Enterprise Azure storage_options + sync LanceDB stack to data plane

geneva_examples/core/config.py

@@ -62,6 +62,8 @@ class Config:
     s3_endpoint: str | None = None
     s3_region: str | None = None
     aws_allow_http: bool = False
+    azure_account_name: str | None = None
+    azure_account_key: str | None = None
     hf_token: str | None = None
 
     @property
@@ -70,7 +72,18 @@ def is_local(self) -> bool:
         return self.mode == "local"
 
     def storage_options(self) -> dict[str, str] | None:
-        """Build S3 ``storage_options``; ``None`` unless all four creds present."""
+        """Build object-store ``storage_options`` for the data plane.
+
+        Azure (account name + key) takes precedence; else S3-compatible when all
+        four creds are present; else ``None``. Enterprise clients write Lance
+        data fragments directly to object storage, so the account-less
+        ``az://<container>`` root URI needs the account named here.
+        """
+        if self.azure_account_name and self.azure_account_key:
+            return {
+                "azure_storage_account_name": self.azure_account_name,
+                "azure_storage_account_key": self.azure_account_key,
+            }
         if not (
             self.s3_access_key
             and self.s3_secret_key
@@ -155,5 +168,7 @@ def load_config(
         s3_endpoint=data.get("s3_endpoint"),
         s3_region=data.get("s3_region"),
         aws_allow_http=_as_bool(data.get("aws_allow_http")),
+        azure_account_name=data.get("azure_account_name"),
+        azure_account_key=data.get("azure_account_key"),
         hf_token=data.get("hf_token") or None,
     )

pyproject.toml

@@ -7,13 +7,15 @@ dependencies = [
     "typer>=0.24.0",
     "pyyaml>=6.0",
     # Pin the LanceDB stack to the versions the deployed cluster was built
-    # against (client/server must match or declare_table 500s). pylance 8.0.x
-    # lives on the lance-format index (see [tool.uv.sources]). pyarrow/numpy are
-    # left to geneva's own constraints. geneva 0.13.0b17+ adds source_task_size
-    # to chunker-view refresh; b18 requires pylance>=8.0.0b11.
-    "geneva==0.13.0b18",
-    "lancedb==0.33.1b2",
-    "pylance==8.0.0b16",
+    # against (client/server must match or declare_table/backfill_column 500s).
+    # pylance 9.0.x lives on the lance-format index (see [tool.uv.sources]).
+    # pyarrow/numpy are left to geneva's own constraints. This set matches the
+    # data plane's geneva 0.13.0b30 build (see sophon lancedb_integtest pins):
+    # older pylance (8.x)/lancedb (0.33) spoke a different namespace API, so
+    # add_columns would succeed but backfill_column returned "column not found".
+    "geneva==0.13.0b30",
+    "lancedb==0.34.0b5",
+    "pylance>=9.0.0b15",
     "pyarrow",
     "numpy",
     "pillow>=10",