Delete validate.py

validate.py

@@ -1,633 +0,0 @@
-#!/usr/bin/env python3
-"""
-Pre-flight validation script for MinerU OCR Service.
-
-Run by entrypoint.sh BEFORE uvicorn starts.
-Exits 0 if all checks pass.
-Exits 1 if any CRITICAL check fails — this crashes the container loudly
-so Hugging Face logs show an actionable error instead of a silent crash
-or a healthy-looking service that fails on every request.
-
-Usage:
-    python validate.py           # run all checks, exit 0/1
-    python validate.py --soft    # run all checks, always exit 0 (log only)
-
-── FORENSIC NOTES (2025-06) ──────────────────────────────────────────────────
-
-OCR engine:
-  The pipeline (full) backend uses paddleocr2pytorch — a self-contained
-  PyTorch reimplementation of PaddleOCR bundled inside the magic-pdf wheel.
-  It uses: torch, cv2, numpy, pyclipper, shapely, yaml.
-  paddlepaddle and paddleocr packages are NOT installed and NOT needed.
-
-  pp_structure_v2.py (which imports paddleocr) is only loaded in 'lite' model
-  mode. Pipeline backend always uses 'full' mode (CustomPEKModel). That file is
-  never imported at runtime.
-
-OCR model path resolution (from pytorch_paddle.py):
-  ocr_models_dir = os.path.join(get_local_models_dir(), 'OCR', 'paddleocr_torch')
-  det_model_path = os.path.join(ocr_models_dir, det_filename)
-  where det_filename comes from models_config.yml keyed by language.
-
-  Default CPU path: lang='ch' → forced to 'ch_lite' on CPU device.
-  After Dockerfile Layer 3.5 patch:
-    ch_lite.det = ch_PP-OCRv5_det_infer.pth  (was ch_PP-OCRv3 — not in HF repo)
-    ch_lite.rec = ch_PP-OCRv5_rec_infer.pth  (unchanged — already in HF repo)
-
-Arch config lookup (from pytorchocr_utility.py):
-  get_arch_config(model_path) uses Path(model_path).stem as the key into
-  arch_config.yaml (bundled in magic-pdf wheel). Both replacement filenames
-  have entries in arch_config.yaml — verified before patch was written.
-
-OpenCV conflict handling:
-  doclayout-yolo, ultralytics, and rapid-table all declare opencv-python
-  (non-headless) as a required dep. pip installs the full build in Layer 3.
-  Layer 4 force-reinstalls opencv-python-headless to overwrite cv2. Both
-  packages expose an identical cv2 API so all callers work correctly at
-  runtime. pip-check shows warnings but they are harmless.
-
-onnxruntime:
-  rapid-table declares onnxruntime>1.17.0 as a required (non-optional) dep.
-  pip resolves it automatically when magic-pdf[full] is installed in Layer 3.
-
-slanet-plus.onnx (table model):
-  Bundled inside the magic-pdf wheel at:
-    magic_pdf/resources/slanet_plus/slanet-plus.onnx
-  NOT downloaded from HF Hub — no separate download needed.
-"""
-
-import importlib
-import json
-import os
-import shutil
-import sys
-import tempfile
-import traceback
-
-SOFT_MODE = "--soft" in sys.argv  # never exit 1, just print
-
-MODELS_DIR = "/app/models"
-EXTRACT_KIT_MODELS = os.path.join(MODELS_DIR, "PDF-Extract-Kit-1.0", "models")
-LAYOUT_MARKER = os.path.join(EXTRACT_KIT_MODELS, "Layout")  # canary directory
-CONFIG_PATH = os.path.expanduser("~/magic-pdf.json")
-
-
-# ── helpers ────────────────────────────────────────────────────────────────────
-def ok(label: str, detail: str = "") -> None:
-    suffix = f"  ({detail})" if detail else ""
-    print(f"  ✓  {label}{suffix}", flush=True)
-
-
-def fail(label: str, detail: str, critical: bool = True) -> None:
-    tag = "CRITICAL" if critical else "WARNING"
-    print(f"  ✗  [{tag}] {label}: {detail}", flush=True)
-
-
-def section(title: str) -> None:
-    print(f"\n{'─' * 60}", flush=True)
-    print(f"  {title}", flush=True)
-    print(f"{'─' * 60}", flush=True)
-
-
-# ── check registry ─────────────────────────────────────────────────────────────
-failures: list[tuple[str, str]] = []
-warnings: list[tuple[str, str]] = []
-
-
-def record_fail(label: str, detail: str, critical: bool = True) -> None:
-    fail(label, detail, critical)
-    if critical:
-        failures.append((label, detail))
-    else:
-        warnings.append((label, detail))
-
-
-# ═══════════════════════════════════════════════════════════════════════════════
-print("\n" + "═" * 60, flush=True)
-print("  MinerU OCR Service — Pre-flight Validation", flush=True)
-print("═" * 60, flush=True)
-
-# ── 1. Python version ──────────────────────────────────────────────────────────
-section("1. Python runtime")
-pv = sys.version_info
-if pv >= (3, 10):
-    ok("Python version", f"{pv.major}.{pv.minor}.{pv.micro}")
-else:
-    record_fail("Python version",
-                f"{pv.major}.{pv.minor} detected — magic-pdf requires >= 3.10")
-
-# ── 2. cv2 ─────────────────────────────────────────────────────────────────────
-section("2. OpenCV (cv2)")
-try:
-    import cv2
-    ok("cv2 import", f"version {cv2.__version__}")
-    build = cv2.getBuildInformation()
-    if "GTK" in build or "Qt" in build:
-        record_fail("cv2 build", "GUI backend detected — use opencv-python-headless",
-                    critical=False)
-    else:
-        ok("cv2 headless", "no GUI backend detected")
-except ImportError as exc:
-    record_fail(
-        "cv2 import",
-        f"{exc}. "
-        "Layer 4 force-reinstall of opencv-python-headless may have failed. "
-        "Check Docker build log for the 'pip install --force-reinstall opencv-python-headless' step.",
-    )
-except Exception as exc:
-    record_fail("cv2 import", f"unexpected error: {exc}")
-
-# ── 3. PyTorch ─────────────────────────────────────────────────────────────────
-section("3. PyTorch + TorchVision")
-try:
-    import torch
-    ok("torch import", f"version {torch.__version__}")
-    if torch.cuda.is_available():
-        record_fail("torch CUDA", "CUDA detected on CPU-only space — unexpected",
-                    critical=False)
-    else:
-        ok("torch device", "CPU-only (expected for free tier)")
-except ImportError as exc:
-    record_fail(
-        "torch import",
-        f"{exc}. "
-        "Install from PyTorch CPU index BEFORE magic-pdf in Dockerfile Layer 2: "
-        "pip install --index-url https://download.pytorch.org/whl/cpu torch torchvision",
-    )
-except Exception as exc:
-    record_fail("torch import", f"unexpected: {exc}")
-
-try:
-    import torchvision
-    ok("torchvision import", f"version {torchvision.__version__}")
-except ImportError as exc:
-    record_fail("torchvision import", str(exc))
-except Exception as exc:
-    record_fail("torchvision import", f"unexpected: {exc}")
-
-# ── 4. ultralytics ─────────────────────────────────────────────────────────────
-section("4. ultralytics (YOLO — required by doclayout_yolo)")
-try:
-    import ultralytics
-    ok("ultralytics import", f"version {ultralytics.__version__}")
-except ImportError as exc:
-    record_fail(
-        "ultralytics import",
-        f"{exc}. "
-        "Provided by magic-pdf[full]. "
-        "ROOT CAUSE: [full-cpu] is NOT a valid extra in magic-pdf 1.3.12 — "
-        "pip silently installed only the base package when given an unknown extra. "
-        "Dockerfile Layer 3 must use magic-pdf[full]==1.3.12 (not [full-cpu]).",
-    )
-except Exception as exc:
-    record_fail("ultralytics import", f"unexpected: {exc}")
-
-# ── 5. doclayout_yolo ──────────────────────────────────────────────────────────
-section("5. doclayout_yolo (layout detection model)")
-try:
-    import doclayout_yolo
-    ok("doclayout_yolo import", f"version {getattr(doclayout_yolo, '__version__', 'unknown')}")
-except ImportError as exc:
-    record_fail(
-        "doclayout_yolo import",
-        f"{exc}. "
-        "Provided by magic-pdf[full] (version 0.0.2b1). "
-        "doclayout-yolo==0.0.2b1 is only on the myhloli custom wheel index — "
-        "Dockerfile Layer 3 must include: "
-        "--extra-index-url https://myhloli.github.io/wheels/",
-    )
-except Exception as exc:
-    record_fail("doclayout_yolo import", f"unexpected: {exc}")
-
-# ── 6. rapid_table ─────────────────────────────────────────────────────────────
-section("6. rapid_table (table extraction)")
-try:
-    import rapid_table
-    ok("rapid_table import", f"version {getattr(rapid_table, '__version__', 'unknown')}")
-except ImportError as exc:
-    record_fail(
-        "rapid_table import",
-        f"{exc}. Provided by magic-pdf[full]. Check Layer 3 install.",
-    )
-except Exception as exc:
-    record_fail("rapid_table import", f"unexpected: {exc}")
-
-# ── 7. onnxruntime ─────────────────────────────────────────────────────────────
-section("7. onnxruntime (required by rapid-table for table model inference)")
-# onnxruntime is a required (non-optional) dep of rapid-table>=1.0.5.
-# pip resolves it automatically when magic-pdf[full] is installed in Layer 3.
-# If it is missing it means rapid-table itself failed to install.
-try:
-    import onnxruntime
-    ok("onnxruntime import", f"version {onnxruntime.__version__}")
-except ImportError as exc:
-    record_fail(
-        "onnxruntime import",
-        f"{exc}. "
-        "onnxruntime is a required dep of rapid-table>=1.0.5. "
-        "Its absence means rapid-table failed to install in Layer 3. "
-        "Check Docker build log for rapid-table install errors.",
-    )
-except Exception as exc:
-    record_fail("onnxruntime import", f"unexpected: {exc}")
-
-# ── 8. magic_pdf core imports ──────────────────────────────────────────────────
-section("8. magic_pdf core imports")
-
-REQUIRED_IMPORTS = [
-    ("magic_pdf.data.dataset",                     ["PymuDocDataset", "ImageDataset"]),
-    ("magic_pdf.data.data_reader_writer",          ["FileBasedDataReader", "FileBasedDataWriter"]),
-    ("magic_pdf.model.doc_analyze_by_custom_model", ["doc_analyze"]),
-    ("magic_pdf.config.enums",                     ["SupportedPdfParseMethod"]),
-]
-
-for module_path, symbols in REQUIRED_IMPORTS:
-    try:
-        mod = importlib.import_module(module_path)
-        missing = [s for s in symbols if not hasattr(mod, s)]
-        if missing:
-            record_fail(f"{module_path}", f"missing symbols: {missing}")
-        else:
-            ok(module_path, ", ".join(symbols))
-    except ImportError as exc:
-        record_fail(module_path, str(exc))
-    except Exception as exc:
-        record_fail(module_path, f"unexpected: {exc}")
-
-# ── 8b. paddleocr2pytorch (OCR engine bundled inside magic-pdf wheel) ──────────
-section("8b. paddleocr2pytorch (PyTorch OCR — bundled in magic-pdf wheel)")
-# This is the actual OCR engine for the pipeline backend.
-# It is NOT a separate pip package — it lives inside the magic-pdf wheel at
-#   magic_pdf/model/sub_modules/ocr/paddleocr2pytorch/
-# If it is missing, the entire magic-pdf package did not install correctly.
-try:
-    from magic_pdf.model.sub_modules.ocr.paddleocr2pytorch.pytorch_paddle import PytorchPaddleOCR
-    ok("PytorchPaddleOCR (paddleocr2pytorch)", "bundled inside magic-pdf wheel — no paddlepaddle pkg needed")
-except ImportError as exc:
-    record_fail(
-        "PytorchPaddleOCR import",
-        f"{exc}. "
-        "This module is bundled inside magic_pdf/model/sub_modules/ocr/paddleocr2pytorch/. "
-        "If missing, magic-pdf itself did not install correctly.",
-    )
-except Exception as exc:
-    record_fail("PytorchPaddleOCR import", f"unexpected: {exc}")
-
-# ── 8c. Deprecated API check ───────────────────────────────────────────────────
-section("8c. Deprecated API check (should NOT exist)")
-OBSOLETE = [
-    "magic_pdf.pipe.UNIPipe",
-    "magic_pdf.rw.DiskReaderWriter",
-]
-for mod_path in OBSOLETE:
-    try:
-        importlib.import_module(mod_path)
-        record_fail(mod_path, "still importable — code may use old API", critical=False)
-    except ImportError:
-        ok(f"{mod_path} (correctly absent)")
-
-# ── 9. End-to-end pipeline smoke test ─────────────────────────────────────────
-section("9. End-to-end pipeline smoke test")
-try:
-    from magic_pdf.model.doc_analyze_by_custom_model import doc_analyze  # noqa: F401
-    import ultralytics  # noqa: F401
-    from magic_pdf.data.dataset import ImageDataset  # noqa: F401
-    from magic_pdf.data.data_reader_writer import FileBasedDataReader, FileBasedDataWriter  # noqa: F401
-    ok("Pipeline imports (doc_analyze + ultralytics + ImageDataset + readers)", "all OK")
-except ImportError as exc:
-    record_fail(
-        "Pipeline smoke test",
-        f"Full pipeline import chain failed: {exc}. "
-        "This means POST /extract will fail on every request.",
-    )
-except Exception as exc:
-    record_fail("Pipeline smoke test", f"unexpected: {exc}")
-
-# ── 10. Config file ────────────────────────────────────────────────────────────
-section("10. MinerU config (magic-pdf.json)")
-_cfg: dict = {}
-if os.path.exists(CONFIG_PATH):
-    try:
-        with open(CONFIG_PATH) as f:
-            _cfg = json.load(f)
-        required_keys = ["models-dir", "device-mode"]
-        missing_keys = [k for k in required_keys if k not in _cfg]
-        if missing_keys:
-            record_fail("Config keys", f"missing: {missing_keys}")
-        else:
-            ok("Config file", CONFIG_PATH)
-            ok("device-mode", _cfg.get("device-mode", "?"))
-            ok("models-dir",  _cfg.get("models-dir",  "?"))
-            ok("formula-enable", str(_cfg.get("formula-config", {}).get("enable", "?")))
-            ok("table-enable",   str(_cfg.get("table-config",   {}).get("enable", "?")))
-    except json.JSONDecodeError as exc:
-        record_fail("Config file", f"invalid JSON: {exc}")
-    except Exception as exc:
-        record_fail("Config file", str(exc))
-else:
-    record_fail(
-        "Config file",
-        f"not found at {CONFIG_PATH}. "
-        "Run download_models.py or check Docker build log.",
-    )
-
-# ── 11. Model directory structure ─────────────────────────────────────────────
-section("11. Model directory structure")
-
-model_dir_checks = [
-    ("PDF-Extract-Kit-1.0 root",  os.path.join(MODELS_DIR, "PDF-Extract-Kit-1.0")),
-    ("Layout models",             os.path.join(EXTRACT_KIT_MODELS, "Layout")),
-    ("Layout/YOLO",               os.path.join(EXTRACT_KIT_MODELS, "Layout", "YOLO")),
-    ("OCR models",                os.path.join(EXTRACT_KIT_MODELS, "OCR")),
-    ("OCR/paddleocr_torch",       os.path.join(EXTRACT_KIT_MODELS, "OCR", "paddleocr_torch")),
-    ("Table models (TabRec)",     os.path.join(EXTRACT_KIT_MODELS, "TabRec")),
-]
-
-for label, path in model_dir_checks:
-    if os.path.isdir(path):
-        try:
-            n = sum(1 for _ in os.scandir(path))
-            ok(label, f"{n} entries  [{path}]")
-        except OSError:
-            ok(label, path)
-    else:
-        record_fail(label, f"directory not found: {path}")
-
-lr_dir = os.path.join(MODELS_DIR, "layoutreader")
-if os.path.isdir(lr_dir):
-    ok("layoutreader (optional)", lr_dir)
-else:
-    record_fail("layoutreader (optional)",
-                "not found — MinerU will use fallback ordering (non-critical)",
-                critical=False)
-
-# ── 11b. Critical model weight files ──────────────────────────────────────────
-section("11b. Critical model weight files")
-#
-# These are the EXACT files MinerU will try to open when processing a document
-# on a CPU deployment (default language = ch → forced to ch_lite on CPU).
-#
-# After Dockerfile Layer 3.5 patch, models_config.yml now references:
-#   ch_lite.det = ch_PP-OCRv5_det_infer.pth  (patched from v3 — v3 NOT in repo)
-#   ch_lite.rec = ch_PP-OCRv5_rec_infer.pth  (unchanged — always in repo)
-#
-# Layout uses doclayout_yolo (from magic-pdf.json layout-config).
-# Table (rapid_table) uses slanet-plus.onnx BUNDLED IN THE WHEEL — not here.
-# Formula is DISABLED — MFD/MFR files not required.
-#
-# Any CRITICAL failure here = service boots but crashes on first document.
-
-_ocr_dir = os.path.join(EXTRACT_KIT_MODELS, "OCR", "paddleocr_torch")
-
-CRITICAL_WEIGHT_FILES: list[tuple[str, str, str]] = [
-    # (label, relative-to-EXTRACT_KIT_MODELS, reason)
-    (
-        "OCR det weight (ch_lite, default CPU lang)",
-        os.path.join("OCR", "paddleocr_torch", "ch_PP-OCRv5_det_infer.pth"),
-        "Patched from ch_PP-OCRv3_det_infer.pth (absent in HF repo). "
-        "Missing = all OCR will crash at model load time."
-    ),
-    (
-        "OCR rec weight (ch_lite)",
-        os.path.join("OCR", "paddleocr_torch", "ch_PP-OCRv5_rec_infer.pth"),
-        "Recognition model for ch_lite. "
-        "Missing = OCR loads det but crashes at recognition."
-    ),
-    (
-        "OCR cls weight (angle classifier)",
-        os.path.join("OCR", "paddleocr_torch", "ch_ptocr_mobile_v2.0_cls_infer.pth"),
-        "Used when use_angle_cls=True. Default is False so non-critical, "
-        "but its absence causes crash if angle classification is enabled."
-    ),
-    (
-        "Layout YOLO weight (doclayout_yolo)",
-        os.path.join("Layout", "YOLO", "doclayout_yolo_docstructbench_imgsz1280_2501.pt"),
-        "Layout detection model. Missing = layout detection crashes on every document."
-    ),
-    (
-        "Layout LayoutLMv3 weight",
-        os.path.join("Layout", "LayoutLMv3", "model_final.pth"),
-        "Alternative layout model. Required even when doclayout_yolo is primary "
-        "because model_configs.yaml always lists it."
-    ),
-    (
-        "Multilingual OCR det (en/latin fallback)",
-        os.path.join("OCR", "paddleocr_torch", "Multilingual_PP-OCRv3_det_infer.pth"),
-        "Patched det for en and latin languages. Missing = crash if lang=en/latin."
-    ),
-]
-
-# cls weight is only critical if use_angle_cls=True (default False)
-NON_CRITICAL_LABELS = {"OCR cls weight (angle classifier)"}
-
-for label, rel_path, reason in CRITICAL_WEIGHT_FILES:
-    full_path = os.path.join(EXTRACT_KIT_MODELS, rel_path)
-    is_critical = label not in NON_CRITICAL_LABELS
-    if os.path.isfile(full_path):
-        size_mb = os.path.getsize(full_path) / (1024 * 1024)
-        ok(label, f"{size_mb:.1f} MB  [{full_path}]")
-    else:
-        record_fail(
-            label,
-            f"FILE NOT FOUND: {full_path}\n"
-            f"    Reason: {reason}",
-            critical=is_critical,
-        )
-
-# ── 11c. models_config.yml consistency check ──────────────────────────────────
-section("11c. models_config.yml consistency check")
-#
-# Reads the installed models_config.yml (inside magic_pdf package) and verifies
-# that every det/rec file it references for the default CPU language (ch_lite)
-# actually exists on disk in the expected location.
-#
-# This catches future version drift between the magic-pdf package and the HF repo
-# BEFORE the service starts, rather than mid-request.
-
-try:
-    import magic_pdf
-    import yaml as _yaml
-    from pathlib import Path as _Path
-
-    _pkg = _Path(magic_pdf.__file__).parent
-    _mcfg = _pkg / 'model/sub_modules/ocr/paddleocr2pytorch/pytorchocr/utils/resources/models_config.yml'
-
-    if not _mcfg.exists():
-        record_fail("models_config.yml", f"not found at expected path: {_mcfg}")
-    else:
-        with open(_mcfg) as _f:
-            _mc = _yaml.safe_load(_f)
-
-        _ocr_torch = os.path.join(EXTRACT_KIT_MODELS, "OCR", "paddleocr_torch")
-
-        # Check the two languages actually used on this CPU deployment
-        _check_langs = ["ch_lite", "ch"]
-        _mc_ok = True
-        for _lang in _check_langs:
-            _entry = _mc.get("lang", {}).get(_lang, {})
-            for _field in ("det", "rec"):
-                _fname = _entry.get(_field)
-                if not _fname:
-                    continue
-                _fpath = os.path.join(_ocr_torch, _fname)
-                if os.path.isfile(_fpath):
-                    ok(f"models_config[{_lang}].{_field}", _fname)
-                else:
-                    record_fail(
-                        f"models_config[{_lang}].{_field}",
-                        f"Config references '{_fname}' but file not found at:\n"
-                        f"    {_fpath}\n"
-                        f"    Dockerfile Layer 3.5 patch may not have run, "
-                        f"or HF repo changed its file structure again.",
-                        critical=True,
-                    )
-                    _mc_ok = False
-
-        if _mc_ok:
-            ok("models_config.yml consistency", "all referenced det/rec files exist on disk")
-
-except Exception as _exc:
-    record_fail("models_config.yml consistency check", f"unexpected error: {_exc}", critical=False)
-
-# ── 11d. Bundled wheel resources ──────────────────────────────────────────────
-section("11d. Bundled wheel resources (inside magic_pdf package)")
-#
-# These files are shipped inside the magic-pdf wheel itself.
-# They do NOT come from the HF download. Their absence means the wheel
-# installed incorrectly or was corrupted.
-
-try:
-    import magic_pdf as _mp
-    from pathlib import Path as _P
-
-    _pkg_root = _P(_mp.__file__).parent
-    _bundled = [
-        ("slanet-plus.onnx (table model)",
-         _pkg_root / "resources" / "slanet_plus" / "slanet-plus.onnx"),
-        ("fasttext langdetect model",
-         _pkg_root / "resources" / "fasttext-langdetect" / "lid.176.ftz"),
-        ("YOLO langdetect model",
-         _pkg_root / "resources" / "yolov11-langdetect" / "yolo_v11_ft.pt"),
-        ("model_configs.yaml (weight path map)",
-         _pkg_root / "resources" / "model_config" / "model_configs.yaml"),
-    ]
-    for _lbl, _p in _bundled:
-        if _p.exists():
-            _sz = _p.stat().st_size / (1024 * 1024)
-            ok(_lbl, f"{_sz:.2f} MB")
-        else:
-            record_fail(_lbl, f"expected inside wheel at {_p} — magic-pdf install may be corrupted")
-
-except Exception as _exc:
-    record_fail("Bundled wheel resources check", f"unexpected: {_exc}", critical=False)
-
-# ── 12. Temp storage ───────────────────────────────────────────────────────────
-section("12. Temp storage")
-try:
-    td = tempfile.mkdtemp(prefix="mineru_validate_")
-    test_file = os.path.join(td, "write_test.bin")
-    with open(test_file, "wb") as f:
-        f.write(b"x" * 4096)
-    assert os.path.getsize(test_file) == 4096
-    shutil.rmtree(td)
-    ok("Temp write + delete", tempfile.gettempdir())
-except Exception as exc:
-    record_fail("Temp storage", str(exc))
-
-# ── 13. System memory (cgroups) ────────────────────────────────────────────────
-section("13. System memory (cgroups)")
-mem_source = "unknown"
-total_mb = used_mb = 0
-
-try:
-    with open("/sys/fs/cgroup/memory.max") as f:
-        raw = f.read().strip()
-    if raw != "max":
-        total_mb = int(raw) // (1024 * 1024)
-        with open("/sys/fs/cgroup/memory.current") as f:
-            used_mb = int(f.read().strip()) // (1024 * 1024)
-        mem_source = "cgroups v2"
-except (FileNotFoundError, ValueError):
-    pass
-
-if total_mb == 0:
-    try:
-        with open("/sys/fs/cgroup/memory/memory.limit_in_bytes") as f:
-            limit = int(f.read().strip())
-        with open("/sys/fs/cgroup/memory/memory.usage_in_bytes") as f:
-            used_bytes = int(f.read().strip())
-        if limit < 128 * 1024 * 1024 * 1024:
-            total_mb = limit // (1024 * 1024)
-            used_mb  = used_bytes // (1024 * 1024)
-            mem_source = "cgroups v1"
-    except (FileNotFoundError, ValueError):
-        pass
-
-if total_mb == 0:
-    try:
-        info: dict[str, int] = {}
-        with open("/proc/meminfo") as f:
-            for line in f:
-                parts = line.split()
-                if len(parts) >= 2:
-                    info[parts[0].rstrip(":")] = int(parts[1])
-        total_mb = info.get("MemTotal", 0) // 1024
-        used_mb  = (info.get("MemTotal", 0) - info.get("MemAvailable", 0)) // 1024
-        mem_source = "/proc/meminfo (may show host RAM)"
-    except Exception:
-        pass
-
-ok("Memory source", mem_source)
-ok("Total memory",  f"{total_mb} MB")
-ok("Used memory",   f"{used_mb} MB")
-ok("Free memory",   f"{total_mb - used_mb} MB")
-
-if total_mb > 32 * 1024:
-    record_fail(
-        "Memory total",
-        f"{total_mb} MB seems too large — cgroups may not be available; "
-        "/proc/meminfo showing host RAM. Memory guard in main.py will be conservative.",
-        critical=False,
-    )
-
-# ── 14. /proc/meminfo sanity ───────────────────────────────────────────────────
-section("14. /proc/meminfo (reference)")
-try:
-    with open("/proc/meminfo") as f:
-        lines = f.readlines()[:5]
-    for line in lines:
-        parts = line.split()
-        if len(parts) >= 2:
-            kb = int(parts[1])
-            ok(parts[0].rstrip(":"), f"{kb // 1024} MB")
-except Exception as exc:
-    record_fail("/proc/meminfo", str(exc), critical=False)
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# Summary
-# ═══════════════════════════════════════════════════════════════════════════════
-print("\n" + "═" * 60, flush=True)
-print("  Validation Summary", flush=True)
-print("═" * 60, flush=True)
-
-if warnings:
-    print(f"\n  ⚠  {len(warnings)} warning(s):", flush=True)
-    for label, detail in warnings:
-        print(f"     • {label}: {detail}", flush=True)
-
-if failures:
-    print(f"\n  ✗  {len(failures)} CRITICAL failure(s):", flush=True)
-    for label, detail in failures:
-        print(f"     • {label}: {detail}", flush=True)
-    print("\n  Service will NOT start until these are resolved.", flush=True)
-    print("  Check Dockerfile pip layers and Docker build log.", flush=True)
-    print("═" * 60 + "\n", flush=True)
-    if not SOFT_MODE:
-        sys.exit(1)
-else:
-    print(f"\n  ✓  All critical checks passed", flush=True)
-    if warnings:
-        print(f"  ⚠  {len(warnings)} non-critical warning(s) — see above", flush=True)
-    print("\n  Service is ready to start.", flush=True)
-    print("═" * 60 + "\n", flush=True)
-    sys.exit(0)