YMatrix 6 Release
6.8.4
Date: 2026-10-08
Enhancements and Improvements
- The optimizer introduces SAOP rewriting and enables it by default: scalar array conditions such as
IN lists are automatically rewritten and merged to reduce repeated computation. If problems occur, disable it via mx_enable_saop_rewriter.
- Shared CTE is now enabled by default (
gp_cte_sharing), so no explicit setup is needed.
- Shared CTEs can generate parallel plans when parallel mode is enabled (
enable_parallel_mode). For troubleshooting, enable mx_debug_cte_replan to output replanning debug information, or use mx_cte_force_parallel to skip cost checks and prefer parallel plans (effective only for queries that can run in parallel).
- When a shared CTE contains only a simple table scan, it no longer uses shared materialization even if
gp_cte_sharing is enabled; each reference scans the base table directly to avoid extra overhead.
- For subqueries that scan replicated tables, the execution plan no longer requires data broadcasting when the conditions are met (for example, no volatile functions and no LIMIT).
- For window functions whose results change monotonically with the row number (such as
row_number()), early termination and two-phase execution are supported to reduce unnecessary intermediate data.
- Optimized equivalence class construction for multi-table joins.
- Optimized plan generation and cost estimation when
LIMIT + DISTINCT uses HashAgg.
- MARS3 adds pattern operator classes for the
text, varchar, and char types. LIKE prefix queries can use indexes, and ON CONFLICT can explicitly specify the corresponding operator classes. Also fixed sorting issues during index build and min/max statistics for DESC indexes. See MARS3 Indexes.
- Optimized resource usage of MARS3 sequential scans, reducing initialization overhead and memory usage for wide-table queries.
- MARS3 can output timing statistics for each execution phase via
EXPLAIN ANALYZE VERBOSE to help locate performance bottlenecks.
- Improved the efficiency of MVCC visibility checks in MARS3.
- mxnumeric now derives precision at the planning stage to improve query performance.
- When expression evaluation or type conversion fails, error messages now include column name context; character and bit-string length overflows and smallint conversion overflows report the specific offending values to make problems easier to locate.
- Using native TID Scan on a MARS3 table now reports an error clearly stating that this scan method is not supported.
- MatrixGate Transfer mode maps columns by name automatically during single-table migration, so the source and target tables do not need to keep the same column order. See Single-Table Migration Tool mxgate.
- MatrixGate no longer parses raw Kafka JSON data repeatedly when scanning it, improving processing performance. A Kafka consumer group can now configure multiple members. See MatrixGate Parameters.
- MatrixGate supports resetting Kafka consumer offsets, and the MXUI data stream page can reset offsets as well. See Resetting Kafka Consumer Offsets.
- MatrixGate and MXUI data streams support the
DisablePAFXFAST parameter. See Kafka SSL Encrypted Connection.
- MXUI adds a disaster recovery cluster status display.
- MXUI data streams support creating non-partitioned tables: no time column is required when creating a table, and the device ID column becomes optional in INSERT mode. Automatically mapped field names are also simplified. See Kafka Access.
- MXUI data streams support saving Kafka profile information for quick reuse later. See Kafka Access.
- MXUI data stream tasks support start and stop. See Managing Kafka Data Streams.
- MXUI hides unsupported action buttons in data stream scenarios.
- The MXUI resource group module automatically detects the resource group version supported by cluster nodes. See Resource Group Management.
- MXUI auto-tuning card adjustment: an optimization can now be skipped only via the Skip button.
- MXUI improves the HAProxy check mechanism so that License operations on a Standby MXUI are correctly forwarded to the Master node.
- When MXTool sets the kernel parameter
vm.min_free_kbytes, it now calculates 5% of total machine memory instead of using a fixed value.
Bug Fixes
- Fixed several known issues
System Variables and Option Changes
| System Variable / Option |
Change Type |
Description |
mx_enable_saop_rewriter |
Added |
Default on
Controls SAOP automatic rewriting. When set to off, scalar array conditions such as IN lists are no longer rewritten. |
gp_cte_sharing |
Changed default value |
Default value changed to on; shared CTE is enabled by default. |
mx_cte_force_parallel |
Added |
Off by default
Skip cost checks and prefer parallel plans for shared CTEs. Effective only for queries that can run in parallel. |
mx_debug_cte_replan |
Added |
Off by default
Print debug information for shared CTE replanning. |
6.8.2
Date: 2026-07-03
Enhancements and Improvements
- Dispatcher optimized the internal connection setup process, reducing unnecessary latency introduced by SSL and GSS checks .
- mxv optimized execution performance for scenarios with many repeated expressions .
- mxv typed encoders now support arbitrary types .
- MARS3 DefaultBrin now supports DynamicSeqScan .
- MARS3 optimized sequential output and cost estimation for MARS3 BRIN and MARS3 btree, and adjusted the estimation model to use stripe/range as the unit.
- MARS3 optimized RangePage and RangeDescPage access paths by referencing RangePage and locally caching RangeDescPage handles to reduce execution overhead.
- Planner now allows replicated tables to participate in parallel plan evaluation by default, so other tables can still generate parallel plans such as parallel hash join when replicated tables exist.
- Planner supports early deduplication of SetOp subplans based on streaming aggregation and supports generating parallel UNION plans .
- Planner optimized Motion generation for UNION, INTERSECT, and EXCEPT, adding Motion nodes only when necessary .
- Planner supports pushing OR conditions in HAVING clauses down to scan nodes for earlier filtering .
- Planner supports adding Motion nodes under Append children inside shared CTEs to support parallel execution .
- Planner supports merging CTE reference filter conditions and pushing them down to shared scans as OR combinations .
- Planner supports converting pushed-down OR conditions to ANY expressions during shared CTE replanning to improve row-count estimation .
- Planner supports pulling up EXISTS/NOT EXISTS sublinks under OR conditions and executing them as left outer joins .
- Planner supports pulling up expression sublinks of type EXPR_SUBLINK as JOIN or left outer join execution .
- Domino stream processing supports executing join and agg in the same stream .
- Domino stream processing removes redundant columns in dual-stream left join. This change is incompatible with existing left join stream tables; related stream tables must be dropped and rebuilt after upgrade .
- PXF and pxf_fdw now support HadoopCatalog for Iceberg v2 tables.
- MXUI added a Pre-Install step that supports entering IP addresses on the Add Node page, matching hostnames, and synchronizing the hosts file .
- MXUI Pre-Install supports configuring sudo privileges for the
mxadmin user, disabling SELinux on hosts, and checking unexpected iptables rules .
- MXUI installation pages now display detailed errors during the node-add stage to make failures easier to diagnose .
- MXUI data table pages now show index information, including index name, type, and size .
- MXUI inspection items now check unexpected
postgresql.auto.conf configurations and unexpected iptables rules .
- MXUI adjusted mail server configuration validation. When AUTH is not enabled, the username is no longer validated .
- When adding mirrors in MXUI, if multiple disks are available, MXUI prompts users to modify the default plan .
- MatrixGate data ingestion now supports Kafka SSL encrypted connections. See MatrixGate Kafka Encrypted Connection (SSL) Configuration .
- MXUI data streams now support Kafka SSL authentication configuration .
- MXCtl adjusted the
kafka_catalog table structure. When creating the table, the id column type changes from int to bigint .
- MXStart supports configuring a wrapper, providing custom extension capability for the postmaster startup process while keeping the original startup workflow unchanged .
- Etcd was upgraded to v3.5.32 to address known CVEs .
- Disaster recovery / high availability now supports switching back and forth between primary and backup clusters .
- Disaster recovery / high availability adopts a new DR design and supports primary-mirror deployment, switch, recovery, and rebuild on the backup cluster. See YMatrix 6.8 Disaster Recovery User Guide .
- MXTool added OOM site collection capability .
- MXTool added the
mxtool init numa-wrapper command, which automatically generates NUMA Wrapper scripts and NUMA mapping configuration based on the current cluster topology and deploys them to each node. See Start a YMatrix Cluster with NUMA Node Binding .
Bug Fixes
- Fixed several known issues
System Variables and Option Changes
| System Variable / Option |
Change Type |
Description |
compresstype=auto |
Added |
MARS3 added the auto compression option, which uses the typed encoder (arraycodec). |
MARS3 btree auto compression |
Default behavior changed |
MARS3 btree enables auto compression by default. |
| rowstore TOAST compression |
Default behavior changed |
MARS3 disables rowstore TOAST compression by default. |
mars3.enable_btree_inorderscan |
Added |
Added the MARS3 btree in-order scan control GUC. The default value is off. |
6.8.1
Date: 2026-05-25
Enhancements and Improvements
- mxv: Optimized plan generation time for
DELETE WHERE IN queries with large lists.
- fastdd: Optimized external parameter pushdown to improve query execution efficiency in related scenarios.
- Optimizer: Improved execution efficiency for scenarios using
EXISTS with SubPlan.
- mars3: Added support for DIMS syntax. See MARS3 DIMS Dimension Column Declaration.
- mars3: Added support for the diskquota plugin.
- mars3: Added default brin information to
EXPLAIN VERBOSE output.
- pxf_fdw: Reorganized the code as a git submodule for easier independent maintenance and version management.
- Domino: Optimized the XLog retention callback performance when
VACUUM FREEZE is executed frequently.
- Matrixgate: Optimized goroutine usage in the internal performance statistics component
load_metric, reducing runtime resource overhead.
- Matrixgate: Introduced
automaxprocs; when GOMAXPROCS is not explicitly configured, Matrixgate automatically sets an appropriate value to improve CPU resource utilization.
- Matrixgate: Enhanced logs by adding table names to
set slot logs, making troubleshooting and operations easier.
- Matrixgate: Optimized the internal fasthttp
flush chunk logic to improve HTTP transfer efficiency.
- Matrixgate: Optimized memory usage by using fixed-size slices as lower-level buffers, reducing memory allocation overhead and fluctuation.
- Matrixgate: Added
http-segmented-body to control whether memory optimization is enabled. It is enabled by default. Set it to false to revert to the legacy Matrixgate memory allocation behavior.
- Matrixgate: Added
http-segment-size-kb to specify the size of each lower-level slice. The default value is 1024KB.
- Matrixgate: Added
write-buffer-size to specify the size of each buffer used for data transfer from Matrixgate to segments.
Bug Fixes
- Fixed several known issues
6.8.0
Date: 2026-04-10
New Features
Enhancements and Improvements
- Enabled work file compression in the vectorized executor by default to improve OLAP performance .
- Optimized the performance of
VHashAgg in the vectorized executor .
- Optimized the performance of
CASE WHEN expressions in the vectorized executor .
- Added support for right anti semi join (NOT IN) in the planner.
- The planner now supports parallel execution of right semi joins in certain scenarios.
- Optimized
HashAgg evaluation for distinct queries with LIMIT clauses in the planner .
- The MARS3 storage engine history snapshots now support the
definitelydead optimization.
- Optimized high memory usage in the MARS3 storage engine for Compact and Scanner operations when operating in UniquMode .
- Enhanced Domino stream computing to correctly retain snapshot files on nodes that have only decoders but no active streams .
VHashAgg now utilizes a single-partition HashMap .- In
mxshift, DDL operations no longer migrate temporary sequences .
- Upgraded the
mxnumeric extension to version 0.3, improving the performance and correctness of type conversions. See mxnumeric User Guide .
- In
matrixts, timebucket_auto now supports automatic sorting by time buckets upon return.
- MatrixUI now supports creating stream table templates.
- Disabled the pprof port in MatrixUI by default; it can be enabled via the web interface .
- SOA service components no longer expose the pprof address by default. It can be temporarily enabled or disabled by sending a signal using
sudo kill -USR2 -p <pid_of_soa_program>.
Bug Fixes
- Fixed several known issues
System Variable Changes
| System Variable |
Modification Type |
Description |
gp_workfile_compression |
Default Value Changed |
Default value changed from off to on. |
gp_interconnect_type |
Default Value Changed |
Default value changed from udpifc to tunnel. |
enable_dynamic_seqscan |
New Variable |
Default value: on.The planner optimizer now supports DynamicSeqScan for scanning partitioned tables, enabling efficient scanning without requiring ORCA. |
mx_enable_motion_runtime_filter |
New Variable |
Default value: on.When enabled, VMotion receivers filter rows that do not meet conditions using Bloom Filters immediately after receiving row data, reducing the volume of data passed to upper-level nodes. |
6.7.1
Date: 2026-02-24
Enhancements and Improvements
- License: Increased the default license validity period from 14 days to 30 days.
- MatrixUI: Added a new health monitoring rule that places the database in read-only mode when disk usage exceeds a configurable threshold. Disabled by default. (Documentation)
- MatrixUI: Added post-installation automatic tuning.
- MatrixUI: Added support for resource group management via the UI.
- Interconnect: Introduced a new interconnect type
tunnel. (Documentation)
- Vectorization: Simplified and enhanced duplicate elimination in vectorized
vtle.
- Vectorization: Added runtime filter support to
VDynamicSeqScan.
- Domino v2: Added support for segment set functionality.
- Query Rewriting: Introduced equivalence class merging during
RULE-based query rewriting.
- Vectorization: Added support for
FULL OUTER JOIN in vhashjoin.
- Dynamic Partition Elimination (DEP): The partition selector node now fully supports parallel execution.
- PXF: Expanded the set of pushdown-capable data types.
- PXF: Added the
clean_nul configuration option to control handling of 0x00 bytes:
- In
replace mode, 0x00 is replaced using the REPLACE() function during SELECT queries.
- In
filter mode, 0x00 is removed during column data parsing by PXF.
- Integration: Integrated PXF/FDW into
matrixdb/gpcontrib/pxf_fdw.
- Error Reporting: Enhanced error messages to include column names and actual values in context.
Bug Fixes
- Fixed several known issues
6.7.0
Date: 2025-12-31
Enhancements and Improvements
- Added pg_audit plugin functionality (Documentation)
- Added postgres_anonymizer plugin functionality (Documentation)
- Enhanced memory usage debugging mechanism
- Refactored VSort to use a Push & Pop model and upgraded to Sorter v2
- Optimized VHashJoin performance when generating results from inner tables with many payload columns
- Added support for expanding a single partition
- mars3 now terminates blocking compact tasks when dropping or truncating a table
- Relaxed column order restrictions in unique indexes for stream joins in Domino, and added support for equivalence relations
- Domino now allows referencing expressions from the right table in the target list of a left stream join
- Domino now supports non-equi joins in dual-stream joins
- MXUI added a toggle to control mounting of the PPROF route
- MXUI removed the error log card from the Dashboard page and replaced it with a health monitoring card
- MXUI enhanced the cluster information card on the Dashboard page to display more detailed cluster metrics
- The optimizer/executor now supports Right Anti Join and Right Semi Join (not yet compatible with Parallel Mode)
- pg_rewind now supports incremental repair for mars3 tables
Bug Fixes
- Fixed several known issues
6.6.1
Date: 2025-11-17
Enhancements and Improvements
- Added cleanup logic for
trigger function during partition split, along with a warning prompting manual cleanup if an error occurs.
- Introduced a new transaction visibility check optimization in mars3.
- Enhanced scalar sequential scans to include
RowRemoveFilter information.
- Improved vectorized sort performance in the vectorized execution engine when handling a large number of by-reference payload columns .
- Added automatic creation of the
dblink extension before Cylinder executes APM tasks.
Bug Fixes
- Fixed several known issues
6.6.0
Date: 2025-11-10
Enhancements and Improvements
- Added support for
default-brin in mars3. (Doc)
- Optimized XLOG size for mars3 in single mode.
- Enabled concurrent
ALTER TABLE SPLIT PARTITION in mars3. Concurrent reads and writes are supported during split (currently unsupported for AO/CO tables, stream tables, or parent tables with triggers).
- Introduced the
mxchangeport utility to modify master/standby port configurations (Doc).
- Optimized data flattening in mxv merge algorithms, improving performance of
vsort, vmergeappend, and vmotion .
- Improved
vmotion performance by optimizing access to interconnect interfaces .
- Enhanced
vhashagg performance with multiple grouping keys .
- Improved
VWindowAgg performance when partitions are small .
- Boosted
vhashagg performance under large memory quotas .
Bug Fixes
- Fixed several known issues
6.5.1
Date：2025-10-10
Enhancements and Improvements
- Supported automatic time synchronization across all machines in the cluster
- TDE now supports rotating master keys
- Added a command-line tool to verify database software version consistency across all cluster machines
- Added logging for the logic calculating shared_buffers, max_connections, and gp_vmem_protect_limit
- Optimized deserialization of pointer-type data with fixed-width content
- Improved reception performance when vmotion is used with the OFFSET clause
- Optimized performance degradation in vmotion caused by overly fine-grained sending
- Optimized memory overconsumption in VSort and granularity control for result set
- Optimized granularity control for vhashjoin result sets
- Optimized excessive execution time of orca plans with large in lists
- Implemented adaptation for mxnumeric type comparisons and row count estimation in orca
- Executor optimized SAOP via hash algorithms and completed orca adaptation
- SF now supports InFilter for mxnumeric types
- InFilter now supports NOT IN
- Vectorized fallback DatumInFilter utilizes hash optimization
- Optimized logic for generating parallel scan paths for replicated tables, avoiding unnecessary data movement
Bug fixes
- Fixed several known issues
6.5.0
Date:2025-09-01
New version features
- MMatrixUI adds database startup/shutdown visualization functionality (Documentation)
- YMatrix introduces Transparent Encryption (TDE), supporting both local file system and KMS system management modes (Documentation)
- Added
mxnumeric extension, providing a high-performance alternative to pg numeric types
- Changed default collate from
en_US.UTF-8 to C
pg_basebackup now disables page checksum verification by default
Enhancements and Improvements
- OpenSSL now supports managing memory used by OpenSSL dependencies via
MemoryContext
- Domino Stream Computing adds the
DROP STREAM .. CASCADE statement, enabling cascading deletion of multiple streams
- Domino Stream Computing adds dual-stream LEFT JOIN support
- Removed non-essential page checksum calculation code from mars3
- Planner now supports performance enhancement by pushing down the limit of left joins under specific conditions
default on update now supports the statement_timestamp() function- Optimized mxvector data processing granularity
- Implemented native
bpchar support for mxvector
- mxvector now supports
dynamic index / indexonly / bitmap scan
- Added GUC
default_null_ordering to control default comparison behavior for null values
- mxnumeric: Optimized AvgState serialization capabilities
- mars3 now supports btree inorder scans
- mars3 now supports incremental
analyzedb and incremental mxbackup
- mars3 relaxed sequence constraints when creating btree indexes in unique mode
- mars3 now supports
delete operations in unique mode scenarios
- mars3 introduces two new GUCs:
mars3.default_storage_options and mars3.default_btree_options, for specifying default mars3 storage parameters
Bug fixes
- Fixed several known issues
6.4.11
Date: 2026-07-03
Bug Fixes
- Fixed several known issues
6.4.10
Date: 2026-05-25
Enhancements and Improvements
- fastdd: Optimized external parameter pushdown to improve query execution efficiency in related scenarios.
- pxf_fdw: Reorganized the code as a git submodule for easier independent maintenance and version management.
- Matrixgate: Optimized goroutine usage in the internal performance statistics component
load_metric, reducing runtime resource overhead.
- Matrixgate: Introduced
automaxprocs; when GOMAXPROCS is not explicitly configured, Matrixgate automatically sets an appropriate value to improve CPU resource utilization.
- Matrixgate: Enhanced logs by adding table names to
set slot logs, making troubleshooting and operations easier.
- Matrixgate: Optimized the internal fasthttp
flush chunk logic to improve HTTP transfer efficiency.
- Matrixgate: Optimized memory usage by using fixed-size slices as lower-level buffers, reducing memory allocation overhead and fluctuation.
- Matrixgate: Added
http-segmented-body to control whether memory optimization is enabled. It is enabled by default. Set it to false to revert to the legacy Matrixgate memory allocation behavior.
- Matrixgate: Added
http-segment-size-kb to specify the size of each lower-level slice. The default value is 1024KB.
- Matrixgate: Added
write-buffer-size to specify the size of each buffer used for data transfer from Matrixgate to segments.
Bug Fixes
- Fixed several known issues
6.4.9
Date: 2026-04-15
Enhancements and Improvements
- Added support for
definitelydead in MARS3 history snapshots to optimize snapshot processing.
- The SOA service component no longer exposes the
pprof address by default. It can be temporarily enabled or disabled by sending the USR2 signal: sudo kill -USR2 -p <pid_of_soa_program>.
- Updated the internal dependency of
supervisor to github.com/cilium/ebpf to address potential security risks (PR25).
- MatrixUI now disables the
pprof port by default and includes a new switch to control pprof route mounting .
- Domino nodes containing only decoders with no active streams now correctly retain snapshot files .
- Disabled operations using
mxinitstandby to remove primary nodes where preferred_role=m. Use mxmoveseg to migrate master nodes to prevent accidental operations that are difficult to recover from .
mxshift DDL migration no longer migrates temporary sequences .
Bug Fixes
- Fixed several known issues
6.4.8
Date: 2026-02-24
Bug Fixes
- Fixed several known issues
6.4.7
Date: 2026-01-09
Bug fixes
- Fixed several known issues
6.4.6
Date: 2025-12-15
Enhancements and Improvements
- Added pg_audit plugin functionality (Documentation)
- Added postgres_anonymizer plugin functionality (Documentation)
- Enhanced memory usage debugging mechanisms .
- mars3 now supports terminating blocking compact tasks that hold locks during
DROP TABLE or TRUNCATE TABLE operations.
Bug Fixes
- Fixed several known issues
6.4.5
Date: 2025-11-20
Bug Fixes
- Fixed several known issues
6.4.4
Date: 2025-11-10
Enhancements and Improvements
- Reduced the log severity when a domino worker exits (previously logged as FATAL: terminating connection due to administrator command)
Bug Fixes
- Fixed several known issues
6.4.3
Date：2025-10-10
Enhancements and Improvements
- Added logging for the logic calculating shared_buffers, max_connections, and gp_vmem_protect_limit
- Optimized the logic for generating the Parallel Scan Path for replication tables to avoid unnecessary data movement
Bug fixes
- Fixed several known issues
6.4.2
Date: 2025-09-01
Enhancements and Improvements
- Removed non-essential code for calculating page checksums in MARS3
Bug fixes
- Fixed several known issues
6.4.1
Date: 2025-08-01
Enhancements and Improvements
- License: The free usage period for newly deployed clusters has been extended from 3 days to 14 days.
- mxinitstandby: Added functionality to delete the current standby (document)
- Starting with this version, the matrixmgr extension will be automatically upgraded after a database upgrade, eliminating the need for manual processing.
- Domino: Improved streaming latency caused by the scheduling module.
- orca now supports mars3 indexes
- Supports two-phase distinct operations
- Strengthened secure use of relcache memory
- Cluster initialization now defaults to disabling autofreeze
- Enhanced mars3 analyze logic using row-based sampling
- Supports mars3 drop index capability on partitioned tables
Bug fixes
- Fixed several known issues
6.4.0
Date: 2025-06-26
New features
- License management function is completely upgraded
- Added support for Centos9 operating system
- Added support for Cgroup V2
- Resource Group has added support for IO_LIMIT function
- MatrixUI has added disk space monitoring, including dashboard heatmap, 4 disk alarms, topology page display Util
- MatrixUI supports displaying query blocking relationships on QE
- By default, use Domino v2 stream calculation to create streams (document)
- Incremental online backup supports normal recovery when failover/recover occurs in the cluster (document)
- Added
mx_eager_memory_release switch to actively reduce the memory usage of idle processes (default off)
- Support migration of mars3 table data through
ALTER TABLE .. SET TABLESPACE statement
- Supports parallel scanning of replicate tables (default off)
- The
auto-vacuum function for system and non-system tables is turned on by default
- Golang and toolchain upgrade to version 1.24
Enhancements and Improvements
- Determine the TLog format of Domino v2, implement buffered read and write, significantly improving disk IO performance
- Reform mxstop to perform streaming and output progress information.
- Incremental online backup supports some tablespace functions (document)
- mars3 supports roi indexing
- mars3 supports automatic conversion of
btree/brin to mars3_btree/mars3_brin index
- The default relfile version of mars3 is adjusted to v1
- Mars3 partition table supports drop index
- mars3 supports shared zstd context for further compression efficiency
- Dynamic bitmap scan on mars3 table is disabled in orca
- Supports the case where the index type is
mars3_btree when modifying the primary key
- mxrule supports the ability to judge whether to project based on the number of distincts
Bug fixes
- Fixed several known issues
6.3.0
Date: 2025-04-18
New features
- Support mxaddmirrors unbalanced deployment scheme (document)
- Supports the addmirror function of graphic pages (document)
- Support incremental cold backup and recovery functions (document)
- mxstate series enhancement, adding parameters such as
-s, -p``-H`
- mxstart adds progress visualization function, adding
-C and -P parameters
- Upgrade the library intrastream to v2 version (as of the failover part, the default is still v1)
- MatrixGate list in mxui integrates the function of deleting tasks
- Matrixshift Migration DDL Added Support for USING, WITH, ECODING Attributes
- Support recursive creation of extended statistics for partition tables
- Support the ability to apply ndistinct to extend statistics to estimate the amount of result set data for composite conditions/index
Enhancements and Improvements
- The replication service does not first crash OOM when system OOM
- Upgrade ETCD to 3.5.x and pass fatigue test verification
- Golang toolchain upgrade to 1.23.7
- Upgrade go-common-libs & matrixapis, clean up CVE
- mxbox refactoring database connection layer
- pgvector upgrade to v0.8.0
Bug fixes
- Fixed several known issues
6.2.4
Date: 2025-03-03
Enhancements and Improvements
- Optimizer
- Optimized ROI logic, no dummy plan is generated when the corresponding partition cannot be found from the ROI
- Tools
- Optimized the success rate of probing all segments before mxstart starts the node
Bug Fix
- Fixed several known issues
6.2.3
Date: 2025-02-24
Enhancements and Improvements
- mxbox
- Optimized redundant information about cluster state in etcd, reduces the pressure to write and update etcd services, and enhances cluster stability
- Extended the function of the
mxbox client deployer get-cluster-service command line tool. In addition to display the currently active cluster/shard service address, it can also display information about inactive highly available services in standby. At the same time, add a required parameter --service-name, with the value cluster or shard
- Caches the information discovered by the SOA service in the deployer service. Other services obtain service discovery results by accessing the deployer API interface, reducing the pressure on direct access to etcd and improving operational stability under large-scale clusters
- High Availability Service Process
cluster/shard/replication/deployer will more actively release physical memory to the operating system and reduce memory usage
- When mxstart/mxstop is operated, if the cluster size is large, the shard/replication will be started in batches to avoid the instantaneous ultra-high load on etcd by concurrent startups of all services, which enhances operational stability
- When primary/mirror under a shard is down, the node in
gp_segment_configuration that retains the state of the last primary role is UP to ensure that the database can still be restarted successfully
- The mxrecover tool has added the
-s option to specify the shard to be restored
- The mxstate tool has added the
-e option to output segment information in an absolute state
- Disaster recovery
- The
restore_point_creation_interval_minute was deleted in the configuration file of the subscriber process, and the restore_point_creation_interval (required) parameter was added. Users can configure the synchronization interval by themselves
- mxaddmirrors plan now supports the generation of mirror node distributions for clusters for unbalanced deployments. It is recommended that when using custom distributions for unbalanced clusters, they should be used after manual review and adjustment to the most reasonable mirror layout
- Changed how mxaddmirror custom configuration is used. Users need to first
mxaddmirrors init, and then mxaddmirrors config --init-file <path_to_file> to generate config templates. Now, even for unbalanced clusters, a reasonable configuration scheme can be generated according to the new usage method
- For unbalanced deployment clusters, mxinitstandby allows to add standby to a blank physical machine without deployed segment nodes
- For unbalanced deployment clusters, mxaddmirror allows adding mirror to a blank physical machine without deployed segment nodes in custom configuration mode
Bug Fix
- Fixed several known issues
6.2.2
Date: 2025-01-10
Enhancements and Improvements
- MatrixShift
- If the configuration file does not specify any migration operation, and only when the verification function is turned on, an error message is added to a database that is migrated from non-mxshift
- When there are duplicate tables or parent tables and some child tables appear in table-list in the scope paragraph of the configuration file, the ability to automatically remove duplicate tables and report warnings
- MatrixGate
- Data transfer between mxgate and segment nodes is adjusted to not enable compression by default to save more memory resources. The original default value is Enable compression
- In transfer mode, allow the migration of data when the mirror node is in Down state
- mxbox
- Added the skip reason output when mxrecover skips certain shards
- When using mxrecover to perform rebalance operations on a cluster, you can shorten the database downtime caused by the operation by adding
CHECKPOINT, and you can skip this behavior by adding --skip-checkpoint parameter
- Improved output content format of this command in the console when rebalanced to the cluster using mxrecover
- mxrecover Adds the
--prepare-timeout parameter. When mxrecover starts, the timeout of cluster status information can be collected through etcd, default is 1 minute
- mxctl
- When installing and deploying, the mxctl setup collect command will check whether localhost can be resolved to a local loopback address, and if it cannot, an error will be reported
Bug fixes
- Fixed several known issues
6.2.1
Date: 2024-12-20
Enhancements and Improvements
- mxbox
- The default timeout for mxstart increases from 1 minute to 5 minutes. It still supports users to customize the timeout with the -t parameter.
- Supports user-defined environment variables when the postmaster process is started.
- When a double failure occurs in the cluster, it allows the nearest primary node to be started by restarting the database instead of failing to start.
- Improved the number of max open files for supervisor service and etcd process, solving the problem of frequent too many open files errors for etcd services under hyperscale clusters. (#6ca5293dcd0)
- When a new cluster is deployed, the --enable-pprof parameter of the etcd service was deleted in the service configuration of the supervisor, making the etcd service more stable and the CPU proportion is lower.
Bug Fix
- Fixed several known issues
6.2.0
Date: 2024-12-09
Enhancements and Improvements
- MatrixUI
- When installing or expanding a new machine, it will actively check whether /etc/ssh/sshd_config contains AllowUsers/AllowGroups/DenyUsers/DenyGroups and other configuration items that may cause the ssh mutual trust operation of the host in the cluster during the installation process, and give specific error information and modification prompts
- MatrixShift
- Introduce stricter checks on configuration files, incorrect configuration items will be reported and prompted when mxshift starts. For example, disable-connector = true should be written in the [scope] paragraph, but the error is written in other paragraphs. mxshift will report an error and remind you when starting
- mxbox
- moveseg / initstandby tool can detect whether the host added by the --add subcommand is already in the physical machine cluster and give an error message
- When the host restarts or runs the YMatrix cluster for the first time, it can automatically create the relevant paths that the resource group cgroup depends on
Bug Fix
- Fixed several known issues
6.1.1
Date: 2024-11-19
Enhancements and Improvements
- MatrixGate
- mxgate optimization: Supports segments connection timeout as configurable GUC
- MatrixShift
- mxshift optimization: /tmp use data dir when there is no permission
- mxbox
- Supports custom unbalanced deployments
- MatrixUI (Document)
- Supports custom unbalanced deployments UI
6.1.0
Date: 2024-10-31
Enhancements and Improvements
-
MatrixShift
- mxshift optimization: migration from gp6 data can be skipped external table
- mxshift optimization: compresslevel is forced to 1 when compresstype is automatically mapped from zlib to zstd
-
Disaster recovery
- Removed useless configuration item slot in subscriber configuration file
- Adjusted content in the log
Bug Fix
- Fixed several known issues
6.0.0
Date: 2024-10-11
New Feature
- Added stream computing function (Document)
- Supports expanded dimensional computing, aggregation computing, multi-level flow computing, dual-stream correlation computing, etc. (Document)
- Supports multiple configuration and management capabilities (document)
2.Integrated flow batch architecture
- The database fusion stream processing and batch processing advantages, supporting the parallel processing of real-time data and historical data.
- Added Disaster Recovery Cluster (Document)
- Supports data transmission and synchronization across data centers
- Supports the operational status monitoring of disaster recovery clusters.
- Supports manual switching of backup clusters to service clusters, and can switch all applications to this cluster for normal business operation.
- The disaster recovery metrics for this disaster recovery cluster are as follows:
- RPO (Recovery Point Objective, recovery point target)
The user can specify a relatively appropriate value to set this target to the minute level without affecting the performance of the source cluster.
- RTO (Recovery Time Objective, recovery time target)
Supports database hot standby implementation, and RTO can reach seconds to minutes.
Performance enhancement
-
TP performance enhancement
- Added support for Blocksize 8K, and the configurable data block (block) is 8KB, which greatly improves the transaction processing performance of small data volumes.
-
MARS3 Storage Engine Enhancement
- Added support for unique key, upsert, row-level lock capabilities.
-
Vectorization engine enhancement
- Added support for Window Function vectorization.
Know issues
-
When there are many partitions, the query will trigger the database OOM.
-
In the KILL QUERY scenario, the recovery process may experience crash problems.
