TiDB 7.2.0 Release Notes

Release date: June 29, 2023

TiDB version: 7.2.0

Quick access: Quick start | Installation packages

7.2.0 introduces the following key features and improvements:

Feature details

Performance

• Support pushing down the following two window functions to TiFlash #7427 @xzhangxian1008

  • FIRST_VALUE
  • LAST_VALUE

• TiFlash supports the pipeline execution model (experimental) #6518 @SeaRise

  Prior to v7.2.0, each task in the TiFlash engine must individually request thread resources during execution. TiFlash controls the number of tasks to limit thread resource usage and prevent overuse, but this issue could not be completely eliminated. To address this problem, starting from v7.2.0, TiFlash introduces a pipeline execution model. This model centrally manages all thread resources and schedules task execution uniformly, maximizing the utilization of thread resources while avoiding resource overuse. To enable or disable the pipeline execution model, modify the tidb_enable_tiflash_pipeline_model system variable.

  For more information, see documentation.

• TiFlash reduces the latency of schema replication #7630 @hongyunyan

  When the schema of a table changes, TiFlash needs to replicate the latest schema from TiKV in a timely manner. Before v7.2.0, when TiFlash accesses table data and detects a table schema change within a database, TiFlash needs to replicate the schemas of all tables in this database again, including those tables without TiFlash replicas. As a result, in a database with a large number of tables, even if you only need to read data from a single table using TiFlash, you might experience significant latency to wait for TiFlash to complete the schema replication of all tables.

  In v7.2.0, TiFlash optimizes the schema replication mechanism and supports only replicating schemas of tables with TiFlash replicas. When a schema change is detected for a table with TiFlash replicas, TiFlash only replicates the schema of that table, which reduces the latency of schema replication of TiFlash and minimizes the impact of DDL operations on TiFlash data replication. This optimization is automatically applied and does not require any manual configuration.

• Improve the performance of statistics collection #44725 @xuyifangreeneyes

  TiDB v7.2.0 optimizes the statistics collection strategy, skipping some of the duplicate information and information that is of little value to the optimizer. The overall speed of statistics collection has been improved by 30%. This improvement allows TiDB to update the statistics of the database in a more timely manner, making the generated execution plans more accurate, thus improving the overall database performance.

  By default, statistics collection skips the columns of the JSON, BLOB, MEDIUMBLOB, and LONGBLOB types. You can modify the default behavior by setting the tidb_analyze_skip_column_types system variable. TiDB supports skipping the JSON, BLOB, and TEXT types and their subtypes.

  For more information, see documentation.

• Improve the performance of checking data and index consistency #43693 @wjhuang2016

  The ADMIN CHECK [TABLE|INDEX] statement is used to check the consistency between data in a table and its corresponding indexes. In v7.2.0, TiDB optimizes the method for checking data consistency and improves the execution efficiency of ADMIN CHECK [TABLE|INDEX] greatly. In scenarios with large amounts of data, this optimization can provide a performance boost of hundreds of times.

  The optimization is enabled by default (tidb_enable_fast_table_check is ON by default) to greatly reduce the time required for data consistency checks in large-scale tables and enhance operational efficiency.

  For more information, see documentation.

Reliability

• Automatically manage queries that consume more resources than expected (experimental) #43691 @Connor1996 @CabinfeverB @glorv @HuSharp @nolouch

  The most common challenge to database stability is the degradation of overall database performance caused by abrupt SQL performance problems. There are many causes for SQL performance issues, such as new SQL statements that have not been fully tested, drastic changes in data volume, and abrupt changes in execution plans. These issues are difficult to completely avoid at the root. TiDB v7.2.0 provides the ability to manage queries that consume more resources than expected. This feature can quickly reduce the scope of impact when a performance issue occurs.

  To manage these queries, you can set the maximum execution time of queries for a resource group. When the execution time of a query exceeds this limit, the query is automatically deprioritized or cancelled. You can also set a period of time to immediately match identified queries by text or execution plan. This helps prevent high concurrency of the problematic queries during the identification phase that could consume more resources than expected.

  Automatic management of queries that consume more resources than expected provides you with an effective means to quickly respond to unexpected query performance problems. This feature can reduce the impact of the problem on overall database performance, thereby improving database stability.

  For more information, see documentation.

• Enhance the capability of creating a binding according to a historical execution plan #39199 @qw4990

  TiDB v7.2.0 enhances the capability of creating a binding according to a historical execution plan. This feature improves the parsing and binding process for complex statements, making the bindings more stable, and supports the following new hints:

  • AGG_TO_COP()
  • LIMIT_TO_COP()
  • ORDER_INDEX
  • NO_ORDER_INDEX()

  For more information, see documentation.

• Introduce the Optimizer Fix Controls mechanism to provide fine-grained control over optimizer behaviors #43169 @time-and-fate

  To generate more reasonable execution plans, the behavior of the TiDB optimizer evolves over product iterations. However, in some particular scenarios, the changes might lead to performance regression. TiDB v7.2.0 introduces Optimizer Fix Controls to let you control some of the fine-grained behaviors of the optimizer. This enables you to roll back or control some new changes.

  Each controllable behavior is described by a GitHub issue corresponding to the fix number. All controllable behaviors are listed in Optimizer Fix Controls. You can set a target value for one or more behaviors by setting the tidb_opt_fix_control system variable to achieve behavior control.

  The Optimizer Fix Controls mechanism helps you control the TiDB optimizer at a granular level. It provides a new means of fixing performance issues caused by the upgrade process and improves the stability of TiDB.

  For more information, see documentation.

• Lightweight statistics initialization becomes generally available (GA) #42160 @xuyifangreeneyes

  Starting from v7.2.0, the lightweight statistics initialization feature becomes GA. Lightweight statistics initialization can significantly reduce the number of statistics that must be loaded during startup, thus improving the speed of loading statistics. This feature increases the stability of TiDB in complex runtime environments and reduces the impact on the overall service when TiDB nodes restart.

  For newly created clusters of v7.2.0 or later versions, TiDB loads lightweight statistics by default during TiDB startup and will wait for the loading to finish before providing services. For clusters upgraded from earlier versions, you can set the TiDB configuration items lite-init-stats and force-init-stats to true to enable this feature.

  For more information, see documentation.

SQL

• Support the CHECK constraints #41711 @fzzf678

  Starting from v7.2.0, you can use CHECK constraints to restrict the values of one or more columns in a table to meet your specified conditions. When a CHECK constraint is added to a table, TiDB checks whether the constraint is satisfied before inserting or updating data in the table. Only the data that satisfies the constraint can be written.

  This feature is disabled by default. You can set the tidb_enable_check_constraint system variable to ON to enable it.

  For more information, see documentation.

DB operations

• DDL jobs support pause and resume operations (experimental) #18015 @godouxm

  Before TiDB v7.2.0, when a DDL job encounters a business peak during execution, you can only manually cancel the DDL job to reduce its impact on the business. In v7.2.0, TiDB introduces pause and resume operations for DDL jobs. These operations let you pause DDL jobs during a peak and resume them after the peak ends, thus avoiding impact on your application workloads.

  For example, you can pause and resume multiple DDL jobs using ADMIN PAUSE DDL JOBS or ADMIN RESUME DDL JOBS:

  ADMIN PAUSE DDL JOBS 1,2;
  ADMIN RESUME DDL JOBS 1,2;
  

  For more information, see documentation.

Data migration

• Introduce a new SQL statement IMPORT INTO to improve data import efficiency greatly (experimental) #42930 @D3Hunter

  The IMPORT INTO statement integrates the Physical Import Mode capability of TiDB Lightning. With this statement, you can quickly import data in formats such as CSV, SQL, and PARQUET into an empty table in TiDB. This import method eliminates the need for a separate deployment and management of TiDB Lightning, thereby reducing the complexity of data import and greatly improving import efficiency.

  For data files stored in Amazon S3 or GCS, when the Backend task distributed execution framework is enabled, IMPORT INTO also supports splitting a data import job into multiple sub-jobs and scheduling them to multiple TiDB nodes for parallel import, which further enhances import performance.

  For more information, see documentation.

• TiDB Lightning supports importing source files with the Latin-1 character set into TiDB #44434 @lance6716

  With this feature, you can directly import source files with the Latin-1 character set into TiDB using TiDB Lightning. Before v7.2.0, importing such files requires your additional preprocessing or conversion. Starting from v7.2.0, you only need to specify character-set = "latin1" when configuring the TiDB Lightning import task. Then, TiDB Lightning automatically handles the character set conversion during the import process to ensure data integrity and accuracy.

  For more information, see documentation.

Compatibility changes

System variables

Configuration file parameters

Improvements

• TiDB

  • Optimize the logic of constructing index scan range so that it supports converting complex conditions into index scan range #41572 #44389 @xuyifangreeneyes
  • Add new monitoring metrics Stale Read OPS and Stale Read Traffic #43325 @you06
  • When the retry leader of stale read encounters a lock, TiDB forcibly retries with the leader after resolving the lock, which avoids unnecessary overhead #43659 @you06
  • Use estimated time to calculate stale read ts and reduce the overhead of stale read #44215 @you06
  • Add logs and system variables for long-running transactions #41471 @crazycs520
  • Support connecting to TiDB through the compressed MySQL protocol, which improves the performance of data-intensive queries under low bandwidth networks and saves bandwidth costs. This supports both zlib and zstd based compression. #22605 @dveeden
  • Recognize both utf8 and utf8bm3 as the legacy three-byte UTF-8 character set encodings, which facilitates the migration of tables with legacy UTF-8 encodings from MySQL 8.0 to TiDB #26226 @dveeden
  • Support using := for assignment in UPDATE statements #44751 @CbcWestwolf

• TiKV

  • Support configuring the retry interval of PD connections in scenarios such as connection request failures using pd.retry-interval #14964 @rleungx
  • Optimize the resource control scheduling algorithm by incorporating the global resource usage #14604 @Connor1996
  • Use gzip compression for check_leader requests to reduce traffic #14553 @you06
  • Add related metrics for check_leader requests #14658 @you06
  • Provide detailed time information during TiKV handling write commands #12362 @cfzjywxk

• PD

  • Use a separate gRPC connection for PD leader election to prevent the impact of other requests #6403 @rleungx
  • Enable the bucket splitting by default to mitigate hotspot issues in multi-Region scenarios #6433 @bufferflies

• Tools

  • Backup & Restore (BR)

    • Support access to Azure Blob Storage by shared access signature (SAS) #44199 @Leavrth

  • TiCDC

    • Optimize the structure of the directory where data files are stored when a DDL operation occurs in the scenario of replication to an object storage service #8891 @CharlesCheung96
    • Support the OAUTHBEARER authentication in the scenario of replication to Kafka #8865 @hi-rustin
    • Add the option of outputting only the handle keys for the DELETE operation in the scenario of replication to Kafka #9143 @3AceShowHand

  • TiDB Data Migration (DM)

    • Support reading compressed binlogs in MySQL 8.0 as a data source for incremental replication #6381 @dveeden

  • TiDB Lightning

    • Optimize the retry mechanism during import to avoid errors caused by leader switching #44478 @lance6716
    • Verify checksum through SQL after the import to improve stability of verification #41941 @GMHDBJD
    • Optimize TiDB Lightning OOM issues when importing wide tables 43853 @D3Hunter

Bug fixes

• TiDB

  • Fix the issue that the query with CTE causes TiDB to hang #43749 #36896 @guo-shaoge
  • Fix the issue that the min, max query result is incorrect #43805 @wshwsh12
  • Fix the issue that the SHOW PROCESSLIST statement cannot display the TxnStart of the transaction of the statement with a long subquery time #40851 @crazycs520
  • Fix the issue that the stale read global optimization does not take effect due to the lack of TxnScope in Coprocessor tasks #43365 @you06
  • Fix the issue that follower read does not handle flashback errors before retrying, which causes query errors #43673 @you06
  • Fix the issue that data and indexes are inconsistent when the ON UPDATE statement does not correctly update the primary key #44565 @zyguan
  • Modify the upper limit of the UNIX_TIMESTAMP() function to 3001-01-19 03:14:07.999999 UTC to be consistent with that of MySQL 8.0.28 or later versions #43987 @YangKeao
  • Fix the issue that adding an index fails in the ingest mode #44137 @tangenta
  • Fix the issue that canceling a DDL task in the rollback state causes errors in related metadata #44143 @wjhuang2016
  • Fix the issue that using memTracker with cursor fetch causes memory leaks #44254 @YangKeao
  • Fix the issue that dropping a database causes slow GC progress #33069 @tiancaiamao
  • Fix the issue that TiDB returns an error when the corresponding rows in partitioned tables cannot be found in the probe phase of index join #43686 @AilinKid @mjonss
  • Fix the issue that there is no warning when using SUBPARTITION to create partitioned tables #41198 #41200 @mjonss
  • Fix the issue that when a query is killed because it exceeds MAX_EXECUTION_TIME, the returned error message is inconsistent with that of MySQL #43031 @dveeden
  • Fix the issue that the LEADING hint does not support querying block aliases #44645 @qw4990
  • Modify the return type of the LAST_INSERT_ID() function from VARCHAR to LONGLONG to be consistent with that of MySQL #44574 @Defined2014
  • Fix the issue that incorrect results might be returned when using a common table expression (CTE) in statements with non-correlated subqueries #44051 @winoros
  • Fix the issue that Join Reorder might cause incorrect outer join results #44314 @AilinKid
  • Fix the issue that PREPARE stmt FROM "ANALYZE TABLE xxx" might be killed by tidb_mem_quota_query #44320 @chrysan

• TiKV

  • Fix the issue that the transaction returns an incorrect value when TiKV handles stale pessimistic lock conflicts #13298 @cfzjywxk
  • Fix the issue that in-memory pessimistic lock might cause flashback failures and data inconsistency #13303 @JmPotato
  • Fix the issue that the fair lock might be incorrect when TiKV handles stale requests #13298 @cfzjywxk
  • Fix the issue that autocommit and point get replica read might break linearizability #14715 @cfzjywxk

• PD

  • Fix the issue that redundant replicas cannot be automatically repaired in some corner cases #6573 @nolouch

• TiFlash

  • Fix the issue that queries might consume more memory than needed when the data on the Join build side is very large and contains many small string type columns #7416 @yibin87

• Tools

  • Backup & Restore (BR)

    • Fix the issue that checksum mismatch is falsely reported in some cases #44472 @Leavrth
    • Fix the issue that resolved lock timeout is falsely reported in some cases #43236 @YuJuncen
    • Fix the issue that TiDB might panic when restoring statistics information #44490 @tangenta

  • TiCDC

    • Fix the issue that Resolved TS does not advance properly in some cases #8963 @CharlesCheung96
    • Fix the issue that the UPDATE operation cannot output old values when the Avro or CSV protocol is used #9086 @3AceShowHand
    • Fix the issue of excessive downstream pressure caused by reading downstream metadata too frequently when replicating data to Kafka #8959 @hi-rustin
    • Fix the issue of too many downstream logs caused by frequently setting the downstream bidirectional replication-related variables when replicating data to TiDB or MySQL #9180 @asddongmen
    • Fix the issue that the PD node crashing causes the TiCDC node to restart #8868 @asddongmen
    • Fix the issue that TiCDC cannot create a changefeed with a downstream Kafka-on-Pulsar #8892 @hi-rustin

  • TiDB Lightning

    • Fix the TiDB Lightning panic issue when experimental.allow-expression-index is enabled and the default value is UUID #44497 @lichunzhu
    • Fix the TiDB Lightning panic issue when a task exits while dividing a data file #43195 @lance6716

Contributors

We would like to thank the following contributors from the TiDB community:

• asjdf
• blacktear23
• Cavan-xu
• darraes
• demoManito
• dhysum
• HappyUncle
• jiyfhust
• L-maple
• nyurik
• SeigeC
• tangjingyu97