apache/beam · error · IllegalArgumentException

Unsupported floating-point precision: ${type.getPrecision().

Error message

Unsupported floating-point precision: ${type.getPrecision().name()}

What it means

Thrown when converting an Arrow FloatingPoint type to a Beam FieldType and the precision is neither SINGLE nor DOUBLE (i.e. HALF float). Beam's schema model in this converter only supports 32-bit FLOAT and 64-bit DOUBLE.

Source

Thrown at sdks/java/extensions/arrow/src/main/java/org/apache/beam/sdk/extensions/arrow/ArrowConversion.java:201

                      case 32:
                        return FieldType.INT32;
                      case 64:
                        return FieldType.INT64;
                      default:
                        throw new IllegalArgumentException(
                            "Unsupported integer bit width: " + type.getBitWidth());
                    }
                  }

                  @Override
                  public FieldType visit(ArrowType.FloatingPoint type) {
                    switch (type.getPrecision()) {
                      case SINGLE:
                        return FieldType.FLOAT;
                      case DOUBLE:
                        return FieldType.DOUBLE;
                      default:
                        throw new IllegalArgumentException(
                            "Unsupported floating-point precision: " + type.getPrecision().name());
                    }
                  }

                  @Override
                  public FieldType visit(ArrowType.Utf8 type) {
                    return FieldType.STRING;
                  }

                  @Override
                  public FieldType visit(ArrowType.Utf8View type) {
                    throw new IllegalArgumentException(
                        "Type \'" + type.toString() + "\' not supported.");
                  }

                  @Override
                  public FieldType visit(ArrowType.Binary type) {
                    return FieldType.BYTES;

View on GitHub (pinned to 12126d8942)

Solutions

Change the Arrow column to FLOAT (SINGLE) or DOUBLE precision before conversion.
Pre-convert HALF floats to DOUBLE in the producer or with an Arrow compute cast.
Patch the visitor to map HALF to FieldType.FLOAT or DOUBLE (accepting precision loss) if upstream Beam does not support half floats.

Example fix

// before
Field v = Field.nullable("v", ArrowType.FloatingPoint.HALF);
// after
Field v = Field.nullable("v", ArrowType.FloatingPoint.DOUBLE); // SINGLE also works

Defensive patterns

Strategy: validation

Validate before calling

for (Field f : schema.getFields()) {
  if (f.getType() instanceof ArrowType.FloatingPoint
      && ((ArrowType.FloatingPoint) f.getType()).getPrecision() == FloatingPointPrecision.HALF) {
    throw new IllegalArgumentException(
        "Column " + f.getName() + " is HALF float; cast to SINGLE or DOUBLE before conversion");
  }
}

Type guard

boolean isConvertibleFloat(ArrowType t) {
  return t instanceof ArrowType.FloatingPoint
      && ((ArrowType.FloatingPoint) t).getPrecision()
          != FloatingPointPrecision.HALF;
}

Try / catch

try {
  Schema beamSchema = ArrowConversion.toBeamSchema(arrowSchema);
} catch (IllegalArgumentException e) {
  if (e.getMessage().startsWith("Unsupported floating-point precision")) {
    arrowSchema = castHalfToDouble(arrowSchema); // then retry
  } else { throw e; }
}

Prevention

Keep FP16 data outside Beam ingestion paths; cast to float/double at export time.
Document supported Arrow types for your pipeline and validate input files early.
Test conversion with a sample batch before running full pipelines.

When it happens

Trigger: Converting an Arrow schema/VectorSchemaRoot containing a FloatingPoint type with precision FloatingPointPrecision.HALF (16-bit float) to a Beam Schema.

Common situations: Half-precision columns produced by ML pipelines (FP16 model weights/embeddings), or GPU-generated Arrow data, fed into a Beam ArrowIO pipeline.

Related errors

AI-assisted analysis of apache/beam@12126d8942 (2026-09-13). Data as JSON: /api/errors/f3d33e86520ff148. Report an issue: GitHub.