Weekly update in TiDB

2017-11-13

Last week, we landed 45 PRs in the TiDB repositories.

Added

• Support more SQL modes in TiDB:
  • the NO_UNSIGNED_SUB sql_mode
  • the REAL_AS_FLOAT sql_mode
  • the PIPES_AS_CONCAT sql_mode
  • the high_not_precedence sql_mode
  • the ONLY_FULL_GROUP_BY sql_mode
• Parse more privilege types like RELOAD, EVENT and so on.
• Support part of window function AST.

Removed

• Remove joinBuilder.
• Remove resolver.go.

Fixed

• Return error instead of panic if a subquery in JOIN ON condition.
• Set the error to be undetermined error if not recovered from an RPC error.
• Fix the issue that DDL is always running and saves the reorg doneHandle regularly.
• Let the in clause handle more cases.

Improved

• Use types.Row to write the result set and prepare to use Chunk.
• Support the alter table add column(col_name column_definition) syntax.
• Build and use the Count-Min sketch.
• Stop reusing the Executor in IndexLookUpJoin and remove doRequestForDatums().
• Begin OpenTracing from dispatch() and change the interface to Execute(ctx, sql) to avoid too many noises.
• Remove the read-only statement from transaction auto retry.
• Make the batch size of index join increase slowly.
• Let select push across projects to support generated column index.
• Support insert into from selectStmt that has brackets.
• Remove the type assertion on types.DatumRow.
• Improve hash join to support all the join types.
• Implement the Count-Min Sketch.
• Enable OpenTracing for the TableReader, IndexReader, IndexLookup executor.

Weekly update in TiSpark

Added

• Add R language support.
• Add Python language support.
• Add Index Read Support and related task concurrent read.

Improved

• Improve integration tests and frameworks.

Fixed

• Fix racing condition in thread pool allocation.

Weekly update in TiKV

Last week, We landed 18 PRs in the TiKV repositories.

Added

• Support splitting table
• Support resolving multiple locks in a batch.
• Use size as a factor of score when scheduling.
• Support OpenTracing for pd-client.
• Support set namespace for meta.
• Support checking the region stats using API.

Fixed

• Verify before destroying a peer.
• Fix the LIKE behavior for coprocessor.
• Fix the do_div_mod bug.
• Fix a nil pointer dereference bug in pd-client.
• Clear hot region related gauges if a region is not hot any more.

Improved

• Add meta decoding for tables.
• Better region key output API.
• Unify ApplyContext and ExecContext.
• Use recover_safe! to hide the panic stack trace.
• Remove the pending peers before adding a new peer.

New contributers (Thanks!)

• wudi
• xiaojian cai