Repository navigation
Databricks runs fail to execute on 0.235.4 with thrift error #5841
Description
Summary
When running with SQLMesh 0.235.4 and Databricks SQL Connector, existing environments fail to run in Databricks with the error:
ERROR:databricks.sql.thrift_backend:ThriftBackend.attempt_request: Exception: encoding without a string argument.
Downgrading to 0.235.3 resolves this issue.
Root Cause
The underlying model SQL I can see being executed successfully against a Databricks Warehouse, the failure occurs when acquiring the built tables' column types.
I believe that this has been caused by PRs #5783 and #5816, as when I remove the DatabricksEngineAdapter.columns method added by those PRs to sqlmesh/core/engine_adapter/databricks.py this error does not occur.
Potentially the fix would be to change line 434 from:
self.cursor.execute(query)to:
self.cursor.execute(query.sql(dialect="databricks"))Environment Tested
Environments:
- Type: Databricks Cluster 17.3 LTS
Connection mode: Databricks SQL Connector
Python: 3.12.3 - Type: Windows 11
Connection mode: Databricks SQL Connector
Python: 3.12.3
The Databricks SQL Connector and thrift versions do not make a difference in success, it's purely whether version 0.235.3 or 0.235.4 is running.
Activity
cschanhniem commentedon Jun 19, 2026
Opened PR #5849 to fix this. The issue is that
self.cursor.execute(query)passes a sqlglot expression object to the Databricks SQL Connector's thrift cursor, which expects a string. Fixed by usingquery.sql(dialect=self.dialect)to render the SQL string first.davem-bis commentedon Jun 19, 2026
Thanks @cschanhniem - I came to the same conclusion and had opened PR #5848 as well.