xxchan

Just for fun.

This blog series is my personal comments about (part of) the development of RisingWave.

Please take it as an unofficial and no-promise supplement.

Features Updates 🌟

NULLS {FIRST | LAST}

feat(common): support NULLS {FIRST | LAST} by richardchien · Pull Request #8485

NULL s can be very tricky in SQL. Do you known you can specify its ordering?

> select * from t order by x /* nulls last */; 
+--------+
| x      |
|--------|
| 1      |
| 2      |
| <null> |
+--------+

> select * from t order by x nulls first;
+--------+
| x      |
|--------|
| <null> |
| 1      |
| 2      |
+--------+

It was tricky to support because previously in RisingWave, we only considered ascending/descending ordering, but not NULL s’ ordering. That caused inconsistent NULL s ordering around the system. When it comes to SQL query results, the NULL values were sometimes the largest, sometimes the smallest.

This also blocked our progress to test RisingWave against other database’s test suites, because in PostgreSQL, NULL s are the largest by default when compared against non-NULL values, while in SQLite and DuckDB they are the smallest by default. With inconsistent ordering of NULL s and no support for NULLS { FIRST | LAST } in order-by clauses, we can neither align with the behavior of one of these databases, nor specify a fixed NULL ordering in test queries to align behaviors of all databases.

Recently, we did a thorough refactoring to force all components in our system to use an unified struct type representing ordering, OrderType . As a result, this week we were finally able to implement NULLS { FIRST | LAST } for order-by clauses at ease, by simply adding a new field, NullsAre , to OrderType , specifying whether NULL s are largest or smallest. By setting Largest as default value of NullsAre , we achieved the same default ordering behavior as PostgreSQL.

Rename relations

feat: support alter rename relations including table/mview/view/sink/index by yezizp2012 · Pull Request #7745

As I mentioned earlier, we are working on DDL support ( ALTER TABLE ), which is quite tricky. We’ve already done quite some work for supporting ADD/DROP COLUMN . Here’s another DDL: renaming relations is supported. This might be slightly easier than ADD/DROP COLUMN , but there are still some tricky parts to consider carefully, like updating all related relations all at the same time.

For example, you can use the command ALTER TABLE t_1 RENAME TO t_2 to rename a table and its related relations will be modified recursively, which you can check with command such as SHOW CREATE MATERIALIZED VIEW mv_x . There is no effect on the stored data.

Performance Optimizations 💪

Bushy tree join ordering

feat: Bushy tree join ordering by KveinAxel · Pull Request #8316

RisingWave is a streaming processing system that aims to provide real-time low latency for our users. By reducing the depth of the join tree, we can effectively minimize latency.

See the illustration below:

To reduce the distance that barriers from join inputs need to travel to the join output, we convert a left-deep tree with a height of 3 into a bushy tree with a height of 5.

https://user-images.githubusercontent.com/9352536/202991793-664ea3f9-3838-4e5f-af6c-e5416140ca40.png

https://user-images.githubusercontent.com/9352536/202991855-998a6d28-a366-4120-8765-be3d5de20474.png

Rusty stuff 🦀️

We ❤️ Rust! This section is about some general Rust related issues.

zld

Cannot run risingwave binary on Mac OS M1 (when linked by zld ) · Issue #8608

@ahmedriza reported that he cannot run the binary on his Macbook M1 due to “symbol not found”.

$ ./target/debug/risingwave
dyld[14116]: symbol not found in flat namespace (__ZN15protobuf_native2io23DeleteCodedOutputStreamEPN6google8protobuf2io17CodedOutputStreamE)
Abort trap: 6

After some investigation, he found that it’s because he used zld in his global cargo config, and RisingWave has a repository-level cargo config that uses lld. Although it is still not clear whether the error is caused by zld or the conflict between the cargo configs, he could solve the problem by just using lld.

This is also one of the reasons why I enjoy open source: hackers are very willing to raise issues and even analyze and solve problems themselves. I can also learn a lot from their thoroughly analysis. From this issue, I learned:

  • How cargo config ($HOME/.cargo/config.toml, and /projects/.cargo/config.toml) works in more detail
  • There’s another linker zld. (But it’s already deprecated in favor of lld .)

New Contributors

(Of course, @ahmedriza is also a great new contributor!)

feat: Add support for array_length function in psql by kamalesh0406 · Pull Request #8636

This week we have another first-time contributor @kamalesh0406. He said “This is my first time writing rust code” 😲.

So it seems RisingWave’s good first issues are a good way to learn and practice Rust. 😄 Don’t hesitate and just join us if you are also interested in the development of an open source database system, or Rust!

Finally, welcome to join the RisingWave Slack community.

So much for this week. See you next week! 🤗

You May Also Enjoy