fix(models): stage downloads on the destination filesystem

.github/workflows/hosting-runtime.yml

@@ -7,6 +7,7 @@ on:
       - k8s/local-hosting/**
       - observability.py
       - main.py
+      - utils/model_download.py
       - .github/workflows/hosting-runtime.yml
   workflow_dispatch: {}
 

.github/workflows/model-install-tests.yml

@@ -0,0 +1,16 @@
+name: Model installation regression
+on:
+  pull_request:
+  push:
+    branches: [main]
+permissions:
+  contents: read
+jobs:
+  model-install:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+      - run: python -m unittest discover -s tests -p test_model_download.py -v

k8s/local-hosting/Dockerfile

@@ -8,6 +8,7 @@ RUN python -c 'from backends.onnx.backend import ONNXBackend; from PIL import Im
 
 RUN python -m pip install --no-cache-dir sentry-sdk==2.73.0
 COPY observability.py main.py /app/
+COPY utils/model_download.py /app/utils/model_download.py
 ARG SENTRY_RELEASE
 ENV SENTRY_DSN=https://8bd44d815901211012fcae52d51371f9@o4512337837686784.ingest.us.sentry.io/4512337894703104 \
     SENTRY_ENVIRONMENT=production \

tests/test_model_download.py

@@ -0,0 +1,37 @@
+"""Exercise installation across actual filesystem boundaries, without a network."""
+import io
+import tempfile
+import unittest
+import zipfile
+from pathlib import Path
+from unittest.mock import patch
+from utils.model_download import ensure_tflite_ssd_mobilenet_v1, ModelDownloadError
+
+
+class ModelInstallTest(unittest.TestCase):
+    def archive(self):
+        data = io.BytesIO()
+        with zipfile.ZipFile(data, 'w') as z:
+            z.writestr('detect.tflite', b'model-fixture')
+        return data.getvalue()
+
+    def test_cross_filesystem_install(self):
+        if not Path('/dev/shm').is_dir():
+            self.skipTest('requires a separate writable tmpfs')
+        with tempfile.TemporaryDirectory(dir='/dev/shm') as directory:
+            root = Path(directory)
+            self.assertNotEqual(root.stat().st_dev, Path(tempfile.gettempdir()).stat().st_dev)
+            with patch('urllib.request.urlopen', return_value=io.BytesIO(self.archive())):
+                ensure_tflite_ssd_mobilenet_v1(model_path=str(root/'model'), labels_path=str(root/'labels'))
+            self.assertEqual((root/'model').read_bytes(), b'model-fixture')
+            self.assertFalse(list(root.glob('.light-object-detect-*')))
+
+    def test_failed_refresh_keeps_existing_model(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            (root/'model').write_bytes(b'known-good')
+            with patch('urllib.request.urlopen', side_effect=OSError('interrupted')):
+                with self.assertRaises(ModelDownloadError):
+                    ensure_tflite_ssd_mobilenet_v1(model_path=str(root/'model'), labels_path=str(root/'labels'), force=True)
+            self.assertEqual((root/'model').read_bytes(), b'known-good')
+            self.assertFalse(list(root.glob('.light-object-detect-*')))

utils/model_download.py

@@ -143,7 +143,7 @@ def ensure_tflite_ssd_mobilenet_v1(
     model_file.parent.mkdir(parents=True, exist_ok=True)
 
     try:
-        with tempfile.TemporaryDirectory(prefix="light-object-detect-") as tmpdir:
+        with tempfile.TemporaryDirectory(prefix=".light-object-detect-", dir=model_file.parent) as tmpdir:
             tmp_dir = Path(tmpdir)
             zip_dest = tmp_dir / "model.zip"
 
@@ -169,7 +169,7 @@ def ensure_tflite_ssd_mobilenet_v1(
             if not extracted_path.exists():
                 raise ModelDownloadError("Extraction succeeded but file is missing on disk.")
 
-            # Atomic-ish replace
+            # Staging and destination share a filesystem, so replacement is atomic.
             tmp_model = tmp_dir / "detect.tmp.tflite"
             tmp_model.write_bytes(extracted_path.read_bytes())
             tmp_model.replace(model_file)