Skip to content

Migrate CI/CD from GCP self-hosted runners to Modal #464

@baogorek

Description

@baogorek

Summary

Replace the self-hosted GCP VM runners with Modal serverless compute for CI/CD workflows.

Motivation

  • Cost reduction: Pay-per-use vs always-on VM (~$180/month → ~$60-80/month)
  • No infrastructure management: No VM maintenance, spot instance handling
  • Parallelization potential: Local area builds can run 435 CDs concurrently

Changes Required

  • Create Modal app with data_build.py for dataset builds
  • Create Modal app with local_area.py for local area publishing
  • Update reusable_test.yaml to trigger Modal instead of self-hosted
  • Update local_area_publish.yaml to trigger Modal
  • Add Modal secrets to GitHub (MODAL_TOKEN_ID, MODAL_TOKEN_SECRET)
  • Create Modal secrets (huggingface-token, gcp-credentials)
  • Validate in production
  • Decommission GCP self-hosted runner VM

Testing

Successfully tested modal run modal_app/data_build.py --test-lite --no-upload locally.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions