TiDB Weekly

News & blog posts

High read latency and network traffic are common issues for a multi-region architecture. At TiDB Hackathon 2019, a team won 2nd place by reducing multi-region read latency and network traffic by 50%.

Last week, we published a post that introduces how the team reduced the read latency and network traffic by half for a multi-region architecture. If you have similar problems, we hope this post will help you.

The full post is here:

How We Reduced Multi-region Read Latency and Network Traffic by 50%

Last week, we also published the third (and last, for now) post in a series on remote work by Nick Cameron. This post talks a bit about some of the practicalities, specifically around work/life balance.

The full post is here:

Remote Work - Part 3

Community

New Committer

Congratulations to TennyZhuang as the new TiKV Committer!

Before becoming a Committer, TennyZhuang had been reviewing PRs for the TiKV community as a Reviewer, and gradually got familiar with the Coprocessor module. He made the following contributions to the TiKV community:

• Greatly improved the performance of LIKE(): tikv/tikv #5866
• Greatly improved the performance of IN(): tikv/tikv #6000
• Refined TiDB's time format to be 8 bytes as TiKV: tikv/tikv #6418
• Participated in implementing the collation for TiKV: tikv/tikv #6592
• Accepted the RPN expression to support complex expressions: tikv/tikv #6313
• Implemented the metadata framework: tikv/tikv #5993

Thank him for his contributions to the community!

New contributors

tikv:

• Poytr1

docs:

• oraluben
• qqbuby

tidb-operator:

• mightyguava

chaos-mesh:

• oraluben

Call for participation

• Issues to solve for TiDB
• Issues to solve for TiKV

Updates of the week

Progress

Last week, we landed 52 PRs in the TiDB repository, 1 PR in the TiSpark repository, and 43 PRs in the TiKV and PD repositories.

Critical PRs

TiDB:

• Support defining decimal for the virtual table in information_schema
• Add more diagnosis rules to check some metrics thresholds
• Make CLUSTER_SLOW_QUERY support querying slow logs at any time
• Specify the time range via the TIME_RANGE hint for the metrics or inspection tables
• Fix the wrong result of BatchPointGet when the plan cache is enabled
• Consider collations when calculating hash values in HashJoin
• Reload the TiKV or TiDB cluster TLS for every newly established connection
• Add a check to reject changing some configuration items during the runtime
• Implement the non-stream physical scan lock for the GC worker
• Resolve the left secondary lock to avoid the dead loop

TiSpark:

• Fix the partition pruning

TiKV and PD:

• Add the stack size and the maximum task number to the unified read pool
• Add the Cmd observer to Raftstore
• Implement non-streaming physical_scan_lock to the GC worker
• Make the Vec-TonN executor aware of the collation of ordering expressions in Coprocessor
• Migrate TiKV to abstract TablePropertiesCollection types
• Add the component sub-command
• Improve the performance of leader transferring by returning multiple operators in batches
• Modify TopN to support the multidimensional load