Summary

sqlmesh dlt_refresh fails with Error: Unknown connection type 'service_account' when the dlt pipeline uses BigQuery with service-account credentials.

Root cause

format_config initialises config = {"type": db_type} (e.g. "bigquery") then iterates all credential object attributes and writes them into config. BigQuery service-account credentials carry a "type": "service_account" field (from the GCP service account JSON format). This overwrites the "bigquery" key. parse_connection_config then raises:

ConfigError: Unknown connection type 'service_account'

This error fires before model generation starts, even though the direct caller (generate_dlt_models) discards the connection config entirely — it is the _ in:

sqlmesh_models, _, _ = generate_dlt_models_and_settings(...)

The bug only surfaces with BigQuery service-account credentials. OAUTH credentials do not carry a conflicting type field, so anyone who tested dlt_refresh with OAUTH would not have seen it.

Fix

Move config["type"] = db_type to after the credential loop so it cannot be overwritten by credential attributes.

 def format_config(configs: t.Dict[str, str], db_type: str) -> str:
     """Generate a string for the gateway connection config."""
-    config = {
-        "type": db_type,
-    }
-
+    config: t.Dict[str, t.Any] = {}
+
     for key, value in configs.items():
         if key == "password":
             config[key] = f'"{value}"'
         elif key == "username":
             config["user"] = value
         else:
             config[key] = value
 
+    config["type"] = db_type
+
     # Validate the connection config fields

Reproduction

1. Create a dlt pipeline targeting BigQuery using service-account credentials
2. Run sqlmesh dlt_refresh <pipeline_name>
3. Observe: Error: Unknown connection type 'service_account'

Does not affect OAUTH credentials.