ci-floor: cgroup cap spec + allocator model (step 4 destination); keep width=4 by gunbai-bot[bot] · Pull Request #5390 · gunb-ai/gunbc

gunbai-bot · 2026-06-20T16:14:48Z

Summary

Revised per warm-crane-135 redirect — do not ship spawn_width=1.

Fail-closed stopgap (operator action): docs/runbooks/ci-runner-cgroup-memory-cap.md — exact systemd commands to set MemoryMax=64G + MemoryHigh=60G on system-actions-runner.slice on srv1/srv2. Sized so runner_cap + OS(8) + sessions(48) + sccache(8) = 128 GiB. Normal width=4 × 14 GiB ≈ 56 GiB fits; pig runs (4×31 GiB) cgroup-OOM exit 137 instead of GLOBAL host OOM.
Keep ci-floor: memory-aware spawn_width — bound batch-2 fan-out by memory budget, not CPU only #5375 throughput: gunbc_ci_floor_spawn_width restored to memory-aware width=4 (not serial).
Keep destination model: AllocationClass + admit() in compute_fabric.dag + witnesses — marked step-4 destination, not wired into CI spawn path.
Supply grounding kept: BMC-grounded 128 GiB / 128 threads via placement_supply_row_from_bmc.

Test plan

ci_floor_plan_witnesses — PASS (width>1, memory-bounded)
compute_fabric_admit_invariant_holds — PASS
Eval gunbc_ci_floor_spawn_width → HardwareThreadCount { count: 4 }

Operator next step

Apply cgroup cap from runbook on srv1 + srv2 (needs root). Coordinate pig slimming with eager-boar-790.

… 1-3 of

…th at 1 Route CI through AllocationClass/ComputeRequest + BMC-grounded supply with a 48 GiB non-CI reserve so grantable fan-out stays serial, preventing claim_executor from reaching ~93 GiB anon-rss on srv2 until step 4 routes all consumers through admit. Co-authored-by: Cursor <cursoragent@cursor.com>

No code change — latest run 27876938344 passed (spawn_width=1, 546 witnesses). Dashboard was reporting failing from the cancelled superseded check. Co-authored-by: Cursor <cursoragent@cursor.com>

… 1-3 of

Per warm-crane-135 redirect: restore #5375 memory-aware spawn_width=4, keep AllocationClass/admit as step-4 destination (not active CI path), and add docs/runbooks/ci-runner-cgroup-memory-cap.md with exact systemd MemoryMax=64G commands for srv1/srv2 operator apply. Co-authored-by: Cursor <cursoragent@cursor.com>

Extract fleet_srv2_bmc_fixture.dag for ci_fleet + bmc_redfish witness. Addresses PR review cleanup debt (duplicated fixture; unused import). Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot · 2026-06-20T16:54:15Z

Review item 1 (cursor/composer-2.5 @ `c936dae`) — addressed

Finding: unused gunbc_ci_floor_compute_request import in ci_floor_plan.dag
Already fixed in ba12cf64 (warm-crane-135 redirect): CI no longer routes through ComputeRequest; import removed. Current imports are only ci_fleet_floor_spawn_width + ci_fleet_floor_spawn_fits_memory_budget.

Finding: duplicated srv2 BMC fixture vs bmc_redfish_grounding_witness_test.dag
Valid — fixed in 11b77f4453: extracted gunbc/fleet_srv2_bmc_fixture.dag as single authority; both gunbc.ci_fleet and the witness import from it. Also dropped unused offer_placement_supply_row import from ci_fleet.dag.

Witnesses green: bmc_redfish_grounding_holds, ci_floor_plan_witnesses.

— sent from neat-wren-326

Replace three parallel matches on AllocationClass with one AllocationClassSpec record and four named data rows (§2 fact-bundle modeling). Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot · 2026-06-20T16:55:49Z

Review item 1 (claude-opus-4-7 @ `c936dae`) — addressed

Finding: AllocationClass parallel-match smell (three matches on threads/memory/label)
Valid — fixed in upcoming commit on this branch: AllocationClassSpec record + four named data rows (allocation_class_small … xlarge); projections are field accesses via single allocation_class_spec() lookup.

Finding: witness_floor_spawn_width_grantable_capped asserts width==1
Stale / already addressed in ba12cf64 (warm-crane-135 redirect): that witness was removed when we dropped the width=1 serial stopgap. Current code uses witness_floor_not_serial (asserts width > 1; evaluates to 4 on fleet). No dissolve-on needed for a witness that no longer exists.

Note: review artifact describes pre-redirect scope (ComputeRequest CI wiring, 48 GiB reserve). Current PR ships cgroup cap runbook + #5375 width=4 + step-4 destination model only.

— sent from neat-wren-326

Re-balance to 110 GiB budgeted caps on 128 GiB (10 GiB headroom): runner 56G, sessions 38G (3×12G concurrent). Add dashboard spawn spec, live set-property apply path, mkdir guard. Runner-only cap is necessary-not-sufficient. Co-authored-by: Cursor <cursoragent@cursor.com>

… 1-3 of

Revert runner to operator-applied 64 GiB (14 GiB is stale scaffold, not measured normal-CI peak). Rebalance headroom via session budget slot (36 GiB illustrative). Part B is recommendation-only — measure peak-RSS distribution before operator routing. Persist host_memory_scaffold_* rows with dissolution triggers in compute_fabric.dag. Co-authored-by: Cursor <cursoragent@cursor.com>

…adCount Fact-bundle rows now store hardware_thread_count() directly; drop the Int projection and re-wrap in allocation_class_resource_envelope (§3 authority). Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot · 2026-06-20T17:24:05Z

Addressed in f202a6c.

AllocationClassSpec.threads is now HardwareThreadCount (same authority as CpuModelCatalogRow.threads). Fact-bundle rows store hardware_thread_count(count: …) directly; allocation_class_threads returns HardwareThreadCount and allocation_class_resource_envelope passes it through without the Int re-wrap.

Witnesses green: compute_fabric_admit_invariant_holds, ci_floor_plan_witnesses.

— sent from neat-wren-326

… 1-3 of

…fold Delete placement_spawn_width_*_with_reserve and compute_request_work_demand (no consumers — step 4 re-adds when dashboard wiring lands). Link scaffold register rows to runbook; admit witnesses read session_spawn_reservation. Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot · 2026-06-20T17:50:09Z

Addressed both non-blocking findings in 73598ed:

Unused step-4 width helpers — deleted placement_memory_width_bound_with_reserve, placement_spawn_width_with_reserve, compute_request_work_demand, and placement_spawn_width_for_request (no tree consumers; step 4 re-adds when dashboard wiring lands). placement_grantable_memory_bytes kept — used by admit().
Scaffold register rows — added bind: comments linking each row to the runbook table; host_memory_scaffold_session_spawn_reservation is now consumed by admit witnesses as reserve_bytes.

Thanks for the APPROVE.

— sent from neat-wren-326

gunbai-bot · 2026-06-20T18:08:19Z

cursor/composer-2.5 APPROVE — verified against HEAD 73598edbb2. No findings to address; prior REQUEST_CHANGES (AllocationClassSpec.threads → HardwareThreadCount) already fixed in f202a6c99b.

Merge readiness @ 73598edbb2:

cursor APPROVE + claude APPROVE (2 distinct api-review approvals)
no open REQUEST_CHANGES
mergeable=CLEAN
CI green (run 27879093818)

Ready for operator merge.

— sent from neat-wren-326

gunbai-bot · 2026-06-20T18:08:49Z

cursor/composer-2.5 APPROVE (review 31478 @ 73598edbb2) — no findings; nothing to change.

Merge readiness unchanged: cursor + claude APPROVE, no REQUEST_CHANGES, mergeable=CLEAN, CI green.

— sent from neat-wren-326

…-compute-fabric-stopgap

…or request) The allocator models consistent, safe *allocation* — fixed AllocationClasses + admit()'s Σ(grants) ≤ grantable invariant — not OS-specific *enforcement*. - grantable = physical RAM − reserve (was 0.50×RAM, a cgroup-MemoryMax heuristic). New typed helper placement_host_total_memory_bytes widens the Nat byte count to Int at the return boundary so the reserve subtraction is well-typed. - Remove dangling cgroup scaffolds host_memory_scaffold_runner_slice_cap and host_memory_scaffold_session_container_cap (0 uses). - Revert claim_executor.rs: drop the Linux /proc + /sys/fs/cgroup memory.peak measurement instrumentation (premature, OS-specific, unconsumed). - Delete docs/runbooks/ci-runner-cgroup-memory-cap.md; de-cgroup the ci_floor_plan / ci_fleet / compute_fabric comments; record the OS-agnostic decision in the plan. - Pre-existing #5375 memory-aware spawn-width kept (load-bearing); enforcement of a grant inside a container is out of scope (a peripheral per-OS realization, §3). Verified by execution: compute_fabric_admit_invariant_holds() = true (grants on empty host; refuses a 2nd Large that would exceed RAM − reserve). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…-compute-fabric-stopgap

briansrls · 2026-06-20T23:30:15Z

Superseded by #5419 — the allocator was unwired spec; #5419 rips out the spawn-width/placement over-modeling and pins the floor to a constant width instead.

…tion; pin floor width=4 (#5419) The memory-aware spawn-width / placement-width / memory-budget model was elaborate spec that nothing live consumed except the CI floor's concurrency — and that is a trivial constant. Cut it: - ci_floor_plan: gunbc_ci_floor_spawn_width -> hardware_thread_count(4) (claim_executor still evaluates it; host never decides width). Drop the memory-budget oracle fn. - ci_fleet: drop the spawn-width/placement fns + their imports; keep the offer + the live gunbc_ci_runner_spec (the runs-on source for ci.yml). - compute_fabric: delete placement_spawn_width / placement_memory_width_bound / placement_floor_memory_budget_bytes (the width parallelization). Keep the feasibility projection (placement_supply_row / offer_placement_supply_row) + satisfies + the types. - ci_floor_plan_witness_test: drop the 4 spawn-width witnesses; keep the plan-structure ones. Adds gunbc/operator_fleet.dag — srv1/srv2 as plain ComputeOffer inventory (real hardware; BMC creds as never-committed std.credentials handles; operator-owned). The authority CI/sessions derive from when a real consumer needs it; no placement modeling attached. Verified: dsl compile 0-diagnostics; the v2 floor closure compiles and ci_floor_plan_witnesses runs GREEN on the constant width. Supersedes #5390 (allocator) + #5417. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

briansrls and others added 2 commits June 20, 2026 16:14

WIP: Stop srv2 host-OOM via the fail-closed 'meantime' stopgap (steps…

b0445eb

… 1-3 of

gunbai-bot Bot marked this pull request as ready for review June 20, 2026 16:21

briansrls and others added 3 commits June 20, 2026 16:47

ci: re-trigger checks after superseded cancelled run

29f8f0f

No code change — latest run 27876938344 passed (spawn_width=1, 546 witnesses). Dashboard was reporting failing from the cancelled superseded check. Co-authored-by: Cursor <cursoragent@cursor.com>

WIP: Stop srv2 host-OOM via the fail-closed 'meantime' stopgap (steps…

504145b

… 1-3 of

gunbai-bot Bot changed the title ~~ci-floor: compute-fabric allocator stopgap caps spawn_width at 1 (srv2 OOM)~~ ci-floor: cgroup cap spec + allocator model (step 4 destination); keep width=4 Jun 20, 2026

refactor(bmc): single-authority srv2 fixture; drop unused imports

11b77f4

Extract fleet_srv2_bmc_fixture.dag for ci_fleet + bmc_redfish witness. Addresses PR review cleanup debt (duplicated fixture; unused import). Co-authored-by: Cursor <cursoragent@cursor.com>

refactor(compute_fabric): fact-bundle AllocationClassSpec rows

b26f20a

Replace three parallel matches on AllocationClass with one AllocationClassSpec record and four named data rows (§2 fact-bundle modeling). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls and others added 4 commits June 20, 2026 16:57

WIP: Stop srv2 host-OOM via the fail-closed 'meantime' stopgap (steps…

cd86264

… 1-3 of

fix(compute_fabric): type AllocationClassSpec.threads as HardwareThre…

f202a6c

…adCount Fact-bundle rows now store hardware_thread_count() directly; drop the Int projection and re-wrap in allocation_class_resource_envelope (§3 authority). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls and others added 2 commits June 20, 2026 17:45

WIP: Stop srv2 host-OOM via the fail-closed 'meantime' stopgap (steps…

e70d428

… 1-3 of

briansrls and others added 4 commits June 20, 2026 19:05

Merge remote-tracking branch 'origin/main' into session/neat-wren-326…

a3c0ca7

…-compute-fabric-stopgap

Merge remote-tracking branch 'origin/main' into session/neat-wren-326…

0724be7

…-compute-fabric-stopgap

Merge remote-tracking branch 'origin/main' into session/neat-wren-326…

9a626c3

…-compute-fabric-stopgap

This was referenced Jun 20, 2026

compute fabric: allocator (Σ≤RAM) + operator fleet authority (srv1/srv2 BMC-grounded) #5417

Closed

compute-fabric: cut the unwired spawn-width/placement over-modeling; pin floor width=4; keep fleet inventory #5419

Merged

briansrls closed this Jun 20, 2026

Navigation Menu

Repository navigation

ci-floor: cgroup cap spec + allocator model (step 4 destination); keep width=4 - #5390

ci-floor: cgroup cap spec + allocator model (step 4 destination); keep width=4#5390
gunbai-bot[bot] wants to merge 17 commits into
mainfrom
session/neat-wren-326-compute-fabric-stopgap

gunbai-bot Bot commented Jun 20, 2026 •

edited

Loading

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

briansrls commented Jun 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

gunbai-bot Bot commented Jun 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Operator next step

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Review item 1 (cursor/composer-2.5 @ c936dae) — addressed

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Review item 1 (claude-opus-4-7 @ c936dae) — addressed

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

gunbai-bot Bot commented Jun 20, 2026

Uh oh!

briansrls commented Jun 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

gunbai-bot Bot commented Jun 20, 2026 •

edited

Loading

Review item 1 (cursor/composer-2.5 @ `c936dae`) — addressed

Review item 1 (claude-opus-4-7 @ `c936dae`) — addressed