ray-project / ray

Ray is a unified framework for scaling AI and Python applications. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.
https://ray.io
Apache License 2.0
33.28k stars 5.63k forks source link

[autoscaler][gcp] wrong values for scheduling in example gcp cluster yaml files #46248

Open guoqiao opened 3 months ago

guoqiao commented 3 months ago

What happened + What you expected to happen

In python/ray/autoscaler/gcp/*.yaml files, the scheduling field is configured as following in a few places:

scheduling:
  - preemptible: true
  - onHostMaintenance: TERMINATE

The equivalent json version will be:

"scheduling": [
    {"preemptilbe": true},
    {"onHostMaintenance": "TERMINATE"}
]

This is wrong, according to GCP Doc:

"scheduling": {
    "onHostMaintenance": enum,
    "automaticRestart": boolean,
    "preemptible": boolean,
    ...
}

Summary: for gcp, scheduling should be object/dict, but misconfigured to be a list of objects in examples.

Versions / Dependencies

all versions up to lastest (2.31)

Reproduction script

in repo root:

cd python/ray/autoscaler/gcp/
grep -A 3 "scheduling:" *.yaml

Issue Severity

Low: It annoys or frustrates me.

anyscalesam commented 3 months ago

Good catch @guoqiao > can you submit a PR and we'll shepherd it through

guoqiao commented 2 months ago

@anyscalesam PR created ^. Thank you!