[GCS FT] Move GCS FT API from annotations to RayCluster CRD

[kevin85421]

Search before asking

• I had searched in the issues and found no similar feature requirement.

Description

Currently, users need to use an annotation ray.io/ft-enabled to enable GCS FT (doc). Move the GCS API from annotations to RayCluster CRD.

Some possible APIs can be something similar to Autoscaler.

enableGcsFaultTolerance: true
gcsFaultToleranceOptions:
  ... # maybe redis address / redis password ... etc

Use case

No response

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

[kevin85421]

/assign @rueian

[rueian]

👌

[rueian]

@kevin85421, I feel the enableGcsFaultTolerance is a bit redundant if we have the gcsFaultToleranceOptions already. Is there any concern so that we should have a dedicated boolean field for enabling the feature? Except that, I think the gcsFaultToleranceOptions field should be like this:

type RayClusterSpec struct {
	...
	GcsFaultToleranceOptions *GcsFaultToleranceOptions `json:"gcsFaultToleranceOptions,omitempty"`
}

type GcsFaultToleranceOptions struct {
	ExternalStorageNamespace string         `json:"externalStorageNamespace,omitempty"`
	RedisAddress             string         `json:"redisAddress"`
	RedisPassword            *RedisPassword `json:"redisPassword,omitempty"`
}

type RedisPassword struct {
	Value     string `json:"value,omitempty"`
	ValueFrom *corev1.EnvVarSource `json:"valueFrom,omitempty"`
}

The RedisAddress is the only mandatory field and we mimic the corev1.EnvVar for the RedisPassword so that we can copy it to the container env fields directly, preserving the valueFrom configuration.

[kevin85421]

SG