Skip to content

lcm: one summarizer timeout parks a session until the summarizer binding changes #4522

Closed

Description

@ScriptedAlchemy

Observed

While fixing the #4518 test flake (PR #4521): when the configured summarizer (e.g. cursor-agent) exceeds its timeout once, the LCM convergence queue parks the session as unavailable with cursor_agent_unavailable (failure count 0). That state is requeued only when the summarizer binding changes, so a single transient timeout (a loaded machine, a slow first start) leaves the session without summaries indefinitely.

Expected

A summarizer timeout is a retryable failure: requeue with bounded backoff (and count it), and park as unavailable only after repeated failures or when the summarizer is genuinely missing, reporting the reason in status.

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions