27380
• Ensures post-split compactions complete even if background compactions are in progress. #27426
• Eliminates duplicate MasterService metrics for improved monitoring clarity. #25277
• Prevents incorrect query results due to invalid seek keys in internal iterator operations. #27641
• Enhances the stability of document reads by validating key prefixes before processing. #27923
• Prevents thread metric data from disappearing over time. #27955
• Fixes unnecessary error logs during transaction loading with CDC enabled. #27986
• Enhances Kubernetes compatibility by using RPC bind address as hostname in masters. #28084
• Ensures tablet server limits in the UI exclude dead or blacklisted nodes, providing accurate data metrics. #27512
• Prevents potential deadlocks in Raft operations by managing flusher state effectively. #27531
• Allows SST files with specific row or column TTLs to remain candidates for compaction. #26014
• Increases reliability of database cloning after a DROP TABLE operation. #28285
• Solves the reversed condition in the rpc_inbound_calls_failed metric, offering more metric information. #28788
• Ensures tablet split handling on the consumer side does not cause incorrect mappings due to race conditions. #28750
• Allows removal of associated intentsdb keys during tablet startup to prevent potential pileup and slower queries. #21831
• Changes prevent unnecessary node crashes due to network issues during remote bootstrapping process cleanup. #28830
• Prevents silent data loss by failing the clone when a table is altering instead of cloning without data. #27621
• Disables TryConsume usage by LogReader to prevent user operations from being incorrectly blocked due to a bug. #29095
• Handles thread creation failure properly and falls back to RPC mechanism when thread cannot start. #28968
• Allows transaction table deletion by updating transaction versions only on live tablet servers. #29067
• Adds extra logging to track data inconsistency when durable_wal_write is set to false. #23453
• Avoids TServer restarts and disruption to existing connections by enabling YSQL operation lease mechanism at TServer start. #27343
• Ensures better handling of schema versions during rapid DDL changes in xCluster replication. #24990, #27275
• Allows namespace_id to be correctly copied in TableInfo constructors to prevent empty return values. #28478
• Prevents unbounded growth of the recently_applied_map by not adding read-only transactions, conserving memory. #26666
• Simplifies replication role management in xCluster by storing the role in the system catalog, enhancing the accuracy of DDL replication by updating the role with each DDL call. #25853
• Fixes handling of db_max_flushing_bytes to properly limit memory usage under high write loads. #26916
• Clears relevant object locks when a PostgreSQL session abruptly ends. #26673
• Fixes crashes when executing SELECT * FROM pg_locks during advisory lock waits. #26972
• Prevents data loss by ensuring databases are flushed before updating flush markers. #26910
• Stops sequences_data table bootstrapping during xCluster switchover to prevent sequence bumps loss. #26957
• Rolls back the DocDB column ID counter when a DDL fails in YSQL databases using automatic mode replication, preventing DocDB column ID divergence. #25787
• Fixes a bug that caused vector indexes to share a folder, preventing crashes. #27049
• Fixes HTTP 503 response formatting on Master API requests. #26914
• Ensures long-running reads access the correct schema version after updates and compactions. #23427
• Lets tservers include their lease epoch in YSQL lease refresh requests to ensure proper bootstrap and restart of PG sessions. #27152
• Changes the class of autoflag enable_export_snapshot_using_relfilenode from kLocalVolatile to kExternal to prevent restore failures during backups. #27155
• Ensures successful remote bootstrap by removing pre-existing directories. #27123
• Adds a new flag xcluster_ddl_queue_statement_timeout_ms to control DDL execution timeout from ddl_queue, with default set to no timeout. #27066
• Fixes crashes in handling session advisory lock statuses. #27030
• Adjusts DocRowwiseIterator to correctly include nulls for NULLS FIRST sorting. #27031
• Stops retrying failed batches when the session is closed. #27171
• Limits the number of entries per trace to prevent memory bloat during long running operations. #27334
• Fixes issue where TServers fail to establish lease with a new master-leader after network partition. #27460, #27479
• Fixes compaction errors from schema version mismatches by storing missing-value info. #27463
• Prevents tasks from being scheduled on the threadpool after setting shutdown flags, avoiding unexpected termination. #27557
• Prevents segmentation faults by ensuring return after executing RPC callbacks with validation failure in TabletService's object lock/release functions. #27561
• Prevents reuse of docdb transactions post-abort to avoid state conflicts in locks. #27559
• Prevents TServer crashes during concurrent vector index and read requests. #27699
• Reduces memory usage by optimizing pg_locks with specific iterator settings and intent limits. #27810
• Ensures safety by updating TabletPeer::GetNamespaceId to use shared_tablet_safe. #27632
• Resolves deadlock during TSHeartbeat by adjusting lock scope. #27773
• Eliminates potential deadlocks in thread pool by fixing a race condition. #27898
• Ensures OID counters in cloned or restored databases avoid conflicts after table drops. #27943
• Stops re-creating the xCluster DDL replication extension on the target. #27928
• Ensures OID counters advance adequately to avoid future table creation collisions. #27944
• Resolves PITR restore deadlocks by fixing lock inversion issues. #27792
• Eliminates redundant rocksdb_block_cache_bytes_read metrics to reduce noise and overhead. #27017
• Enables re-acquisition of object locks during statement retries in extended query mode execution. #24877
• Ensures all fastpath object locks appear in pg_locks output. #28163
• Disables catalog version check for ReleaseObjectLocks API to prevent TServer crashes. #28138
• Increases YSQL DDL RPC timeout to 180s and introduces configurable ysql_ddl_rpc_timeout_sec flag. #28168
• Prevents potential crashes during tablet shutdown when vector index compaction is running. #28511
• Prevents fatal errors in xCluster by ensuring all schema versions are present during replication. #28326
• Reduces master Write-Ahead Log (WAL) data by excluding db_catalog invalidation messages from initial ReleaseObjectLockRequestPB persistence. #28573
• Fixes exclusive lock intents acquisition before shared memory setup. #28595
• Adds a session-level advisory lock for ddl_queue_handler to prevent multiple handlers running concurrently. #27186
• Allows Parallel Query feature to work correctly with object locking by limiting object lock and finish transaction calls to the leader worker. #28878
• Adds a new enable_object_locking_infra flag and YSQL configuration parameter to activate table-level locks only after upgrading all nodes. Both this flag and enable_object_locking_for_table_locks must be true to activate the feature. This flag is tagged as PGC_INTERNAL preventing users from toggling it. #28875
• Fixes PgConnTest.SessionExpiration failure when table locking is enabled. #29010
• Reduces pg_locks transient errors by verifying tablet leaders during lookup and increasing retry attempts. #28180
• Fixes a TServer crash issue during hybrid scans caused by incorrect read requests from Postgres. #29347
• Fixes the issue of recording "query id 0" in Active Session History samples. #26808
• Improves error handling for shared memory operations in DocDB. #26740
• Removes 60-second timeout upper bound on admin RPCs and adds new yb_client_admin_rpc_timeout_sec flag. #26722
• Prevents deadlocks during background compaction and transaction loading. #26880
• Prevents yb-admin crashes by correctly handling argument count for create_database_snapshot. #26899
• Enables yb-pbc-dump to decrypt and process encrypted protobuf files using specified keys. #26917
• Prevents crashes in GCC fastdebug builds when vlogs are enabled by adjusting lambda storage in ObjectLockManager. #26960
• Prevents master process crashes by fixing an iteration modification issue in TriggerDdlVerificationIfNeeded. #26967
• Blocks nonconcurrent index creation on xCluster replicated tables. #27006
• Ensures pg_locks query reliability when a node is down by continuing to serve requests. #27035
• Eliminates the risk of deadlocks during table alterations and snapshot operations. #26565
• Adds a lock-free SemiFairQueue class to the rpc::ThreadPool to manage tasks execution order. #27121
• Enables efficient memory usage and fast data access through dynamic loading and unloading of data blocks in YbHnsw. #27075
• Adjusts primary key creation on collection tables for better distribution of records, fixes bug in make_expanded_record_from_datum, and switches ysql_enable_documentdb to a preview flag. #27082, #27058
• Adjusts DDL replication tables to be non-colocated directly in the extension using SQL for better compatibility with Postgres. #27100
• Ensures column ID continuity in backups by preserving next_column_id. #27064
• Enables YSQL transaction tracing with pg_client_use_shared_memory enabled. #27392
• Ensures no deadlock when threads wait on PgResponseCache for query results. #27412
• Prevents duplicate vector insertion after TServer restart during backfill. #27457
• Allows unlimited threads in the catalog manager async_task_pool to prevent xCluster failures. #26617
• Allows retrieval of table schema with vector column via java/yb-client by adding VECTOR type. #26384
• Ensures consistent backup by exporting only committed DocDB tables at snapshot hybrid time. #27449, #27590
• Eliminates undefined behavior by correcting the use of moved variables in pg_client_session.cc. #27707
• Allows skipping object locking during initdb to prevent failures. #27444
• Adds a wait event for transactions with SERIALIZABLE READ ONLY DEFERRABLE. #27836
• Enhances metrics to better estimate cluster balancing times and validate the accuracy of the balancing algorithm. #23908
• Ensures TServer handles lock releases correctly after bootstrapping. #27735
• Prevents master crashes during index backfill by handling empty indexes correctly. #27697
• Fixes placement logic to support wildcards in cloud info for tablet selection. #27285
• Ensures xCluster metrics display correct table IDs and namespaces for better monitoring accuracy. #27667
• Restores deletion of table info objects to prevent memory overload. #23721
• Disables object locking when ysql_enable_db_catalog_version_mode is false. #27872
• Prevents deadlocks by setting status tablet ID for shared memory lock requests. #27921
• Eliminates DFATAL error for unregistered lock owners handling exclusive locks. #27968
• Fail restore for unmatched DocDB tables during import_snapshot phase. #27820
• Fixes a bug to ensure correct data storage in PostgreSQL table cache requests. #28001
• Disables unnecessary logging when FLAGS_txn_slow_op_threshold_ms is set to 0. #27996
• Object locks now respect yb_locks_min_txn_age and yb_locks_max_transactions settings. #27851
• Resolves transaction status resolution hang during shutdown by using a CoordinatorLock and initiating shutdown for Rpcs before resolver. #28030
• Ensures replication of transaction application status even when the last pending tablet is deleted. #28059
• Defers transaction priority assignment to the first read/write operation. #28125
• Ensures TSLocalLockManager is only created when both enable_ysql and enable_object_locking_for_table_locks are enabled. #28126
• Enables ALTER DEFAULT PRIVILEGES commands in xCluster's automatic mode. #27051
• Reduces noise in compiler output for x86 OS X builds by fixing regular expression errors. #28292
• Ensures master metadata snapshots are consistent with snapshot_hybrid_time for accurate backups. #27575
• Fixes issues with aborted transactions during promotions preventing stuck operations. #27853
• Ensures cloning databases functions correctly with table locks enabled. #27800
• Enables system tables to always fetch fresh tombstone time, preventing stale cache issues. #28327
• Prevents an infinite loop and memory overhead in RocksDBOptionsParser when reading OPTION files. #28343
• Stops XCluster polling when excessive DDL errors occur, preventing retries beyond limits. #28363
• Adds CRC checksum to RPC calls to detect data corruption. #18675
• Restores previous deadlock detection by retrying lock requests from the host tserver. #28386
• Enforces TLS v1.2 for secure webserver connections. #28417
• Reduces table tombstone check time significantly by adding an iterator upperbound, improving performance. #28451
• Preserves cluster balancer metrics correctly across all placements including read replicas. #28366
• Prevents table deletion during snapshot creation by implementing namespace anchoring. #28290
• Increases data metric's capacity to handle over 4 GB for balancing. #28500
• Resolves two compile errors in GCC 12 build by adjusting parameters to be compatible with GCC and suppressing warnings. #28519
• Displays "(unenforced)" in master UI under "Tablet peer limit" when tablet limit enforcement is off. #28479
• Fixes an issue where aborted transactions could get stuck with object locks. #28520
• Removes dangling references in certain code areas identified by the -Wdangling-references warning. #28560
• Eliminates an extra space before the period in the output of yb-admin flush_tablet/compact_table command. #28572
• Fixed a deadlock issue in the pg client service shutdown path, improving system stability. #28597
• Reduces chances of hitting kDeadlock for first statement in RC transactions and addresses rollback issue for session advisory lock requests. #28593
• Adds a memory tracker to monitor temporary memory used when reading WALs, preventing TServer crashes. #28623
• Removes a function call that caused tablet server shutdowns during RPC reception. #28615
• Adds time units to relevant Prometheus metric help descriptions for better understanding. #28634
• Resolves the issue of queue_time_us_stats being incremented twice, enhancing TestThreadPool.TestMetrics stability. #28668
• Adds a long operation tracker to CatalocustomeranagerBgTasks::RunOnceAsLeader for easier debugging. #28712
• Adjusts active log segment size tracking for accurate wal_size metric when reusing existing segments. #28720
• Prevents yb-master crash by making MessageDifferencer thread-safe and limiting its use to debug builds. #28729
• Allows deletion of hidden tablets during point-in-time restores that are no longer in use. #28704
• Allows AutoAnalyze to work consistently, regardless of table level locks feature, by always acquiring the acquire_for_update_lock lock. #28783
• Adjusts advisory lock release timing to prevent handler blockages during shutdown in time-sensitive builds. #28889
• Redundant finish transaction calls no longer request a new transaction when releasing object locks if a DocDB transaction doesn't exist. #28761
• Allows table-level locks to respect the lock_timeout setting, returning an error if lock is not obtained in time. #27677
• Reduces master_rpc_timeout_ms to 1.5s to lessen outages by quickly detecting dead/unavailable master nodes. #28924
• Enables "SELECT FOR KEY SHARE" to avoid crashes during cluster restart by starting each run with a clean state. #28881
• Reduces Unable to find schema name for YSQL table errors and optimizes the retrieval of YSQL Table Oid. #28793, #27590
• Eliminates shutdown deadlock in CQL service, enabling consistent and graceful server shutdown. #28438
• Modifies fast path transaction criteria to exclude object locking writes. #28922
• Enables transactions to start with the correct locality in AcquireObjectLocks RPC handler, reducing latency and eliminating extra global roundtrips for global transactions. Also passes force_global_transaction and tablespace OID for AcquireObjectLocks RPCs, allowing transactions to start under the appropriate locality. #28317, #28979
• Adds object lock auto-flag checks to YbMaybeLockMasterCatalogVersion and AcceptInvalidationMessages for upgrade safety. #29054
• Ensures that failed ALTER TABLE statements roll back even in the event of a master crash. #27021
• Prevents vector index compaction failure when all vectors are filtered out during compaction. #29016
• Eliminates unexpected system exits when DDL transactions enter the tablet wait-queue in a colocated database with index. #29215
• Adds debug logs to assist in debugging issues related to concurrent DDL execution without object locking. #29080
• Enables use of PostgreSQL memory allocation for BSON by updating to the latest version of the BSON library. #27104
• Corrects xCluster status display for sequence data streams. #26388
• Disables pg_cron jobs on xCluster target by default, adjustable with cron.yb_enable_on_xcluster_target. #26799
• Enables extension build in Ubuntu 22.04-Clang19-debug by using pcre from the latest third-party update. #26948
• Fixes a memory leak by using condition variables for signal handling in the connection manager. #28191
• Reduces segmentation faults by isolating cascading timeout handling to auth processes. #28206, #28218, #28231, #28242
• Eliminates stale hashmap entries on parse errors, enhancing protocol-level prepared statement handling. #28576

CDC

• Ensures CDC correctly processes transactions involving vector columns by skipping orphan tombstone intents. #27942
• Prevent loss of records due to bugs in the PeerMessageQueue::ReadReplicatedMessagesForConsistentCDC implementation. #29060
• Ensures only relevant COMMIT records are streamed, fixing gRPC connector crashes. #26861
• Prevents CDC crashes by logging a warning for dropped indexes on colocated tables. #26834
• Prevents data loss by not streaming records during transaction load. #26835
• Ensures reliable CDC stream functionality during index creation, preventing schema packing errors. #26891
• Fixes issue where data streaming with colocation could skip intents, causing record or column loss. #27052
• Prevents deletion of child tablet entries during metadata cleanup. #27069
• Sets the default value of dcheck_for_missing_schema_packing to true to enhance CDC testing accuracy. #26891
• Ensures hidden tablets are accessible during CDC stream fetches to prevent errors. #26911
• Optimizes CDC streams by sending a single BEGIN and COMMIT record for colocated tables. #27248, #2605
• Ensures CDC streams data for non-transactional YCQL tables correctly. #28034
• Ensures unique sorting of batched shard records using write_id to prevent data loss. #28166
• Eliminates race conditions in CDC session management, preventing crashes. #28336
• Introduces the yb_ignore_read_time_in_walsender parameter for logical replication continuity during upgrades. #27177
• Ensures only tables in a publication are checked for CHANGE replica identity. #28629
• Handles error propagation from CDC to client for missing footer in segments. #28592
• Allows large transactions for unqualified tables to stream correctly, preventing the virtual WAL from getting stuck. #28744
• Fixes CDC issue with virtual WAL restart after a DDL, preventing record decoding failure. #29136
• Prevents core dumps by returning early from WalSndWakeup due to irrelevant callers in the YugabyteDB context. #28379
• Filters out data from rolled-back savepoints in CDC streaming, ensuring only valid data is received by CDC consumers. #22817
• Prevents the "Trying to fetch already GCed intents" error by correctly updating and managing checkpoints for child tablets after a tablet split. #28840
• Changes the enable_cdcsdk_setting_get_changes_response_byte_limit flag to default as true, ensuring the correct maximum response size. #27892
• Stops applying the LogReader memory limit to CDC, a setting intended only for xCluster. #29066
• Adds "Log index cache entry for op index not found" error to retryable errors in cdcsdk_virtual_wal. #29061
• Reverses changes to support savepoints due to an exposed bug, specifically for version 2025.2.0. #29404
• Reduces the frequency of log spam due to missing tablet warnings by changing logging parameters. #28086

Other

• Updates the Knowledge base -LlamaIndex with the latest YugabyteDB Vector Index dependency details. #27913