Weekly update in TiDB

2017-09-25

Last week, we landed 63 PRs in the TiDB repositories.

Added

• Use new expression framework by default.
• Support the DOT explain format.
• Support the syntax for EXPLAIN FORMAT = stringlit
• Support the TIME/TIMESTAMP literal

Removed

• Remove expression/typeinfer.go entirely.
• Abandon the selection controller.

Fixed

• Roll back the ID allocator when a transaction fails to commit.
• Fix the returned column length of all the SHOW statements.
• Fix the Navicat for MySQL compatibility issue of the SHOW CREATE TABLE statement.
• Fix a bug in GC worker.
• Fix a bug of merge JOIN/stream aggregation and ORDER BY.
• Support the aggregation function that contains an aggregation function.
• Abort an unsafe transaction subject to GC command
• Fix a bug in parseDatetime.

Improved

• Refactor the aggregation to reduce memory usage.
• Use continuous-value assumption to estimate cost.
• Speed up the add index operation.
• Don’t return the killed session when SHOW PROCESSLIST.
• Refine the return type, charset and collation for the get variable expression.
• First step to make parser to be totally compatible with MySQL.
• Add real tables for global/session status in the performance schema.
• Implement the ANALYZE columns pushing down.
• Rewrite the builtin function MOD
• Use the Goroutine pool to avoid runtime.morestack.
• Remove the usage of “TypeClass” in:
  • builtin_arithmetic.go
  • builtin_string.go
  • expression.go
  • builtin_cast.go
  • builtin_math.go
  • builtin_op.go

Weekly update in TiSpark

Fixed

• Fix the Region client error in handling for store no match and leader switch
• Fix an issue in PD Client leader switch.

Weekly update in TiKV

Last week, We landed 30 PRs in the TiKV repositories.

Added

• Add benchmarks for JSON binary codec.
• Implement fmsketch for coprocessor.
• Add the SplitRegion API.
• Synchronize the Raft log when necessary.
• Add the debug framework.
• Add the namespace checker for PD.
• Support setting the store state for PD.
• Support initial schedulers’ name from the config file.
• Add the table namespace classifier for PD.
• Support scheduling based on regions’ read flow for PD.
• Support getting region’s namespace in PD.

Fixed

• Fix setting a nonempty store to the tombstone state.

Improved

• Replace opp_neg with overflowing_neg.
• Enhance synchronizing the WAL log.
• Add metrics for transactions.
• Use 96MB as the default size for regions.
• Use counters instead of histogram when statistics scan details.
• Fix resource downloading and extracting when bench JSON.
• Add more statistics for RocksDB.
• Update the futures.
• Refactor the cluster information for PD.

New contributors (Thanks!)

• David Ding
• Hu Ming
• OuYang Jin
• Liang SHANG