Skip to content

Correlate scheduler admission and request admission observability #648

@eric-tramel

Description

@eric-tramel

Priority Level

Medium

Epic: #645
Depends on: #657, #635, #647
Related: #644, #641, #649, #660
Target branch: epic/645-async-scheduling while the epic is active.

Source of Truth

Specification details live in PR #658 under plans/645/, especially observability.md, architecture.md, contracts.md, benchmark-plan.md, and issue-map.md. This issue tracks the correlation slice and gates only.

Implementation Scope

Correlate scheduler admission and request admission without merging the two telemetry systems.

This issue owns:

  • RuntimeCorrelationProvider, CorrelatedRuntimeView, and the final plan-equivalent correlation context boundary;
  • async-safe context propagation across scheduled task execution;
  • model request start/complete timing from ModelRequestExecutor for joined timelines;
  • shared event naming/correlation-field conventions for scheduler and request telemetry;
  • benchmark/trace artifacts that show scheduler wait, request wait, provider execution, and completion in one view.

This issue must not put scheduler types into request-admission protocols or make either telemetry subsystem require the other.

Quality Gates

  • Scheduler admission events and request-admission events remain independently useful when the other subsystem is disabled.
  • Correlation can distinguish dependency wait, ready queue wait, scheduler admission wait, request admission wait, provider cooldown/rate-limit behavior, provider execution, and task completion.
  • Context propagation is isolated across concurrent tasks and reset correctly.
  • No prompts, completions, row values, secrets, raw provider payloads, or unbounded request IDs become metric labels.
  • Tests cover context propagation, set/reset behavior, concurrent isolation, disabled subsystems, attribute conversion, low-cardinality constraints, and sink failure isolation.

Validation

Run correlation tests, affected scheduler/request-admission instrumentation tests, make check, and at least one joined trace/JSON artifact per plans/645/benchmark-plan.md.

Metadata

Metadata

Assignees

No one assigned

    Labels

    planAgent-assisted development plantaskInternal development task

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions