Skip to content

ci_budget_tree_witness: three live-tree reds were stale premises, re-derived from the producers - #12907

Closed
gunbai-bot[bot] wants to merge 8 commits into
mainfrom
session/deep-deer-663-budget
Closed

gunbai-bot[bot] wants to merge 8 commits into
mainfrom
session/deep-deer-663-budget

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Oct 1, 2026

Copy link
Copy Markdown
Contributor

Stacked on #12905, which admits this module at entry. Follow-up (2) from deep-ferret-305.

I read each claim's observed value with a temporary probe on 2026-10-01. In all three cases the claim was wrong and the subject was right.

claim observed why the claim was stale repair
witness_live_is_fail_closed gha1 = 466,266,685,440; sess1 = 55,834,574,848 It asserted the Unestablished arm (runner pool 0, session = all allocatable). srv1 now resolves ReservationBytes (52 GiB), so a positive pool is the correct derivation. Each arm's relation is asserted over session_reservation_bytes(srv1), so the fail-closed half still applies if srv1 falls back to Unestablished.
witness_srv2_symmetric_to_srv1 ram1 = 539.3e9; ram2 = 134.3e9 srv1 was rebuilt to 512 GiB on 2026-08-22, and since #11625 the trees read usable RAM per host. Equal pools stopped being true. p1 - p2 == ram1 - ram2. The overhead and the slice are the same on both hosts, so nothing else may differ.
witness_srv3_outbudgets_srv1_by_exactly_the_overhead_gap p3 - p1 = gap + 1,392,640 Usable RAM differs between srv1 and srv3 by 1,392,640 B (#11625). p3 - p1 == overhead gap + RAM gap, both read from the producers.

All three are removed from floor_expected_red_chunk_live_tree_admission, because they now pass.

Evidence: local claim_batch --hermetic over the whole module: 13 PASS, 0 FAIL.

🤖 Generated with Claude Code

gunbc-ci-auto-heal and others added 8 commits October 1, 2026 11:41
… LiveTreeDisposition

(1)(2) live_deploy.emit sudoers claims: the needle is now the install's own node
(gunbc.ci_deploy_sudoers deploy_sudoers_elevated over the fleet visudo row) rendered by the same
serializer. Stale since #12602 (/usr/bin/sudo) and #12525 (bash builder quotes every word). The
two probe negatives are deleted: unmatchable under quoting, and since #12168 those probes are
emitted on purpose after the install.
(3) twin claim: count equality replaced by an identity join on artifact kind, host singletons
(fabric storage + #12747's approval broker front door) subtracted by the functions that decide them.
(4) CPUQuota grant: sudoers side read through sudoers_argument_word (escape since #12563).
(5) tasks verdict: bare `Absent ==` never named the ConvergeVerdict arm; typed match + a Drifted
discriminating conjunct.
(6) runner_lifecycle: fabric rows from srv3/srv4_fabric_first_slot, each controlled by the
slot below it on its own host (srv4-06 is fabric since 2026-09-18).
Admission: build_cache_endpoint_observe, ci_budget_tree_witness, host_allocation_conservation
import v2.std.live_tree (the #12540 class #12819 fixed once); variant rows retired ImportsFixed.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…derived from the producers

witness_live_is_fail_closed asserted the Unestablished arm while srv1 now resolves ReservationBytes;
it asserts each arm's relation over session_reservation_bytes(srv1). witness_srv2_symmetric_to_srv1
assumed equal RAM (srv1 is 512 GiB since 2026-08-22; usable RAM per host since #11625); it asserts
the pools differ by exactly the RAM gap. witness_srv3_outbudgets_srv1_by_exactly_the_overhead_gap
missed the 1,392,640-byte usable-RAM difference; gap = overhead gap + RAM gap, both read from the
producers. Removed from floor_expected_red_chunk_live_tree_admission (they now pass).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The floor's changed cost-debt edit judgment lexes the whole 125 KB emit_test.dag at base and at
head per identity in the interpreter; with these two identities changed, site projection ran past
the 90-minute cap (run 36856989404). Their fix moves to its own PR, held on that floor defect.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…-budget

# Conflicts:
#	dag/test/claim/ci/ci_budget_tree_witness_test.dag
…w 73848)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@gunbai-bot

gunbai-bot Bot commented Oct 1, 2026

Copy link
Copy Markdown
Contributor Author

Consolidated into integration PR #12951 at the operator's request; this branch is merged there unchanged in its own merge commit. Closing here; the branch is kept. — sent from deep-ferret-305

@gunbai-bot gunbai-bot Bot closed this Oct 1, 2026
@github-merge-queue
github-merge-queue Bot removed this pull request from the merge queue due to a manual request Oct 1, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

0 participants