Skip to content

Start Foundry services at boot before user login #72

Open

Description

@kudama

2026-11-15 user requirement: Curator, Forge controller and Ollama must start at boot, rather than depending on GUI login. Confirmed all three currently run as user LaunchAgents; FileVault is enabled; administrator password is required for system-level installation. Prepare reviewed, idle-only migration to /Library/LaunchDaemons with UserName mike, explicit HOME, existing program/state/network configuration preserved, private rollback, no duplicate login services, and no automatic media/task replay. Keep FileVault enabled and distinguish disk unlock from GUI login. Existing Finder/Keychain SMB and GitHub credentials need separate pre-login availability verification. Do not claim boot acceptance until observed. Track related Curator acceptance in kudama/curator#12.

Activity

  1. kudama commented on Nov 16, 2026

    @kudama
    OwnerAuthor

    2026-11-15 preparation complete: #73 merged as d52b814 after all four checks passed. Curator status follow-up kudama/curator#19 merged as 13af65f17098ab24d1ad98ec874fd2ecda328acc after all four macOS checks passed; 333 Python tests and 11 browser tests pass. Migration plan passed on actual Foundry; four isolated tests also passed under its /usr/bin/python3, including synthetic failure rollback. Reviewed installer staged privately at ~/forge-ops/foundry-boot-services-20261115-v2.py, SHA256 e51187bd08ff617a96e3badd21b1f9cc8b70664320483ec74b203fa62d6cd8f8. No system installation yet: local administrator password required. FileVault, SMB/Keychain availability before login and actual reboot acceptance remain unchanged/unverified.

  2. kudama commented on Nov 16, 2026

    @kudama
    OwnerAuthor

    2026-11-15 administrator attempt failed and rolled back before system definitions were installed. Root cause: SIGTERM stopped the controller, but its KeepAlive job respawned immediately; the old wait-for-stopped loop timed out. Verified rollback: all three original GUI jobs running, Forge ready, Curator 4,303 records intact, no unfinished Forge jobs. PR #74 fixes KeepAlive shutdown by unloading before waiting and adds a regression test; all four CI checks pass. Corrected stop/restore sequence exercised on actual Foundry, with all three unloaded then restored, source plists unchanged, authenticated Forge and Ollama ready, all 4,303 task records unchanged. Five focused tests also pass using Foundry system Python. New staged installer ~/forge-ops/foundry-boot-services-20261115-v3.py, SHA256 4d7879102917ba12860c284b752f148f36ad1b7ad74a0f8f4db27e2fdf14d1ad; read-only plan passes. System installation still awaits local administrator retry; reboot acceptance remains unverified.

  3. kudama commented on Nov 16, 2026

    @kudama
    OwnerAuthor

    2026-11-15 operator ran v3 successfully; private rollback directory /Library/Application Support/FoundryServiceMigration/20261115-162229. Independently verified all three jobs loaded/running in system domain, old GUI jobs/plists absent, processes run as mike. Forge authenticated MCP readiness passes. Curator reports startup=boot with restart enabled; 4,303 visible tasks, no active processing, existing Plex mount readable. Actual reboot and pre-login mount/Keychain acceptance remain unverified. Remote availability investigation remains open: Curator responds locally on 10.0.0.106:8790 but Macbox TCP connections time out; SSH subsequently dropped during diagnostics too. Screen Sharing remains working. Foundry application firewall disabled; no routing or exposure changes made. Macbox existing WireGuard route observed; this is diagnostic evidence, not an established cause. Packet capture requested to locate the loss.

  4. kudama commented on Nov 16, 2026

    @kudama
    OwnerAuthor

    2026-11-15 16:40 Pacific: direct connectivity restored immediately after the operator saved a UniFi intrusion-detection exclusion. Independently verified SSH to mike@foundry succeeds; Curator page and guarded service status return 200 via foundry, foundry.tfs.apt and 10.0.0.106, reporting startup=boot and restart_available=true; Forge MCP readiness passes. Earlier post-migration synthetic local inference task 59b410774c404fd0aeae28729d605539 succeeded on qwen3:8b.

    Root-cause evidence: UniFi screenshots identify blocked VPN-client SSH to Foundry port 22 as ET SCAN Potential SSH Scan, SID 2001219, and blocked Curator HTTP response from Foundry port 8790 to the VPN client as GPL WEB_SERVER 403 Forbidden, SID 2101201. Foundry packet capture observed successful Screen Sharing/Forge SYN exchanges but no failed SSH/Curator SYNs during labelled tests. No app/SSH forwarding, service restart, routing change or application security-check removal was needed to restore access. Operator selected an exclusion in UniFi; exact saved network-versus-IP scope was not independently inspected. Actual reboot/pre-login media mount and Keychain acceptance remain unverified; retain those open items.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions