[alphaliujin]

Problem

On container deployments the backend runs as the unprivileged deeptutor user (supervisor user=deeptutor), but HOME is inherited from supervisord, which runs as root. mineru_models_ready() therefore probes /root/.cache/modelscope/hub/models, and calls .is_dir() outside its try/except:

PermissionError: [Errno 13] Permission denied: '/root/.cache/modelscope/hub/models'

build_settings_readiness() catches every exception and records the row as parser_probe_failed, so Settings → Document Parsing shows MinerU as misconfigured with no actionable cause — while MinerU parsing itself works.

Changes

• mineru_models_ready(): resolve the cache roots inside the guarded block and treat an unreadable/absent home as “cannot confirm”, returning False (models_missing). This matches the function's documented fail-closed contract instead of letting a PermissionError escape.
• build_settings_readiness(): log the probe exception with a traceback before recording parser_probe_failed, so future probe failures are diagnosable.
• Dockerfile: create /home/deeptutor and set HOME="/home/deeptutor" for both supervisor programs. This is the root cause; the other two changes keep the failure mode safe for existing images and non-container installs.
• Regression test covering an unreadable cache root (returns False) and a readable cache (returns True).

Verification

• Reproduced on a 1.7.0 container: with HOME=/root, the probe raised PermissionError and the API reported parser.mineru misconfigured / parser_probe_failed.
• With the patch applied, the same scenario returns False, and a populated cache still returns True (manual test in the 1.7.0 container, all other dependencies identical).
• A container started with HOME=/home/deeptutor reports mineru_models_ready() == True, and parser.mineru readiness is enabled_verified.