Shared data/CN requires closed source dependency?

[Samrose-Ahmed]

From my understanding, currently to build starrocks using --enable-shared-data which allows building shared data/CN features, it will require using a thirdparty dependency that is not open source (the starlet/starcache libraries) which are vended as precompiled objects.

Essentially, as i understand, this makes the entire shared data feature closed source by default, since it is not currently possible to compile using both --enable-shared-data and --without-starcache (it seems --enable-shared-data/USE_STAROS=ON option requires startlet/starcache libraries).

As this a Linux Foundation Apache 2.0 licensed project, can you clarify this? Even if the starcache/starlet is not to be open source, I believe the shared data functionality of SR should be open source usable without any closed source third party dependency (e.g. even if requires using a dummy cache).

We can do some internal patching but I want to understand if this is just a build scripting oversight or what the intended behavior is here.

Thanks!

StarRocks version (Required)

• Any > 3.x

[alvin-celerdata]

@Samrose-Ahmed

Great question.

The current status is that when StarRocks is running in shared-nothing mode, there is no closed-source dependency. When it is running in shared-data mode, it will depend on StarOS and StarCache, which have not been open-sourced now. They are distributed in the binary format under Apache License V2, so there are no differences between shared-data and shared-nothing if you don't have customized modification requirements.

Before, when the shared-data mode was under development, we made it closed-source. As the shared-data mode becomes more and more mature, it is a good time to consider open-sourcing StarOS as well as StarCache.

Thank you for that question to take it back to our spot, we will consider it seriously. We will get back to you when we have some conclusions.