Repository navigation
test: validate v1.0.0-rc.1 across macOS, Windows, KR, and US #150
Description
Objective
Validate published v1.0.0-rc.1 (1.0.0rc1, commit 70f99b2e66846043f2b607d01277fe5bb7d0f0e2) on Mac mini, MacBook, Windows quintet-desktop, KR and US before stable promotion. User approved installation, coordinated KR runtime replacement, actual agent ACK checks and a four-hour soak.
Installation and evidence
- Published wheel/sdist provenance and fresh installs
- Mac mini: standalone CLI/skill upgrade and exact MCP/relay venv
- MacBook: standalone CLI/skill upgrade and exact MCP/relay venv
- Windows: local-agent handoff, native core/MCP; WSL relay if available
- KR: user CLI and coordinated production control/relay replacement after verified snapshot
- US: exact package venv under the existing ubuntu account
Behavioral validation
- Local discovery, JSON contracts, diagnostics, dry-run and empty/missing-agent behavior
- Bidirectional Mac SSH, server-to-Mac and available Windows SSH routes
- Dedicated Claude/Codex/Antigravity sessions: unique marker and actual ACK, separate from submitted/queued
- MCP initialization, tools, denied policy, dry-run, approved send
- Forced direct, forced relay and automatic routing
- Duplicate/conflicting request IDs, revoke, reconnect, key rotation, lost-response reconciliation and durable replay preservation
- KR public health, device-code/OAuth, authenticated WSS and operational readiness
- Four-hour soak: health every 60s, direct/relay probes every 5m, actual ACK at T0/T2h/T4h, one coordinated relay restart at T2h
- Final resource/restart/state checks, release gate and redacted evidence
Acceptance and triage
No unexplained failures, duplicate effects, lost messages, revision regression or secret exposure. All required actual ACK checks pass. Planned relay restart recovers within 60 seconds. Missing prerequisites remain explicitly blocked, not passed.
Create a separate deduplicated issue for each reproducible finding with environment, exact artifact, reproduction, expected/actual result, redacted evidence and ACK status. Runtime/build/packaging/auth/state fixes belong to v1.0.0-rc.2 (target 2026-09-25). Documentation/metadata/promotion-only work belongs to v1.0.0 (target 2026-09-29). Security findings use a private advisory. Dates are targets; failed gates block promotion.
After runtime fixes, install RC2 on all five hosts and repeat the affected routes and baseline matrix. Relay/control/state/recovery changes repeat the four-hour soak; core-only changes require a 30-minute all-host regression run. With no runtime changes, RC2 can be closed as not required after complete evidence. Merges and release publication require separate approval.
Use isolated test sessions/state. Preserve unknown send outcomes and reconcile instead of retrying. Preserve production identity, receipts, tombstones and replay state. KR rollback switches to the previous runtime only when schema-compatible; do not restore an old live database automatically. Do not copy credentials between hosts.
Activity
abruption commentedon Sep 22, 2026
RC1 rollout / first validation checkpoint (2026-09-22 KST)
All five machines now resolve
session-peer 1.0.0rc1. Existing standalone installs on both Macs and US were preserved; native Windows retained pip-user installation. KR has a user venv and the production relay uses an immutable RC1 venv. Optional MCP/relay environments are exact-version pinned. Windows WSL2 Ubuntu 24.04 has its own exact relay venv.Published wheel SHA256
7f4f4f79e4f17303badb37d4310cd6c22dad4d03f9da747071e91121fe0371d8; sdist SHA2565c9184021fea0145fd8ea28dcd4c10308f36cc9868a84fd286eec4a1ca165710. Both match PyPI metadata; fresh wheel/extras and sdist installs succeeded. CI and publication runs for the exact RC1 commit are successful.KR production
Passed live / protocol evidence
message_id_conflict.Findings and remaining gates
lock_probe_unsupported— RC2 blocker.disk I/O error— RC2 blocker. One authorized send returnednative_outcome_unknown, request ID7e641e37-b5fa-4df4-8475-3465a9fa7015; it was not retried.pip checkreports an existing unrelated mysql-connector/protobuf dependency gap; the isolated WSL venv passes. Claude posting alone is not counted as ACK.A four-hour operational soak has started for the two Macs, US and KR: 60-second health/resource/state samples, five-minute direct/relay probes, actual-message checkpoints at T0/T2h/T4h and one relay restart at T2h. Windows native/WSL gates remain failed, so even a passing soak will not close the full five-host gate. Runtime fixes and RC2 revalidation are required before stable. No merge/tag/release publication was performed.
abruption commentedon Sep 22, 2026
Windows follow-up: the user confirms claude-main shows neither an incoming message nor an approval/startup prompt. Registry, running process and actual named-pipe server PID match; one auth-only probe disconnects with ERROR_BROKEN_PIPE before any user message is sent. Filed #157 as an RC2 blocker; exact auth rejection cause remains unconfirmed. No ambiguous send was retried. T0 actual ACKs were observed for all five dedicated Mac targets, but mini Codex required canceling an unrelated SSH approval prompt, so it was not unattended. Operational soak runs 2026-09-22 14:04–18:04 KST, with planned relay restart at 16:04. A bounded reporter will attempt one summary comment at completion/abort; it will not infer T2/T4 actual ACKs or close the release gate. Live rotation/recovery/revoke and remaining actual ACK review are still pending. GitHub and Google browser logins were both validated; provider accounts remained distinct.
abruption commentedon Sep 22, 2026
MacBook handoff reconciled (2026-09-22 14:18 KST): the local validator reports exact RC1 install, rollback backup, discovery and real MCP initialize/tools/list/list_sessions success. Coordinator independently rechecked current standalone version 1.0.0rc1, SHA256 bb642c323c4e69b98c9c28c711441cac17b821aad56bbf452f997984ec4d1790, RC venv Python 3.14.7 and clean pip check. Its report of missing dedicated ACK targets and abruptly@Windows SSH denial describes an earlier/local validation scope: coordinator subsequently created dedicated MacBook Claude/Codex test sessions and observed actual SSH/T0 relay ACKs, and authenticated Windows with the correct account. These prerequisite resolutions do not resolve the Windows product defects #151/#153/#156/#157. Do not rerun installation or start a second soak from this handoff. Existing four-hour soak remains running; this is not a completion report.
abruption commentedon Sep 22, 2026
Windows blockers now have draft PRs: #158 (main/RC2: #151/#153/#156/#157) and #159 (release/0.9.x → 0.9.2 minimal hotfix: #151/#157). User independently confirmed exact ACKs WIN-CODEX-153-4712e8d0, WIN-RELAY-156-a29d13e4, WIN-CLAUDE-092-6b9e40a2. Native Windows MCP adapter list/dry-run passes. #151 live SSH diagnostic verified. RC full local suite: 477 tests OK, 11 skips. Stable suite: same 8 fixture-isolation failures as untouched v0.9.1 (381 vs 370 tests), no new failures; wheel/sdist fresh installs passed. CI started for both draft PRs. No release/tag/PyPI publication or production Relay policy change. Original unknown sends/journals preserved and not retried. Windows default RC1 is unchanged; isolated WSL receiver currently runs the candidate on preserved state/policy with a bounded lifetime. Existing KR/US/Mac soak remains separate and ongoing. #152 remains separate. Maintenance publication route must be reviewed before 0.9.2 publication.
abruption commentedon Sep 22, 2026
Stable hotfix v0.9.2 is published: https://github.com/abruption/session-peer/releases/tag/v0.9.2 and https://pypi.org/project/session-peer/0.9.2/. Exact merged source a751140, protected maintenance preflight, publishing run 35695005641 and post-upload checks all passed. Independent GitHub/PyPI/workflow artifact hash comparison and fresh wheel/sdist installs passed; standalone 0.9.1 sees v0.9.2 as latest stable. v0.9.2 milestone closed. This completes the stable hotfix publication for #151/#157, not the RC2 campaign: #158 and RC2 issues remain open/unmerged, and existing fleet RC installations were not overwritten.
abruption commentedon Sep 22, 2026
RC2 preparation is now in #160, based on merged #158 (5549be3; main push CI passed). It bumps only the runtime version to 1.0.0rc2 and adds four-language notes/runbooks with issue mapping #151/#153/#156/#157, mandatory WSL codexPython migration, and the exact-package five-host validation plan. Local full suite: 480 tests OK (11 skips); two builds match; archive tests, fresh wheel/sdist installs, extras pip check and installed MCP initialize/tools-list pass with zero messages sent. RC2 is NOT published/installed on the fleet. Because receiver transport code changed, the plan requires a new four-hour campaign; existing RC1 soak/candidate ACK evidence does not substitute. Issues/milestone remain open pending published RC2 acceptance. No production restart/state change, old unknown-message retry, or installation-manager replacement was performed.
abruption commentedon Sep 22, 2026
RC2 package publication complete: https://github.com/abruption/session-peer/releases/tag/v1.0.0-rc.2 and https://pypi.org/project/session-peer/1.0.0rc2/. Source fe9d8d8, successful publish run 35696656167. Public GitHub/PyPI/workflow hashes match; fresh wheel+extras/sdist installation, pip check, MCP initialization/tools and empty Claude discovery pass. v0.9.2 remains latest stable. This is package publication only: no fleet default install, production restart, identity/replay mutation, or live send was performed. Next acceptance step remains approved exact-RC2 installation on Mac mini/MacBook/quintet-desktop/KR/US, WSL codexPython migration, fresh actual ACK route checks and the new four-hour campaign. Existing RC1/candidate results are not promoted to RC2 acceptance. Keep all four Windows findings and milestone open until evidence is complete.
abruption commentedon Sep 22, 2026
RC1 stopped; RC2 five-host revalidation running
User explicitly requested stopping the entire RC1 campaign, reinstalling RC2, and repeating real ACK plus four-hour validation.
abortedat 112.8 minutes, with recorded health/Relay failures; it is not a passed soak. Old T2 restart was canceled. Evidence, identities, paired state, policies and unknown request journals are preserved.1.0.0rc2, tagv1.0.0-rc.2, commitfe9d8d83a8f9eb7bb26d47c6c3ebbf6c9b211608) is installed on Mac mini, MacBook, Windows native and WSL, KR user/runtime, and US.pip checkreports an existing unrelated MySQL/protobuf dependency problem; no unrelated dependency change was made.New four-hour clock: 2026-09-22 16:12:59–20:12:59 KST (checkpoint completion/report may finish shortly afterward). Health/resource/state checks every minute; forced-direct/public-Relay route checks every five minutes; new single-attempt ACK tokens at T0/T2h/T4h; one planned Relay restart at T2h (18:12:59 KST). Windows public Relay/native-core paths are included. Checkpoint notifications request independent live-PTY/user ACK review; the bounded final reporter never infers consumption.
Current result is not PASS: intermittent public Relay failures already reproduced during preflight/T0 and are being tracked separately. The new run records failures rather than silently restarting the clock or masking them. No issue closure, milestone completion, stable promotion, or new release is implied by starting this test.
Evidence: local RC2 campaign
BuZTaE; operational log, checkpoint request IDs, ACK observations, installation transition and final reporter outputs kept separately. No transcript polling, credentials, private keys or raw auth material are published.abruption commentedon Sep 22, 2026
RC2 T0 independent observation update: Windows Codex exact ACK RC2-T0-win-native-codex-d7941d4c is user-confirmed. For Windows Claude RC2-T0-win-native-claude-1b1a95e7, the user clarified their initial both-ACKs statement: the request is visible but the agent is waiting for approval. Record request reception confirmed / approval pending / actual ACK unconfirmed, NOT a passed Claude ACK gate. No permission bypass or resend. The earlier preflight confirmation is a separate checkpoint and does not prove unattended Claude behavior. Mini Claude public-Relay T0 exact ACK was independently observed on its owned live PTY.
abruption commentedon Sep 22, 2026
RC2 diagnostic boundary — issue #161
User authorized enabling privacy-filtered Relay access logs and a bounded causal investigation, while retaining the existing RC2 campaign as diagnostic evidence.
abruption commentedon Sep 23, 2026
RC2 campaign reconciliation / RC3 preparation (2026-09-23)
The RC2 operational soak has no terminal
completerecord: its last recorded minute is 172 (19:05:01 KST). It therefore did not meet the four-hour acceptance gate, irrespective of the reason its coordinator stopped. Public Relay route failures were present from the beginning and are tracked in #161; the possible coordinator restart explains only the missing ending, not those failures. Do not count previous T0 or one-shot direct/Cloudflare ACKs as T4 validation.An independent Claude read-only investigation and code cross-check are summarized in #161. Receiver idle/reconnect and stale-room hypotheses are split into #163. A fresh exact-artifact five-host RC3 matrix and uninterrupted four-hour criterion are specified in #164 under milestone
v1.0.0-rc.3; draft PR #162 is only a candidate for that milestone. No RC3 build, fleet install, production switch, new soak, release or issue closure has occurred. The previously prepared 15-minute receiver wrapper's completion gate was not met by the incomplete RC2 run.abruption commentedon Sep 24, 2026
The RC1 validation plan was superseded by the completed RC4 five-host campaign tracked in #164. RC1/RC2 defects were triaged into #151/#153/#156/#157/#161/#163 and addressed or explicitly carried forward. The RC4 rerun has a complete 14,436.65 s record and documented T120 Windows Claude operational exception; see #164. Close this obsolete RC1 plan without rewriting its historical acceptance state.