Skip to content

feat(aws-healthomics): add steering for CPU/GPU resource fallback via… - #193

Open
alxawan wants to merge 1 commit into
kirodotdev:mainfrom
alxawan:main
Open

feat(aws-healthomics): add steering for CPU/GPU resource fallback via…#193
alxawan wants to merge 1 commit into
kirodotdev:mainfrom
alxawan:main

Conversation

@alxawan

@alxawan alxawan commented Sep 1, 2026

Copy link
Copy Markdown

feat(aws-healthomics): add steering for CPU/GPU resource fallback via…

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

- `docker` and `maxRetries` are shared, not per-profile — set them once at the top level as
you would for any other task. See the [Workflow Development SOP](./workflow-development.md#task-parameters).

## Example

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add equivalent Nextflow and CWL examples if these are supported

- A value as low as 1 minute is accepted, but only allows roughly one capacity-search cycle
before advancing — this is USUALLY too aggressive for anything but a deliberately fast
first-choice profile.
- For scarce accelerator types (e.g. `nvidia-l40s`), RECOMMEND >= 20 minutes so the profile

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How would an agent know what is scarce? At any given time any GPU could be relatively plentiful or scarce? It will also depend on the region. Some regions/ AZs only have newer GPU types and these might not be scarce in those regions?

- For scarce accelerator types (e.g. `nvidia-l40s`), RECOMMEND >= 20 minutes so the profile
gets a fair chance before advancing (see rule 9 above).
- The sum of every profile's wait window SHOULD stay well under the run's overall timeout
(maximum 7 days) — a long fallback list with generous per-profile timeouts can otherwise

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Default maximum is 7 days.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants