Skip to content

2026-09-17: macOS 27 Post-Reboot Estate Audit and Service Stabilization

Operational telemetry command center visualization of a post-reboot systems audit on Apple silicon with emerald and gold telemetry streams.

Host manoir rebooted into macOS 27.0 at dawn. We opened the diagnostic telemetry to verify our estate. Tommy watched from the mat. The Council demanded proof on disk before declaring all clear.

While core daemon processes survived the kernel upgrade, several friction points demanded operator attention: system indexing pressure driving high load, an orphaned process contending on port 8199, permissions on /var/log/sanctum blocking the SharePoint bridge, and provider credentials requiring rotation.

1. System Load & Post-Upgrade Indexing Triage

Section titled “1. System Load & Post-Upgrade Indexing Triage”

Upon host boot, macOS 27 initiated comprehensive migration passes across CoreSpotlight, Metadata, Photos/MediaAnalysis, and Mail:

  • hybridsearchd, corespotlightd, CGPDFService, and mediaanalysisd consumed heavy background cycles.
  • Mail.app ran /System/Applications/Mail.app/Contents/MacOS/Mail -MailLibraryUpgradeOnly YES against ~/Library/Mail/V10/MailData/Envelope Index.
  • In addition, a recursive grep across the runner workspace had pegged one core at 100%.

Once the runaway search process was terminated and memory pressure subsided, host load plummeted from an initial peak of 151.02 down to 11.37, restoring normal CPU scheduling and preventing starvation of timing-sensitive canaries.

Initial inspection raised concerns regarding the Airtight VM (10.10.10.10), where diagnostic probes initially looked for retired port assignments. A live inspection over SSH confirmed that all core containerized and system services were running under systemd user mode:

  • Graphiti Server: active on 127.0.0.1:31416 (graphiti-server.service).
  • Health Ingester API: active on 10.10.10.10:10101 (health-ingester.service).
  • OpenClaw Gateway: active on 0.0.0.0:1977 (openclaw-gateway.service).
  • VM-hosted Docker Containers: Documenso, Outline, Dragonpit Conduit, and Neo4j (127.0.0.1:7474) all healthy.

The comprehensive test harness ~/.sanctum/tests/test-connections.sh confirmed 26 passed, 0 failed across Home Assistant Green (10.0.0.3 and Tailscale 100.95.147.70), Health Ingester, Cloudflare Tunnels, Tailscale MagicDNS, and Memory Vault (localhost:42069 holding 6,967 notes).

3. SharePoint Bridge LaunchDaemon & Directory Permissions

Section titled “3. SharePoint Bridge LaunchDaemon & Directory Permissions”

Cloudflare Tunnel logs reported HTTP 502 errors when routing https://bridge.nepveu.name to origin http://localhost:8443.

Investigation revealed that system LaunchDaemon /Library/LaunchDaemons/name.nepveu.sanctum-bridge.plist (running as user bert) had failed to spawn with exit code 78 (EX_CONFIG):

  1. Directory Traversal Restriction: Following the macOS upgrade, /var/log/sanctum was recreated with permissions drwxr--r-- root:wheel. Non-root processes could not traverse the directory to open bridge.out.log and bridge.err.log.
  2. Audit Log Failure: The underlying service failed with PermissionError: [Errno 13] Permission denied: '/var/log/sanctum/audit.jsonl'.

Bert restored proper ownership and group permissions:

Terminal window
sudo chown -R root:sanctum /var/log/sanctum
sudo chmod 775 /var/log/sanctum

Launchd immediately spawned name.nepveu.sanctum-bridge on port 8443. The public bridge endpoint responded with HTTP 200 (ok v0.1.0 modules: m365, sharepoint), and sanctum-bridge-canary successfully completed its document version upload (version 23 verified).

4. Provider Key Synchronization & Local Route Alignment

Section titled “4. Provider Key Synchronization & Local Route Alignment”

The sanctum doctor CLI command reported failures on gemini and mlx_local.

The Keychain entry google-ai-api-key contained an expired key that failed with 400 INVALID_ARGUMENT. Bert validated the live secret at ~/.sanctum/secrets/gemini-api-key against Google’s API (200 OK) and updated the macOS Keychain in-place:

Terminal window
KEY=$(tr -d '[:space:]' < ~/.sanctum/secrets/gemini-api-key)
security add-generic-password -U -a sanctum -s google-ai-api-key -w "$KEY"

sanctum-cli had defaulted to legacy port 8900 for mlx_local. Bert configured ~/.sanctum/instance.yaml under cli.providers:

providers:
claude:
via: proxy
endpoint: http://127.0.0.1:3456
mlx_local:
endpoint: http://127.0.0.1:8055
model: council-secure

Probing sanctum-server on 127.0.0.1:8055 confirmed HealthSnapshot(ok=True, latency_ms=1).

To eliminate passphrase prompts during unattended scripts, Bert updated ~/.sanctum/ssh/config for openclaw, 10.10.10.10, and yoda to place ~/.ssh/sanctum_automation_ed25519 first with IdentityAgent none. The configuration was committed and pushed upstream to sanctum-config (1b8dc40).

All four operational suites confirmed full estate health:

  • Qui-Gon Infrastructure Doctor (quigon-doctor.py): 16/16 checks passing (0 failing).
  • Sanctum Connection Tests (test-connections.sh): 26/26 passing (0 failed).
  • Council Parity Smoke (parity-smoke.sh): 10/10 prompts verified on MLX :1337.
  • Sanctum Doctor: 3/3 providers operational (claude 3ms, gemini 177ms, mlx_local 0ms; 0 failed).