Skip to content

[Feature]: Use fixed polling interval for vLLM instance readiness #455

@waltforme

Description

@waltforme

Feature Area

Other

Problem Statement

While a vLLM instance starts up, the dual-pods controller polls /is_sleeping through the work queue's exponential backoff (5ms growing to 20s). Early retries are wastefully frequent; later ones add unnecessary latency. A fixed ~5s interval would be more appropriate.

From #443 (comment) (point 4).

Proposed Solution

TBD

Alternatives Considered

No response

Willingness to Contribute

Yes, I can submit a PR

Additional Context

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions