TiDB 3.0 GA Release Notes

Release date: June 28, 2019

TiDB version: 3.0.0

TiDB Ansible version: 3.0.0

Overview

On June 28, 2019, TiDB 3.0 GA is released. The corresponding TiDB Ansible version is 3.0.0. Compared with TiDB 2.1, this release has greatly improved in the following aspects:

  • Stability. TiDB 3.0 has demonstrated long-term stability for large-scale clusters with up to 150+ nodes and 300+ TB of storage.
  • Usability. TiDB 3.0 has multi-facet improvements in usability, including standardized slow query logs, well-developed log file specification, and new features such as EXPLAIN ANALYZE and SQL Trace to save operation costs for users.
  • Performance. The performance of TiDB 3.0 is 4.5 times greater than TiDB 2.1 in TPC-C benchmarks, and over 1.5 times in Sysbench benchmarks. Thanks to the support for Views, TPC-H 50G Q15 can now run normally.
  • New features including Window Functions, Views (Experimental), partitioned tables, the plugin framework, pessimistic locking (Experimental), and SQL Plan Management.

TiDB

  • New Features
    • Support Window Functions; compatible with all window functions in MySQL 8.0, including NTILE, LEAD, LAG, PERCENT_RANK, NTH_VALUE, CUME_DIST, FIRST_VALUE , LAST_VALUE, RANK, DENSE_RANK, and ROW_NUMBER
    • Support Views (Experimental)
    • Improve Table Partition
      • Support Range Partition
      • Support Hash Partition
    • Add the plug-in framework, supporting plugins such as IP Whitelist (Enterprise) and Audit Log (Enterprise).
    • Support the SQL Plan Management function to create SQL execution plan binding to ensure query stability (Experimental)
  • SQL Optimizer
    • Optimize the NOT EXISTS subquery and convert it to Anti Semi Join to improve performance
    • Optimize the constant propagation on the Outer Join, and add the optimization rule of Outer Join elimination to reduce non-effective computations and improve performance
    • Optimize the IN subquery to execute Inner Join after aggregation to improve performance
    • Optimize Index Join to adapt to more scenarios
    • Improve the Partition Pruning optimization rule of Range Partition
    • Optimize the query logic for _tidb_rowidto avoid full table scan and improve performance
    • Match more prefix columns of the indexes when extracting access conditions of composite indexes if there are relevant columns in the filter to improve performance
    • Improve the accuracy of cost estimates by using order correlation between columns
    • Optimize Join Order based on the greedy strategy and the dynamic programming algorithm to speed up the join operation of multiple tables
    • Support Skyline Pruning, with some rules to prevent the execution plan from relying too heavily on statistics to improve query stability
    • Improve the accuracy of row count estimation for single-column indexes with NULL values
    • Support FAST ANALYZE that randomly samples in each Region to avoid full table scan and improve performance with statistics collection
    • Support the incremental Analyze operation on monotonically increasing index columns to improve performance with statistics collection
    • Support using subqueries in the DO statement
    • Support using Index Join in transactions
    • Optimize prepare/execute to support DDL statements with no parameters
    • Modify the system behavior to auto load statistics when the stats-lease variable value is 0
    • Support exporting historical statistics
    • Support the dump/load correlation of histograms
  • SQL Execution Engine
    • Optimize log output: EXECUTE outputs user variables and COMMIT outputs slow query logs to facilitate troubleshooting
    • Support the EXPLAIN ANALYZE function to improve SQL tuning usability
    • Support the admin show next_row_id command to get the ID of the next row
    • Add six built-in functions: JSON_QUOTE, JSON_ARRAY_APPEND, JSON_MERGE_PRESERVE, BENCHMARK ,COALESCE, and NAME_CONST
    • Optimize control logics on the chunk size to dynamically adjust based on the query context, to reduce the SQL execution time and resource consumption
    • Support tracking and controlling memory usage in three operators - TableReader, IndexReader and IndexLookupReader
    • Optimize the Merge Join operator to support an empty ON condition
    • Optimize write performance for single tables that contains too many columns
    • Improve the performance of admin show ddl jobs by supporting scanning data in reverse order
    • Add the split table region statement to manually split the table Region to alleviate hotspot issues
    • Add the split index region statement to manually split the index Region to alleviate hotspot issues
    • Add a blocklist to prohibit pushing down expressions to Coprocessor
    • Optimize the Expensive Query log to print the SQL query in the log when it exceeds the configured limit of execution time or memory
  • DDL
    • Support migrating from character set utf8 to utf8mb4
    • Change the default character set fromutf8 to utf8mb4
    • Add the alter schema statement to modify the character set and the collation of the database
    • Support ALTER algorithm INPLACE/INSTANT
    • Support SHOW CREATE VIEW
    • Support SHOW CREATE USER
    • Support fast recovery of mistakenly deleted tables
    • Support adjusting the number of concurrencies of ADD INDEX dynamically
    • Add the pre_split_regions option that pre-allocates Regions when creating the table using the CREATE TABLE statement, to relieve write hot Regions caused by lots of writes after the table creation
    • Support splitting Regions by the index and range of the table specified using SQL statements to relieve hotspot issues
    • Add the ddl_error_count_limit global variable to limit the number of DDL task retries
    • Add a feature to use SHARD_ROW_ID_BITS to scatter row IDs when the column contains an AUTO_INCREMENT attribute to relieve hotspot issues
    • Optimize the lifetime of invalid DDL metadata to speed up recovering the normal execution of DDL operations after upgrading the TiDB cluster
  • Transactions
    • Support the pessimistic transaction mode (Experimental)
    • Optimize transaction processing logics to adapt to more scenarios:
      • Change the default value tidb_disable_txn_auto_retry to on, which means non-auto committed transactions will not be retried
      • Add the tidb_batch_commit system variable to split a transaction into multiple ones to be executed concurrently
      • Add the tidb_low_resolution_tso system variable to control the number of TSOs to obtain in batches and reduce the number of times that transactions request for TSOs, to improve performance in scenarios with relatively low requirement of consistency
      • Add the tidb_skip_isolation_level_check variable to control whether to report errors when the isolation level is set to SERIALIZABLE
      • Modify the tidb_disable_txn_auto_retry system variable to make it work on all retryable errors
  • Permission Management
    • Perform permission check on the ANALYZE, USE, SET GLOBAL, and SHOW PROCESSLIST statements
    • Support Role Based Access Control (RBAC) (Experimental)
  • Server
    • Optimize slow query logs:
      • Restructure the log format
      • Optimize the log content
      • Optimize the log query method to support using the INFORMATION_SCHEMA.SLOW_QUERY and ADMIN SHOW SLOW statements of the memory table to query slow query logs
    • Develop a unified log format specification with restructured log system to facilitate collection and analysis by tools
    • Support using SQL statements to manage TiDB Binlog services, including querying status, enabling TiDB Binlog, maintaining and sending TiDB Binlog strategies.
    • Support using unix_socket to connect to the database
    • Support Trace for SQL statements
    • Support getting information for a TiDB instance via the /debug/zip HTTP interface to facilitate troubleshooting.
    • Optimize monitoring items to facilitate troubleshooting:
      • Add the high_error_rate_feedback_total monitoring item to monitor the difference between the actual data volume and the estimated data volume based on statistics
      • Add a QPS monitoring item in the database dimension
    • Optimize the system initialization process to only allow the DDL owner to perform the initialization. This reduces the startup time for initialization or upgrading.
    • Optimize the execution logic of kill query to improve performance and ensure resource is release properly
    • Add a startup option config-check to check the validity of the configuration file
    • Add the tidb_back_off_weight system variable to control the backoff time of internal error retries
    • Add the wait_timeoutand interactive_timeout system variables to control the maximum idle connections allowed
    • Add the connection pool for TiKV to shorten the connection establishing time
  • Compatibility
    • Support the ALLOW_INVALID_DATES SQL mode
    • Support the MySQL 320 Handshake protocol
    • Support manifesting unsigned BIGINT columns as auto-increment columns
    • Support the SHOW CREATE DATABASE IF NOT EXISTS syntax
    • Optimize the fault tolerance of load data for CSV files
    • Abandon the predicate pushdown operation when the filtering condition contains a user variable to improve the compatibility with MySQL’s behavior of using user variables to simulate Window Functions

PD

  • Support re-creating a cluster from a single node
  • Migrate Region metadata from etcd to the go-leveldb storage engine to solve the storage bottleneck in etcd for large-scale clusters
  • API
    • Add the remove-tombstone API to clear Tombstone stores
    • Add the ScanRegions API to batch query Region information
    • Add the GetOperator API to query running operators
    • Optimize the performance of the GetStores API
  • Configurations
    • Optimize configuration check logic to avoid configuration item errors
    • Add enable-two-way-merge to control the direction of Region merge
    • Add hot-region-schedule-limit to control the scheduling rate for hot Regions
    • Add hot-region-cache-hits-threshold to identify hotspot when hitting multiple thresholds consecutively
    • Add the store-balance-rate configuration item to control the maximum numbers of balance Region operators allowed per minute
  • Scheduler Optimizations
    • Add the store limit mechanism for separately controlling the speed of operators for each store
    • Support the waitingOperator queue to optimize the resource race among different schedulers
    • Support scheduling rate limit to actively send scheduling operations to TiKV. This improves the scheduling rate by limiting the number of concurrent scheduling tasks on a single node.
    • Optimize the Region Scatter scheduling to be not restrained by the limit mechanism
    • Add the shuffle-hot-region scheduler to facilitate TiKV stability test in scenarios of poor hotspot scheduling
  • Simulator
    • Add simulator for data import scenarios
    • Support setting different heartbeats intervals for the Store
  • Others
    • Upgrade etcd to solve the issues of inconsistent log output formats, Leader selection failure in prevote, and lease deadlocking
    • Develop a unified log format specification with restructured log system to facilitate collection and analysis by tools
    • Add monitoring metrics including scheduling parameters, cluster label information, and time consumed by PD to process TSO requests, Store ID, and address information.

TiKV

  • Support distributed GC and concurrent lock resolving for improved GC performance
  • Support reversed raw_scan and raw_batch_scan
  • Support Multi-thread Raftstore and Multi-thread Apply to improve scalabilities, concurrency capacity, and resource usage within a single node. Performance improves by 70% under the same level of pressure
  • Support batch receiving and sending Raft messages, improving TPS by 7% for write intensive scenarios
  • Support checking RocksDB Level 0 files before applying snapshots to avoid write stall
  • Introduce Titan, a key-value plugin that improves write performance for scenarios with value sizes greater than 1KiB, and relieves write amplification in certain degrees
  • Support the pessimistic transaction mode (Experimental)
  • Support getting monitoring information via HTTP
  • Modify the semantics of Insert to allow Prewrite to succeed only when there is no Key
  • Develop a unified log format specification with restructured log system to facilitate collection and analysis by tools
  • Add performance metrics related to configuration information and key bound crossing
  • Support Local Reader in RawKV to improve performance
  • Engine
    • Optimize memory management to reduce memory allocation and copying for Iterator Key Bound Option
    • Support block cache sharing among different column families
  • Server
    • Reduce context switch overhead from batch commands
    • Remove txn scheduler
    • Add monitoring items related to read index and GC worker
  • RaftStore
    • Support Hibernate Regions to optimize CPU consumption from RaftStore (Experimental)
    • Remove the local reader thread
  • Coprocessor
    • Refactor the computation framework to implement vector operators, computation using vector expressions, and vector aggregations to improve performance
    • Support providing operator execution status for the EXPLAIN ANALYZE statement in TiDB
    • Switch to the work-stealing thread pool model to reduce context switch cost

Tools

  • TiDB Lightning
    • Support redirected replication of data tables
    • Support importing CSV files
    • Improve performance for conversion from SQL to KV pairs
    • Support batch import of single tables to improve performance
    • Support separately importing data and indexes for big tables to improve the performance of TiKV-importer
    • Support filling the missing column using the row_id or the default column value when column data is missing in the new file
    • Support setting a speed limit in TIKV-importer when uploading SST files to TiKV
  • TiDB Binlog
    • Add the advertise-addr configuration in Drainer to support the bridge mode in the container environment
    • Add the GetMvccByEncodeKey function in Pump to speed up querying the transaction status
    • Support compressing communication data among components to reduce network resource consumption
    • Add the Arbiter tool that supports reading binlog from Kafka and replicate the data into MySQL
    • Support filtering out files that don’t require replication via Reparo
    • Support replicating generated columns
    • Add the syncer.sql-mode configuration item to support using different sql-modes to parse DDL queries
    • Add the syncer.ignore-table configuration item to support filtering tables not to be replicated
  • sync-diff-inspector
    • Support checkpoint to record verification status and continue the verification from last saved point after restarting
    • Add the only-use-checksum configuration item to check data consistency by calculating checksum
    • Support using TiDB statistics and multiple columns to split chunks for comparison to adapt to more scenarios

TiDB Ansible

  • Upgrade the following monitoring components to a stable version:
    • Prometheus from V2.2.1 to V2.8.1
    • Pushgateway from V0.4.0 to V0.7.0
    • Node_exporter from V0.15.2 to V0.17.0
    • Alertmanager from V0.14.0 to V0.17.0
    • Grafana from V4.6.3 to V6.1.6
    • Ansible from V2.5.14 to V2.7.11
  • Add the TiKV summary monitoring dashboard to view cluster status conveniently
  • Add the TiKV trouble_shooting monitoring dashboard to remove duplicate items and facilitate troubleshooting
  • Add the TiKV details monitoring dashboard to facilitate debugging and troubleshooting
  • Add concurrent check for version consistency during rolling updates to improve the update performance
  • Support deployment and operations for TiDB Lightning
  • Optimize the table-regions.py script to support displaying Leader distribution by tables
  • Optimize TiDB monitoring and add latency related monitoring items by SQL categories
  • Modify the operating system version limit to only support the CentOS 7.0+ and Red Hat 7.0+ operating systems
  • Add the monitoring item to predict the maximum QPS of the cluster (hidden by default)