Files
open-wc/coordination/claims/M03-RND-SCHEDULER-001.md
T

88 lines
3.5 KiB
Markdown

# M03-RND-SCHEDULER-001 — Bounded render budget scheduler
<!-- OPENWC_CLAIM:M03-RND-SCHEDULER-001:sindo-main-codex:2026-07-18 -->
## Ownership
- Target: M03
- Program: RND
- Owner/Agent ID: sindo-main-codex
- Branch: `work/sindo-main-codex/m03-render-budget-scheduler`
- Lease expires UTC: 2026-07-18
- Integrator: M03 milestone integrator
## Outcome
Extract the streamer's per-frame main-thread operation quotas into a scene-free
`RenderBudgetScheduler`, preserving existing queue priority and quota values and
stopping permit issuance after renderer teardown cancellation.
## Non-goals
- Move queue, worker-task, cache, Node, Resource or RID ownership out of the streamer.
- Change queue ordering, concurrency limits, exported defaults or visual behavior.
- Add a generic job framework, priority graph, time-based adaptive budgeting or dependency.
- Change cache formats, renderer profiles or target milestone status.
## Paths
- Exclusive: `src/render/streaming/render_budget_scheduler.gd`,
`src/tools/verify_render_budget_scheduler.gd`,
`docs/modules/render-budget-scheduler.md`, this claim
- Shared/hotspots: `src/scenes/streaming/streaming_world_loader.gd`,
`docs/modules/world-renderer.md`, `docs/modules/README.md`, `RENDER.md`,
`targets/03-renderer-facade.md`
- Generated/ignored: `.godot`, native DLL, generated cache resources and
proprietary renderer corpus
## Contracts and data
- Input: immutable per-frame map from documented operation lane IDs to
non-negative operation limits
- Output: boolean permits and remaining/consumed diagnostic snapshots
- Lifecycle: `begin_frame` resets permits; `cancel` permanently rejects permits
for the current scheduler instance
- Queue contents, ordering, task concurrency and schema/cache versions: unchanged
- Consumer: `StreamingWorldLoader` main-thread queue/finalize drains
## Dependencies
- Requires: integrated M03 facade and streaming planner on master `650d584`
- Blocks: renderer performance regression gate and per-service extraction
- External state: none; tests use synthetic lane limits and no world assets
## Verification
- Commands: dedicated scheduler verifier, facade/planner/focus/coordinate
regressions, renderer manifest/dry-run and repository gates
- Fixtures: shared lane priority, independent lanes, zero/negative limits,
frame reset, unknown lane and cancellation
- Fidelity evidence: operation defaults and loop order are captured before
migration; all seven renderer checkpoint plans remain valid
- Performance budget: permit checks are constant-time dictionary operations and
create no Node, Resource, worker task, cache entry or RID
## Documentation deliverables
- Inline API docs for scheduler lifecycle, permits and diagnostics
- New module specification with input/output table, data-flow, sequence and
lifecycle diagrams
- Updated world-renderer flow/ownership/source map, module registry and renderer notes
## Simplicity and naming
- Important name: `RenderBudgetScheduler`
- Simplest approach: one small main-thread service with named quota lanes
- Rejected complexity: generic job graph, callbacks, queue ownership, adaptive
timing controller, service container or third-party dependency
- Unavoidable complexity: chunk removal and creation intentionally share one
lane so removal retains priority while consuming the same historical budget
- Measured optimization evidence: no optimization claim; exact bounded extraction
## Status
- State: active
- Done: claim and extraction contract defined
- Next: implement scheduler, migrate existing bounded loops and add verification/docs
- Blocked by: