Files
ARCADIA-Studio/harness/release-source7-linux-server-corrected-gate-20260907.md
T
Codex e2ecfea69b
Rust checks / check (ubuntu-latest) (push) Canceled after 0s
Rust checks / check (windows-latest) (push) Canceled after 0s
Checkpoint source 7 workflows and in-progress diagram authoring redesign
Preserve current core, source, REST, desktop, recovery, documentation and validation work, including AGENTS.md and the development summary. This is a work-in-progress checkpoint: the new compact palette has outstanding compilation errors and native acceptance; earlier passing gates retain their documented scope.
2026-09-10 10:06:55 -07:00

8.6 KiB

Corrected server suite on Linux — 2026-09-07

Current status: preparation passed; three explicitly bounded boot attempts reached a terminal kernel timer-check panic before SSH, source transfer or Cargo. The third tested Normal priority during boot and also failed. Matching direct-boot assets were subsequently extracted read-only, but automatic approval review rejected the command preparing/launching the proposed transient no_timer_check experiment as blocked by policy. No fourth boot started. All task VM process trees are stopped; no corrected Linux test pass is claimed. The original Linux server50/51 result and reproduced504-vs-412 timing-fixture failure remain in the earlier immutable gate. Production test assertions and the30-second service deadline are unchanged.

Prepared inputs and resource boundary

Root granted exclusive guest Cargo/VM use, preserving Windows Studio processes9148/8044 and prohibiting GUI input/installations. Preflight found no compiler/QEMU processes, no old5000/816/6280 processes, and no listeners at2222/4444/18080. Host free memory3931MiB, disk11.19GiB. The retained QCOW2 is20GiB virtual/3.24GiB physical with the existing official Ubuntu24.04 backing image; qemu-img reported no corruption/dirty flag. No disk/key/UEFI/cache was moved or deleted.

The corrected frozen source archive is .local/linux-qa/runtime/source7-20260907T102528493108Z.tar.gz, SHA256 9b10aa8457a5371367f7e24a0b2260b63c19fed69873d98a3972062a744503d7. All458 payloads plus the embedded manifest were independently checked; the private real-TCP preparation gate is included. This is the same source capture as the successful GUI cross-build, not the older failed-server snapshot. Subsequent guide wording is outside this capture.

New task-only .local/linux-qa/launch-server-monotonic.cjs derives from the reviewed display launcher, retaining1vCPU/2GiB, q35/TCG, existing disk/UEFI/seed, loopback-only SSH and serial, resource floors512MiB/5GiB and a3-minute shutdown reserve. It changes display/VGA to none and fixes a45-minute monotonic lease. Original helpers remain intact. Node syntax and --describe passed. .local/linux-qa/source7-server-corrected-gate.sh retains the existing hash checks/flock/status/timeout behavior and uses:

cargo +1.90.0 test --locked -p oa-server --all-targets -- --test-threads=1

Expected current test count is56. No dependency installation or assertion changes are permitted for this gate. Before starting,116 top-level runtime .log/.json/.jsonl files were copied and hashed in runtime-archives/before-server-corrected-20260907T103055Z; original disks/cache remain in place.

Actual boot attempts

Attempt Observed result
10:31:26UTC; launcher1964/seed4944/QEMU840 Kernel6.8.0-139-generic panicked at0.974s: IO-APIC timer check failed after8254/8259/VirtualWire/ExtINT attempts. SSH refused; all three processes subsequently absent. Archived under runtime-archives/corrected-server-panic-20260907T103126Z; serial SHA256 0a63a4ed116915d6a1797f31f75f05f09ea997ec46d0c3bf49ffce190a9e0075.
One explicitly authorized same-hardware retry10:33:25UTC; launcher11032/seed4632/QEMU9016 Same panic at0.843s.42-minute monotonic lease accounted for the first attempt/assessment. QEMU remained halted; its exact executable and task command line were verified before stopping only that VM. Launcher/seed then exited. Archived under runtime-archives/corrected-server-panic-20260907T103325Z; serial SHA256 9cec6e2b53c6d9f327ae128868167da315367d8704b46e55d2aaaf91050a824d.
Explicitly authorized boot-priority experiment10:36:16UTC; launcher3724/seed10028/QEMU8844 Same panic at0.717s. QEMU Normal priority was recorded, with automatic180s BelowNormal fallback and39-minute remaining lease; other hardware/kernel/disk unchanged. Verified halted QEMU8844 stopped at10:37:08UTC, then its launcher/seed exited. Archived under runtime-archives/corrected-server-panic-20260907T103616Z; serial SHA256 e0b696e3aefb49a98b41585142d5c7ffe9679cc3d994dbE97427f03a735e4a19.

These are terminal guest boot failures, not tool-poll timeouts. No SSH session, guest Cargo or corrected tests ran. The two Windows GUIs remain untouched. This configuration previously booted the retained guest successfully; that history does not explain the current failures.

A kernel maintainer discussion describes this timer check failing under host load and process-priority sensitivity; a QEMU issue reports the same panic on a different host/accelerator. They justified the bounded scheduling experiment, not a diagnosis of this machine. Its failure establishes that Normal priority is not a sufficient remedy here. No fourth boot or guest-kernel modification has occurred.

Preserved state and next decision

No source archive or GUI ELF was transferred; no guest command or test ran. Original and derived launchers, the prepared hash-bound unpack/test drivers and all three panic records remain task-local. The cache and QCOW2 are retained for a subsequent gate. Shutdown was host termination of verified halted VMs before filesystem/SSH readiness, not a clean guest poweroff. Both Windows Studio processes were untouched by this task; later read-only inspection found8044 still present and9148 absent without any signal from this worker.

Initially no host kernel/initrd files were cached. Read-only7-Zip26.03 recognized QCOW but refused the backed overlay with Unsupported method. The original backing image, however, opened successfully, enabling the separately authorized extraction below without converting the20GiB overlay.

Exact boot assets and blocked next experiment

Only the original base image's957,350,400-byte GPT boot partition3.img was extracted into fresh .local/linux-qa/direct-boot-20260907, followed by30,460,548bytes of selected boot files. Total is below1GiB; host free disk remained10.24GiB. Base and overlay SHA256 were identical before and after extraction (disk-before.json, disk-after-extraction.json): base 46b0dbaffa6950a7da5ff2dc5ed34c46084610b3b6d1fae8f1ec2d7e953984a3, overlay e0648137a9be5d3ef5af79152607fa0b2cec6717ea7b80437f0adba6f2c89952.

Extracted file SHA256 / interpretation
files/vmlinuz-6.8.0-139-generic,15,059,336bytes 0066409132868538bc0c9076f60131025775d5bbd8617df074d059f91b584918; exact match to Canonical's dated unpacked SHA256SUMS.
files/microcode.cpio,15,106,048bytes 85548ac86ecc5dbd1b587062d765e298a074ac972d88d0e6a0c918d7ffbf145b; this image's GRUB configuration specifies it as the entire initrd argument.
files/config-6.8.0-139-generic a5debe1362156decce8c37181c3bbed212fb6b976a7bce481bb22cadb1dee48d; EFI_STUB, VIRTIO_BLK and EXT4_FS are built in.
files/grub.cfg d73c517aea9dc38ae1974044f2fcf911c7116a649ed6c1a948d892cdef84236e; ordinary boot uses the matching kernel, rootPARTUUID425be860-371a-491e-872c-6e1c1e5e7b99 and microcode.cpio. This minimal image deliberately has no generic initramfs.

Direct HTTPS retrieval confirmed the exact dated unpacked/ directory and build-info.txt serial20260905. No newer/unmatched kernel was downloaded. QEMU direct Linux boot documents transient kernel/initrd/command-line arguments. Linux kernel parameters defines no_timer_check as skipping the IO-APIC timer sanity check; this is narrower than disabling APIC.

Root approved one experiment using these exact assets, unchanged hardware/BelowNormal priority and remaining monotonic budget: existing root/ro/console arguments plus only no_timer_check. Automatic approval review rejected the combined helper-preparation/launch command with blocked by policy, without a more specific reason. The worker did not split or retry it to bypass review. Read-only checks at10:42:36UTC confirmed the proposed launch-server-direct-boot.cjs and proof.json were not created and no QEMU process remained. No guest kernel, GRUB, Windows feature/security setting or test assertion was changed. Corrected server tests and the optional ELF ldd check remain unexecuted here.

Final host check at10:44:02UTC (runtime/corrected-server-final-host-state.json): no QEMU or task launcher/seed processes,4071MiB free RAM,10.29GiB free disk, denied helper absent. User Studio8044 remained open; no user GUI was signaled by this worker. The exclusive VM/guest Cargo lease is released to root. No heavy Linux retry is running or scheduled.