[review_comments_retriever] Avoid populating database from scratch

bugbug/tools/code_review.py

@@ -1285,8 +1285,15 @@ def clean_comment(self, comment):
         return comment
 
     def add_comments_by_hunk(self, items: Iterable[tuple[Hunk, InlineComment]]):
+        point_ids = self.vector_db.get_existing_ids()
+
         def vector_points():
+            nonlocal point_ids
+
             for hunk, comment in items:
+                if comment.id in point_ids:
+                    continue
+
                 str_hunk = str(hunk)
                 vector = self.embeddings.embed_query(str_hunk)
                 payload = {

bugbug/vectordb.py

@@ -46,6 +46,10 @@ def insert(self, points: Iterable[VectorPoint]):
     def search(self, query: list[float]) -> Iterable[PayloadScore]:
         ...
 
+    @abstractmethod
+    def get_existing_ids(self):
+        ...
+
 
 class QdrantVectorDB(VectorDB):
     def __init__(self, collection_name: str, *args, **kwargs):
@@ -83,3 +87,25 @@ def insert(self, points: Iterable[VectorPoint]):
     def search(self, query: list[float]) -> Iterable[PayloadScore]:
         for item in self.client.search(self.collection_name, query):
             yield PayloadScore(item.score, item.id, item.payload)
+
+    def get_existing_ids(self):
+        point_ids = set()
+        offset = None
+
+        while True:
+            points, next_page_offset = self.client.scroll(
+                collection_name=self.collection_name,
+                limit=100,
+                with_payload=False,
+                with_vectors=False,
+                offset=offset,
+            )
+
+            point_ids.update(point.id for point in points)
+
+            if not next_page_offset:
+                break
+
+            offset = next_page_offset
+
+        return point_ids