my_backup: Daily Backup silently failing 8 days

Date: 2026-07-26

What happened

my_backup STATUS check flagged “no backup runs detected in last week.” The Windows Scheduled Task \TaTasks\Daily Backup had been failing (exit code 2) every day from 07-19 through 07-26.

Root cause

.venv’s own pyvenv.cfg showed a Linux Python home — only a WSL uv can write that. Cross-referenced against git log + Core/IT/Logs/2026-07-18_my_backup-notification-email-investigation.md:20: on 2026-07-18 at 16:45, a manual uv run send_status_report (run from WSL, verifying an unrelated fix) rebuilt the project’s default .venv — which lives on the shared D: drive both WSL and Windows touch — as a Linux venv. Windows uv.exe couldn’t sync it afterward (Access is denied removing the Linux lib64 symlink), so the Windows Scheduled Task failed silently every day after.

Not a code regression — the WSL cron path was never at risk (it already used UV_PROJECT_ENVIRONMENT=/home/ta/.venvs/my_backup, off D:). The Windows task called bare uv.exe run my_backup with no override, so it silently depended on the ambiguous shared default path. That architecture was the actual root cause; the manual command was just the trigger.

Fix

1. Deleted the poisoned .venv, rebuilt clean, verified a full backup run (exit 0).
2. Guardrail: gave the Windows Scheduled Task its own isolated UV_PROJECT_ENVIRONMENT (C:\Users\Admin\.venvs\my_backup, off D:) via a new wrapper script D:\FSS\Software\Utils\Windows\my_backup-daily-task.bat, mirroring the isolation WSL already had. Neither production entrypoint touches the shared default .venv anymore — a future stray uv run in the repo dir just poisons an unused folder.
3. Verified the guardrail under the real execution context (schtasks /run, Admin/InteractiveToken), not just a manual re-run: exit 0, isolated venv confirmed Windows-native, unchanged after the run.
4. Disabled Maximizer SQL dump (config.yaml) — deprecated 2026-07-22, Twenty CRM is SSOT — and removed it from the Critical Files health check so it doesn’t alarm forever.
5. Registered my_backup_daily in job_monitor (max_age_hours: 26) — this job was never monitored, which is why an 8-day outage took a week to surface instead of alarming same-day.

Gotchas discovered (promoted to GlobalDevRules.md § 3.9)

• Shared-drive default .venv is a silent WSL/Windows collision trap when neither entrypoint sets UV_PROJECT_ENVIRONMENT.
• schtasks /query /xml output piped through WSL declares encoding="UTF-16" but writes plain single-byte bytes — re-importing via /create /xml fails with “unable to switch the encoding” until properly re-encoded.

Commits

• my_backup e8a85e0 (Maximizer disable), CHANGELOG updated
• job_monitor afe81db (heartbeat registration)
• KB f898e72

Follow-up (deferred, not blocking)

Artifact-freshness checking (independent verifier pattern) for my_backup_daily — Talbot confirmed weekly status-report cadence is sufficient for now.