Week 12

(August 19, 2025 – August 25, 2025)

Scheduler Agent Work

This week, despite exploring multiple approaches, the scheduler is still not fully functioning as expected. This has been the main challenge and blocker in pushing other dependent services forward.

Key Challenges Faced:

• Path mismatches related to fo_scheduler install paths, which created uncertainty in the setup.
• Deciding between continuing with the existing scheduler implementation or starting fresh from scratch to avoid legacy issues.
• Lack of clarity on evaluation metrics, e.g., whether to focus on job queue performance, agent execution times or resource usage in a containerized setup.

At this point, I am leaning towards rebuilding the scheduler from scratch as it seems like the more sustainable path forward, but I am awaiting mentor guidance to ensure alignment.

Overall, while the scheduler has not yet reached a running state, the time invested has helped surface important blockers and open questions that must be addressed before progress can be made.

Open Questions / Confusions

During my work this week, I noted down a couple of key confusions that need mentor input:

📌 Based on my research in the wiki and older methods for scheduler, what is the correct handling of install path mismatches for fo_scheduler?

📌 What metrics should we prioritize (e.g., job queue performance, task throughput, resource usage) to evaluate the scheduler in a containerized microservices setup?

Resolving these will help me focus my efforts in the upcoming weeks and avoid wasting time in the wrong direction.

Kustomization & Infrastructure

While the bulk of my time was on the scheduler, I continued revisiting my Kustomization setup from Week 11. This structure will be useful once the scheduler stabilizes, as it ensures we can maintain both dev and prod overlays with cleaner configurations.

I also reviewed some Dockerfiles, but this work will be resumed in later weeks once the scheduler issue is clarified.

Collaboration and Mentorship

• I actively discussed progress and blockers with Avinal. We explored different approaches for the scheduler (Reuse vs. Rebuild). The consensus so far is to carefully evaluate before committing to a rebuild.
• Omar is still pending to respond, but his past GSoC experience could bring valuable input.

Meeting 1

(August 20, 2025)

The call did not take place due to the unavailability of Avinal.

Meeting 2

(August 21, 2025)

In the weekly community meeting call, I shared my updates:

• Tried different approaches but it is still not working as expected
• Dedicated time to make the Scheduler pod run
• Raised confusions regarding install paths and evaluation metrics
• Suggested that a fresh rebuild of the scheduler might be more reasonable moving forward
• Highlighted that I am still seeking guidance to refine my direction

I concluded my update by handing over to Avinal for further discussion.