Repository navigation
id(world): F12.12 many worlds per node: 1,000 idle / 100 active with a per-world memory budget #557
Description
What and why
Lazy open, idle eviction with checkpoint and a bounded memory budget per world, measured against 1,000 idle and 100 active (20 ev/s, p99 < 50 ms) targets with the F11.8 harness. Lazy open, eviction and per-guest memory caps exist, but nobody has measured the targets and a world's total memory is not bounded.
Source
.opencode/plans/phase-12-platform-flexibility.md:51.opencode/plans/phase-12-platform-flexibility.md:19-20
Current status
Partial. Mechanisms exist from Phase 9/10: pkgs/id/src/world_hub.rs:7-15 (lazy single-flight open, session and open-world ceilings, idle eviction), and a per-guest linear memory cap of 16 MiB (pkgs/id/src/sandbox.rs:25-36). The per-world total memory budget with refusal tests landed in #4628. The F11.8 many-world measurement now exists; its p99 is 5,650.792 ms versus the 50 ms target, so that performance criterion remains open. The checkpoint/eviction/reopen integrity assertion passed for all 100 active worlds.
Acceptance criteria
- A per-world total memory budget is enforced, with refusal tests
- 1,000 idle and 100 active worlds (20 ev/s each, p99 < 50 ms) are measured with the F11.8 harness and the results recorded (measured p99: 5,650.792 ms; target missed)
- Idle eviction checkpoints and reopens without loss under that load
Size
L
Prerequisites
- id(ci): F11.8 benchmark harness with committed baselines #539 — D1-008 F11.8 Benchmark harness with committed baselines
Activity
dezren39 commentedon Oct 26, 2026
Intake review
Fleet-monitor intake pass (2026-10-26). Body already carries a full
## Acceptance criteriasection, so this pass is label normalization plus a liveness check — no body edit.Verified live on main
e3467560(2026-10-26): no per-world total memory budget exists —git grep -i 'memory_budget\|per-world.*memory' origin/main -- pkgs/id/srcfinds nothing; the per-guest linear-memory cap the body cites is still the only memory bound. The measurement half is also untouched (no harness run recorded). Still live; the body's "Partial" status assessment still holds.Related: prerequisite #539 (F11.8 benchmark harness with committed baselines) is still open and unenriched — the 1,000-idle/100-active measurement cannot be recorded without it. The metrics from #553 would make the budget observable once enforced.
Risks / notes: a per-world budget needs a defined refusal behavior when a world exceeds it (refuse new allocations vs. evict vs. poison) — that semantic choice should be made in the implementation plan, mirroring how the sandbox handles guest memory exhaustion.
Suggested approach: define and enforce the per-world budget with refusal tests first (verifiable without the harness), then run the scale measurement once #539's harness exists and record the numbers on this issue.
209 remaining items
developing-today commentedon Nov 3, 2026
agent-note:v1
state: reviewed (coverage loop-back)
Coverage pass 2026-11-03, loop-back review: this issue was re-reviewed (drift / overlap / already-handled / stale / blockers / ordering) so every open issue in the four repos is reviewed and in the system. Verdict: still live; in the system as task 8085 (triage, queued, priority 50) in repo-health-code-pilot. Labels audited and synced. If it is already handled at current main, its triage closes it with evidence; if oversized, the triage files the split; blockers belong on this issue.
Found via: coverage loop-back 2026-11-03
dezren39 commentedon Nov 4, 2026
🤖 Agent note · triaged