KernelCI Weekly Newsletter (Week 50)

KernelCI Weekly Newsletter (Week 50)

[Denys Fedoryshchenko]

**Summary**

The issues with KCIDB submissions have been largely resolved, though
KCIDB is still experiencing minor problems with large batches (should
not cause data loss). We continue to monitor the situation closely.

This week, we also began planning a redesign of the storage subsystem.
The goal is to improve performance and reliability while providing
artifact storage for both KCIDB and partner CI systems. Since these two
subsystems require significant work, most other features are on hold
until their completion.

**Known Issues**

- `/latest/docs` and `openapi.json` are currently returning a 500
error. We are investigating the issue.

### Key Updates

- **Storage Subsystem Design Proposal**  
  A design proposal for the storage subsystem has been drafted:  
  [Storage Subsystem Design
Proposal](https://gist.github.com/nuclearcat/2d34f8266ccd1b8356bd993eadcf2eed
)  
  Feedback and suggestions from the community are welcome at the weekly
meeting and via the mailing list.

- **New Pub/Sub Subsystem Design**  
  Work is underway on a new Pub/Sub subsystem to address fundamental
issues with the current implementation, including non-guaranteed
message delivery and the lack of load-balancing features.  
  - The new subsystem will use Redis Streams as its foundation.  
  - It will be backward-compatible, requiring no API changes, enabling
seamless migration for clients.

- **Experimental Features in kci-dev**  
  The `kci-dev` team has started implementing experimental features for
retrieving results from the Dashboard API. Details can be found here:  
  [GitHub Issue #57](https://github.com/kernelci/kci-dev/issues/57)

- **Re-enabling ARM64 QEMU Support**  
  Efforts are ongoing to re-enable ARM64 QEMU support.

- **Buildroot Rootfs Fixes**  
  Long-pending fixes for Buildroot rootfs (pending for two years) are
being addressed.

- **Dashboard Improvements and Fixes**  
  As always, continuous improvements and fixes are being made to the
Dashboard.

Re: KernelCI Weekly Newsletter (Week 50)

[Guillaume Tucker]

Hello,

Thanks Denys for the comprehensive weekly news.

On 11/12/2024 10:53 pm, Denys Fedoryshchenko wrote:
> - **New Pub/Sub Subsystem Design**
>    Work is underway on a new Pub/Sub subsystem to address fundamental
> issues with the current implementation, including non-guaranteed
> message delivery and the lack of load-balancing features.
>    - The new subsystem will use Redis Streams as its foundation.
>    - It will be backward-compatible, requiring no API changes, enabling
> seamless migration for clients.

This looks similar to the Task concept which was introduced in Renelick 
after the initial fork, see some WIP documentation here:

https://gtucker.gitlab.io/renelick/tasks/

Some polishing is still needed but the main concept has now been 
stabilised.  Basically they rely on message queues which guarantee each 
message to be delivered to exactly one consumer.  Any API client can 
schedule a task, typically this is done by an orchestrator service, then 
the task object will need to be accepted by a single compatible 
scheduler which will be responsible for executing it.  If you have 
multiple instances of the same scheduler, it balances the load between 
them by design.  It also makes it possible to implement retries and 
other features related to task management.

By contrast, Pub/Sub events aren't guaranteed to be delivered since 
they're not destined to any particular recipient.  They enable a 
decoupled, modular architecture where you can add and remove subscribers 
without changing anything in the existing system.

Let me know if you want to discuss it further, also I'll be covering 
this in an upcoming blog post.

Guillaume

Re: KernelCI Weekly Newsletter (Week 50)

[Guillaume Tucker]

On 11/12/2024 10:53 pm, Denys Fedoryshchenko wrote:
> - `/latest/docs` and `openapi.json` are currently returning a 500
> error. We are investigating the issue.

Not sure if you've fixed this already but I've hit the same issue
with Renelick.  The workaround I've found was to downgrade Beanie
to v1.27.0 since v1.28.0 which was released in December appears
to be breaking Pydantic validation.  The latest Beanie v1.29.0
also shows the same problem, and upgrading Pydantic didn't solve
the error either.

I've created this GitLab issue to keep track of it, aiming to
have it resolved by the end of next week:

   https://gitlab.com/gtucker.io/renelick/-/issues/133

Hope this helps,
Guillaume