Initial curated RAGcore source from c16667f0c9397e2bfc8dd0255b114fba23c8d05e

tests/unit/infrastructure/qdrant/test_qdrant_embedding_selection_verifier.py

@@ -0,0 +1,96 @@
+from __future__ import annotations
+
+from types import SimpleNamespace
+from uuid import uuid4
+
+import pytest
+from qdrant_client.http import models
+
+from ragcore.application.models.selection import EmbeddingCutoverRejected
+from ragcore.domain.control import DataPolicy
+from ragcore.domain.models import EmbeddingProfile, EmbeddingProvider, EmbeddingSelection
+from ragcore.infrastructure.qdrant.selection_verifier import QdrantEmbeddingSelectionVerifier
+
+
+def _profile() -> EmbeddingProfile:
+    return EmbeddingProfile(
+        id=uuid4(),
+        profile="selected-index",
+        version="v2",
+        provider=EmbeddingProvider.MODELFORGE,
+        model_identifier="rag.embedding@1",
+        model_digest="b" * 64,
+        dimensions=3,
+        context_limit=4096,
+        tokenizer="capability-managed",
+        data_policy=DataPolicy.LOCAL_ONLY,
+        embedding_space_id=uuid4(),
+    )
+
+
+class FakeQdrant:
+    def __init__(self, *, dimension: int, payloads: list[dict[str, object]]) -> None:
+        self.dimension = dimension
+        self.payloads = payloads
+        self.collections: list[str] = []
+
+    async def get_collection(self, collection_name: str) -> SimpleNamespace:
+        self.collections.append(collection_name)
+        return SimpleNamespace(
+            config=SimpleNamespace(
+                params=SimpleNamespace(
+                    vectors={
+                        "dense": models.VectorParams(
+                            size=self.dimension, distance=models.Distance.COSINE
+                        )
+                    },
+                    sparse_vectors={"sparse": models.SparseVectorParams()},
+                )
+            )
+        )
+
+    async def scroll(self, **_kwargs: object) -> tuple[list[SimpleNamespace], None]:
+        return [SimpleNamespace(payload=value) for value in self.payloads], None
+
+
+def _payload(profile: EmbeddingProfile) -> dict[str, object]:
+    return {
+        "embedding_model_digest": profile.model_digest,
+        "chunk_profile_version": profile.version,
+        "embedding_space_id": str(profile.embedding_space_id),
+    }
+
+
+@pytest.mark.asyncio
+async def test_collection_verifier_rejects_a_stale_vector_dimension() -> None:
+    profile = _profile()
+    selection = EmbeddingSelection.from_profile(profile, experiment_route_key="lab-route-next")
+    client = FakeQdrant(dimension=2, payloads=[_payload(profile)])
+
+    with pytest.raises(EmbeddingCutoverRejected, match="vector shape"):
+        await QdrantEmbeddingSelectionVerifier(client).verify(selection, profile)  # type: ignore[arg-type]
+
+    assert client.collections == ["rag_dense_selected-index_v2"]
+
+
+@pytest.mark.asyncio
+async def test_collection_verifier_rejects_stale_profile_identity_pins() -> None:
+    profile = _profile()
+    selection = EmbeddingSelection.from_profile(profile, experiment_route_key="lab-route-next")
+    stale = {**_payload(profile), "embedding_space_id": str(uuid4())}
+    client = FakeQdrant(dimension=3, payloads=[stale])
+
+    with pytest.raises(EmbeddingCutoverRejected, match="embedding-space pin"):
+        await QdrantEmbeddingSelectionVerifier(client).verify(selection, profile)  # type: ignore[arg-type]
+
+
+@pytest.mark.asyncio
+async def test_collection_verifier_counts_only_the_exact_pinned_collection() -> None:
+    profile = _profile()
+    selection = EmbeddingSelection.from_profile(profile)
+    client = FakeQdrant(dimension=3, payloads=[_payload(profile), _payload(profile)])
+
+    result = await QdrantEmbeddingSelectionVerifier(client).verify(selection, profile)  # type: ignore[arg-type]
+
+    assert client.collections == ["rag_dense_selected-index_v2"]
+    assert result.verified_point_count == 2

tests/unit/infrastructure/qdrant/test_qdrant_selected_collection.py

@@ -0,0 +1,91 @@
+from __future__ import annotations
+
+from types import SimpleNamespace
+from uuid import uuid4
+
+import httpx
+import pytest
+from qdrant_client.http import models
+from qdrant_client.http.exceptions import UnexpectedResponse
+
+from ragcore.application.indexes.models import IndexBuildError
+from ragcore.infrastructure.qdrant.adapter import CollectionProfileMismatchError
+from ragcore.infrastructure.qdrant.models import CollectionKind, CollectionProfile
+from ragcore.infrastructure.qdrant.rebuild import QdrantRebuildStore
+
+
+class MissingQdrantCollection:
+    def __init__(self) -> None:
+        self.reads = 0
+        self.creates = 0
+
+    async def get_collection(self, _name: str) -> object:
+        self.reads += 1
+        raise UnexpectedResponse(404, "Not Found", b"missing", httpx.Headers())
+
+    async def create_collection(self, **_kwargs: object) -> None:
+        self.creates += 1
+
+
+@pytest.mark.asyncio
+async def test_selected_collection_verification_fails_without_recreating_missing_collection() -> (
+    None
+):
+    client = MissingQdrantCollection()
+    profile = CollectionProfile(
+        kind=CollectionKind.DENSE,
+        profile="selected-profile",
+        version="v2",
+        dimensions=4,
+        model_digest="a" * 64,
+        embedding_space_id=uuid4(),
+    )
+
+    with pytest.raises(CollectionProfileMismatchError, match="refusing to recreate"):
+        await QdrantRebuildStore(client).verify_existing_selection(profile)  # type: ignore[arg-type]
+
+    assert client.reads == 1
+    assert client.creates == 0
+
+
+@pytest.mark.asyncio
+async def test_selected_collection_verification_checks_every_existing_point_pin() -> None:
+    space_id = uuid4()
+
+    class StalePinnedCollection:
+        async def get_collection(self, _name: str) -> object:
+            return SimpleNamespace(
+                config=SimpleNamespace(
+                    params=SimpleNamespace(
+                        vectors={
+                            "dense": models.VectorParams(size=4, distance=models.Distance.COSINE)
+                        },
+                        sparse_vectors={"sparse": models.SparseVectorParams()},
+                    )
+                )
+            )
+
+        async def scroll(self, **_kwargs: object) -> tuple[list[SimpleNamespace], None]:
+            return [
+                SimpleNamespace(
+                    payload={
+                        "embedding_model_digest": "b" * 64,
+                        "chunk_profile_version": "v2",
+                        "embedding_space_id": str(space_id),
+                    }
+                )
+            ], None
+
+    profile = CollectionProfile(
+        kind=CollectionKind.DENSE,
+        profile="selected-profile",
+        version="v2",
+        dimensions=4,
+        model_digest="a" * 64,
+        embedding_space_id=space_id,
+    )
+
+    with pytest.raises(IndexBuildError, match="incompatible identity pins"):
+        await QdrantRebuildStore(StalePinnedCollection()).verify_existing_selection(  # type: ignore[arg-type]
+            profile
+        )