Release 3.0.7
Behavior Changes
- Adjust the permission requirements for
show frontends
andshow backends
to align with the corresponding RESTful API, i.e., requiring theSELECT_PRIV
permission on theinformation_schema
database - Admin and root users with specified domains are no longer considered system users
- Storage: The default number of concurrent transactions per database is adjusted to 10000
New Features
Query Optimizer
- Support MySQL's aggregate roll-up syntax
GROUP BY ... WITH ROLLUP
Query Execution
Like
statement supportsescape
syntax
Semi-structured Data Management
- Support building non-tokenized inverted indexes and ngram bloomfilter indexes only for new data by setting the session variable
enable_add_index_for_new_data=true
New Functions
- Added data functions:
cot
/sec
/cosec
Improvements
Data Ingestion
- Optimize error message prompts for
SHOW CREATE LOAD
Primary Key Model
- Add segment key bounds truncation capability to avoid single large import failures
Storage
- Enhance the reliability of compaction and imported data
- Optimize balance speed
- Optimize table creation speed
- Optimize compaction default parameters and observability
- Optimize the issue of query error -230
- Add system table
backend_tablets
- Optimize the performance of querying
information_schema.tables
from follower nodes in cloud mode
Storage-Compute Decoupled
- Enhance observability of Meta-service recycler
- Support cross-compute group incremental preheating during import compaction
- Optimize Storage vault connectivity check
- Support updating storage backend information via MS API
Lakehouse
- Optimize ORC zlib decompression performance in x86 environment and fix potential issues
- Optimize the default number of concurrent threads for external table reading
- Optimize error messages for Catalogs that do not support DDL operations
Asynchronous Materialized Views
- Optimize the performance of transparent rewriting planning
Query Optimizer
- The
group_concat
function now allows parameters of non-string types - The
sum
andavg
functions allow parameters of non-numeric types - Expand the scope of support for delayed materialization in TOP-N queries, enabling delayed materialization when querying partial columns
- When creating partitions, list partitions allow inclusion of
MAX_VALUE
- Optimize the performance of sampling and collecting statistical information for aggregate model tables
- Optimize the accuracy of NDV values when sampling and collecting statistical information
Inverted Index
- Unify the order of properties displayed for inverted indexes in
show create table
- Add per-condition profile metrics (such as hit rows and execution time) for inverted index filter conditions to facilitate performance analysis
- Enhance the display of inverted index-related information in profiles
Permissions
- Ranger supports setting permissions for storage vault and compute group
Bug Fixes
Data Ingestion
- Fix the correctness issue that may occur when importing CSV files with multi-character separators
- Fix the issue where the result of
ROUTINE LOAD
task display is incorrect after modifying task properties - Fix the issue where the one-stream multi-table import plan becomes invalid after primary node restart or Leader switch
- Fix the issue where all scheduling tasks are blocked because
ROUTINE LOAD
tasks cannot find available BE nodes - Fix the concurrent read-write conflict issue of
runningTxnIds
Primary Key Model
- Optimize the import performance of mow tables under high-frequency concurrent imports
- mow table full compaction releases space of deleted data
- Fix the potential import failure issue of mow tables in extreme scenarios
- Optimize the compaction performance of mow tables
- Fix the potential correctness issue of mow tables during concurrent imports and schema changes
- Fix the issue where schema change on empty mow tables may cause import stuck or schema change failure
- Fix the memory leak issue of mow delete bitmap cache
- Fix the potential correctness issue of mow tables after schema change
Storage
- Fix the missing rowset issue in clone process caused by compaction
- Fix the issue of inaccurate size calculation and default value for autobucket
- Fix the potential correctness issue caused by bucket columns
- Fix the issue where single-column tables cannot be renamed
- Fix the potential memory leak issue of memtable
- Fix the inconsistent error reporting issue for unsupported operations in empty table transaction writes
Storage-Compute Decoupled
- Several fixes for File cache
- Fix the issue where cumulative point may roll back during schema process
- Fix the issue where background tasks affect automatic restart
- Fix the unhandled exception issue in data recycling process in azure environment
- Fix the issue where file cache is not cleaned up in time when compacting a single rowset
Lakehouse
- Fix the transaction commit failure issue for Iceberg table writes in Kerberos environment
- Fix the query issue for hudi in kerberos environment
- Fix the potential deadlock issue in multi-Catalog scenarios
- Fix the metadata inconsistency issue caused by concurrent Catalog refresh in some cases
- Fix the issue where ORC footer is read multiple times in some cases
- Fix the issue where Table Valued Function cannot read compressed json files
- SQL Server Catalog supports identifying IDENTITY column information
- SQL Convertor supports specifying multiple URLs for high availability
Asynchronous Materialized Views
- Fix the issue where partition compensation may be performed incorrectly when the query is optimized to an empty result set
Query Optimizer
- Fix the issue where factors other than
sql_select_limit
affect DML execution results - Fix the issue where materialized CTEs may report errors in extreme cases when starting local shuffle
- Fix the issue where prepared insert statements cannot be executed on non-master nodes
- Fix the result error issue when casting
ipv4
to string
Permissions
- When a user has multiple roles, the permissions of the multiple roles will be merged before authorization
Query Execution
- Fix issues with some json functions
- Fix the potential BE Core issue when the asynchronous thread pool is full
- Fix the incorrect result issue of
hll_to_base64
- Fix the result error issue when casting
decimal256
to float - Fix two memory leak issues
- Fix the be core issue caused by
bitmap_from_base64
- Fix the potential be core issue caused by
array_map
function - Fix the potential error issue of
split_by_regexp
function - Fix the potential result error issue of
bitmap_union
function under extremely large data volumes - Fix the potential core issue of
format round
function under some boundary values
Inverted Index
- Fix the memory leak issue of inverted indexes in abnormal situations
- Fix the error reporting issue when writing and querying empty index files
- Capture IO exceptions in inverted index string reading to avoid process crash due to exceptions
Complex Data Types
- Fix the potential type inference error when Variant Nested data types conflict
- Fix the parameter type inference error of
map
function - Fix the issue where data is incorrectly converted to NULL when specifying
'$.'
as the path in jsonpath - Fix the issue where the serialization format cannot be restored when a subfield of Variant contains
.
Others
- Fix the insufficient length issue of the IP field in the auditlog table
- Fix the issue where the query id recorded in the audit log is that of the previous query when SQL parsing fails