Skip to content

P3: watchdog "marking idle" conflates idle with blocked-on-send #894

@khaliqgant

Description

@khaliqgant

Source: reliability spec cloud/docs/specs/agent-relay-reliability-spec.md §P3. Filed from spec-triage on PR #861.

Problem

watchdog: no PTY output for 120s — marking idle fired for an agent that was not idle — it was actively retrying a failing send (see #892). "Idle" conflated "no terminal output" with "no work / nothing to do", masking a blocked-on-send deadlock.

Required

Distinguish idle (nothing queued) from stalled (blocked on send / long op). The watchdog state must not report "idle" when the agent has pending outbound work it cannot deliver.

Scope

Broker watchdog semantics (Rust). Coupled with #892 (delivery state is the signal that disambiguates).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions