2017-06-05-tidb-weekly.md

layout	title
post	Weekly Update

Weekly update in TiDB

Last two weeks, we landed 53 PRs in the TiDB repositories.

Added

• Support using After and First to specify column position in the Alter Table Statement.

• Support the password builtin function.

• Support the inet6_ntoa builtin function.

• Support the Extract and Unquote function for Json.

• Support json_{set/insert/replace} and json_merge for Json.

• Support batched Index Lookup Join.

Fixed

• Fix a goroutine leak problem.

• Fix a bug in double read executor.

• Fix a bug for type inferrer of Bit literal value.

• Fix a bug about context cancellation.

Improved

• Refactor expression evaluation:

  • Refactor the cast function.
  • Add the GetTypeClass() interface in Expression.

• Update stored password in mysql.user table to make it consistent with MySQL.

• Speed up the DDL process:

  • Use etcd to synchronise schema version instead of waiting two leases.
  • Enable DDL speed up.

• Add the references_priv column in mysql.user.

• Close DDL worker gracefully when shutting down the TiDB server.

Weekly update in TiKV

Last two weeks, We landed 33 PRs in the TiKV repositories.

Added

• Introduce gRPC, see 1850, 1865, 1868, 1879.
• Report the written keys to PD for better scheduler.
• Balance the hot regions by peer and leader together.
• Record the cluster bootstrap time.
• Add the state for operators.
• Add the metrics for balance.
• Add the disaster recovery tool for PD.
• Output the allocation statistics when receiving SIGUSR1.
• Add the direct IO configuration for RocksDB.

Fixed

• Check the cluster ID when re-connecting to PD.
• Reuse the timer to avoid spawning thread frequently.
• Remove the realtime signal handler.
• Prevent accessing nil store when recovering cluster.
• Check the environment TZ setting.
• Use the forward mode for scanning data in GC command.
• Check the last apply index to determine handing snapshot.

Improved

• Increase the region size from 64 MB to 96 MB.
• Split total region peer cache.
• Use FlatMap instead of HashMap to improve performance.
• Modify the default compression in RocksDB.