CI: 2023 12 01
Zack Galbreath edited this page Dec 1, 2023
·
1 revision
- Aashish Chaudhary
- Caetano Melone
- Jacob Nesbitt
- Massimiliano Culpo
- Dan LaManna
- Mike VanDenburgh
- Ryan Krattiger
- Scott Wittenburg
- Tammy Dahlgren
- Todd Gamblin
- Zack Galbreath
- We discussed Caetano's research into dynamic resource Allocation for our Spack CI builds. We are excited to implement his recommendations to improve the efficiency of our CI system! Our plan to do so is as follows:
- Use Caetano's results to update the hardcoded
KUBERNETES_MEMORY_REQUEST
values in [share/spack/gitlab/cloud_pipelines
]. - Update our
generate
jobs to dynamically receive memory request values from a new service, falling back to the hardcoded values above if this service is unavailable.
- Use Caetano's results to update the hardcoded
- We are investigating how to enable/disable stacks without creating a PR to spack. (PR #41381)
- Update resource requests based on Caetano's findings
- Continue pursuing "costs per job" metric
- Content-addressable tarballs