What causes "UDF returned no columns" in risingwavelabs/risingwave?

Fix the UDF to always return at least the result column

UDF returned no columns

Error message

UDF returned no columns

What it means

After converting the UDF's Arrow RecordBatch into RisingWave's column format, the expression takes the first column as the result. If the batch has zero columns there is no result to return, so it bails. A scalar/vectorized UDF must always produce at least one output column.

Source

Thrown at src/expr/core/src/expr/expr_udf.rs:135

        self.metrics
            .memory_usage_bytes
            .set(self.runtime.memory_usage() as i64);

        let arrow_output = arrow_output_result?;

        if arrow_output.num_rows() != input.cardinality() {
            bail!(
                "UDF returned {} rows, but expected {}",
                arrow_output.num_rows(),
                input.cardinality(),
            );
        }

        let output = self.arrow_convert.from_record_batch(&arrow_output)?;
        let output = output.expand_vis(input.visibility().clone());

        let Some(array) = output.columns().first() else {
            bail!("UDF returned no columns");
        };
        if !array.data_type().equals_datatype(&self.return_type) {
            bail!(
                "UDF returned {:?}, but expected {:?}",
                array.data_type(),
                self.return_type,
            );
        }

        // handle optional error column
        if let Some(errors) = output.columns().get(1) {
            if errors.data_type() != DataType::Varchar {
                bail!(
                    "UDF returned errors column with invalid type: {:?}",
                    errors.data_type()
                );
            }
            let errors = errors

View on GitHub (pinned to 6469eb736d)

Solutions

Fix the UDF to always return at least the result column
Check the remote UDF service response and its serialization
Verify the Arrow->RisingWave conversion path isn't dropping columns
Test the UDF with a simple constant function to isolate where columns are lost

Defensive patterns

Strategy: try-catch

Validate before calling

// In UDF: ensure at least one output column
assert!(!record_batch.columns().is_empty(), "UDF must return a result column");

Type guard

fn has_output_columns(b: &RecordBatch) -> bool { !b.columns().is_empty() }

Try / catch

match res { Err(e) if e.to_string().contains("UDF returned no columns") => log_and_report_udf_bug(&e), Ok(v) => v, Err(e) => return Err(e) }

Prevention

Always append the result column in UDF implementations
Smoke-test UDFs with a trivial constant function
Check remote UDF service responses are non-empty

When it happens

Trigger: A UDF implementation or remote UDF service returns an empty RecordBatch (zero columns) — e.g. the function forgot to append its result column, or a conversion layer dropped all columns.

Common situations: Custom UDFs returning only an error column or nothing; remote UDF services returning malformed payloads; SDK conversion bugs.

Understand the failure class

Background: "empty response", "returned no data", "empty embeddings": what HTTP 200-with-empty-body errors mean across libraries — this error's family across 36 libraries.

Related errors

AI-assisted analysis of risingwavelabs/risingwave@6469eb736d (2026-09-11). Data as JSON: /api/errors/970ccf8cee05f264. Report an issue: GitHub.