TiDB Weekly

News & blog post

A community user at Zhihu.com (the Quora of China) shared a detailed TiDB case study:

Zhihu is the Quora of China with 220 million registered users, and it has 30 million questions with more than 130 million answers. With approximately 100 billion rows of data accruing each month and growing, this number will reach 3 trillion in two years. They can foresee great challenges in scaling their backend system while maintaining good user experience. With TiDB, they managed to harness the power of data and ensure good user experience because they get 99th percentile response time over these 1.3 trillion rows of data was about 25 ms, and the 999th percentile response time was about 50 ms.

The case study shows why they chose TiDB, how they are using it, what they learned and best practice, and some thoughts about the future with TiDB.

Full story is here:

Lesson Learned from Queries over 1.3 Trillion Rows of Data Within Milliseconds of Response Time at Zhihu.com

Community

New contributors

tikv:

• jadireddi
• aknuds1

parser:

• hey-kong
• yangwenmai
• yufeiminds

rust-prometheus:

• brndnmtthws

Call for participation

TiDB:

• Vectorize all the built-in functions
• Issues to solve

TiKV:

• Issues to solve

Updates of the week

Progress

TiDB:

Last week, we landed 55 PRs in the TiDB repository.

• Release TiDB v3.0.3
• Add a test framework for the functions of vectorized expression evaluation
• Enable the vectorized expression evaluation by default
• Support dividing a Region into multiple Regions

TiSpark:

• Release TiSpark v2.1.5 and v2.2.0

TiKV and PD:

Last week, we landed 29 PRs in the TiKV and PD repositories.

• Release TiKV and PD v3.0.3
• Fix the issue that a wrong value might be returned when the number of keys is 0
• Fix the issue that a Region does not merge to the sibling with smaller size
• Support updating the lock TTL through heartbeats when executing a large transaction
• Support checking configuration by running the config-check command-line option

Critical PRs

TiDB:

• Support the vectorized bitmap operations for nulls in Column
• Fix the modify column operation on the bit column
• Add a test framework for the functions of vectorized expression evaluation
• Eliminate the redundant projection
• Support dividing a Region into multiple Regions
• Support index_lookup_merge_join in the physical plan
• Support IndexNestedLoopHashJoin
• Do not send TSO requests when point get uses maxUint64 as its transaction timestamp
• Enable the vectorized expression evaluation by default
• Split AVG to COUNT and SUM for the TableReader Coprocessor task

TiKV and PD:

• Use the new approximate keys function when the result is 0
• Add the txn_heart_beat API
• Fix the issue that a Region does not merge to the sibling with smaller size
• Add the backup service
• Add the config-check flag for tikv-server
• Add the config-check flag for pd-server
• Continue to use OpenSSL
• Fix a thread-safe bug and improve code
• Fix the precision loss while unflattening Duration in codec

Proposals

TiDB:

• Add Contributor Covenant Code of Conduct
• Update the Github workflow

Upcoming TiDB event

Paper reading