private CommandExecutorStat ExecuteDataFrameSqlCommand(
Stream inputStream,
Stream outputStream,
SqlCommand[] commands)
{
var stat = new CommandExecutorStat();
ICommandRunner commandRunner = CreateCommandRunner(commands);
SerDe.Write(outputStream, (int)SpecialLengths.START_ARROW_STREAM);
IpcOptions ipcOptions = ArrowIpcOptions();
ArrowStreamWriter writer = null;
foreach (RecordBatch input in GetInputIterator(inputStream))
{
FxDataFrame dataFrame = FxDataFrame.FromArrowRecordBatch(input);
var inputColumns = new DataFrameColumn[input.ColumnCount];
for (int i = 0; i < dataFrame.Columns.Count; ++i)
{
inputColumns[i] = dataFrame.Columns[i];
}
DataFrameColumn[] results = commandRunner.Run(inputColumns);
var resultDataFrame = new FxDataFrame(results);
IEnumerable <RecordBatch> recordBatches = resultDataFrame.ToArrowRecordBatches();
foreach (RecordBatch result in recordBatches)
{
stat.NumEntriesProcessed += result.Length;
if (writer == null)
{
writer =
new ArrowStreamWriter(outputStream, result.Schema, leaveOpen: true, ipcOptions);
}
// TODO: Remove sync-over-async once WriteRecordBatch exists.
writer.WriteRecordBatchAsync(result).GetAwaiter().GetResult();
}
}
WriteEnd(outputStream, ipcOptions);
writer?.Dispose();
return(stat);
}
private static void TestRoundTripRecordBatch(RecordBatch originalBatch, IpcOptions options = null)
{
using (MemoryStream stream = new MemoryStream())
{
using (var writer = new ArrowStreamWriter(stream, originalBatch.Schema, leaveOpen: true, options))
{
writer.WriteRecordBatch(originalBatch);
writer.WriteEnd();
}
stream.Position = 0;
using (var reader = new ArrowStreamReader(stream))
{
RecordBatch newBatch = reader.ReadNextRecordBatch();
ArrowReaderVerifier.CompareBatches(originalBatch, newBatch);
}
}
}
private CommandExecutorStat ExecuteDataFrameGroupedMapCommand(
Stream inputStream,
Stream outputStream,
SqlCommand[] commands)
{
Debug.Assert(commands.Length == 1,
"Grouped Map UDFs do not support combining multiple UDFs.");
var stat = new CommandExecutorStat();
var worker = (DataFrameGroupedMapWorkerFunction)commands[0].WorkerFunction;
SerDe.Write(outputStream, (int)SpecialLengths.START_ARROW_STREAM);
IpcOptions ipcOptions = ArrowIpcOptions();
ArrowStreamWriter writer = null;
foreach (RecordBatch input in GetInputIterator(inputStream))
{
FxDataFrame dataFrame = FxDataFrame.FromArrowRecordBatch(input);
FxDataFrame resultDataFrame = worker.Func(dataFrame);
IEnumerable <RecordBatch> recordBatches = resultDataFrame.ToArrowRecordBatches();
foreach (RecordBatch batch in recordBatches)
{
RecordBatch final = WrapColumnsInStructIfApplicable(batch);
stat.NumEntriesProcessed += final.Length;
if (writer == null)
{
writer =
new ArrowStreamWriter(outputStream, final.Schema, leaveOpen: true, ipcOptions);
}
writer.WriteRecordBatch(final);
}
}
WriteEnd(outputStream, ipcOptions);
writer?.Dispose();
return(stat);
}
private CommandExecutorStat ExecuteDataFrameGroupedMapCommand(
Stream inputStream,
Stream outputStream,
SqlCommand[] commands)
{
Debug.Assert(commands.Length == 1,
"Grouped Map UDFs do not support combining multiple UDFs.");
var stat = new CommandExecutorStat();
var worker = (DataFrameGroupedMapWorkerFunction)commands[0].WorkerFunction;
SerDe.Write(outputStream, (int)SpecialLengths.START_ARROW_STREAM);
IpcOptions ipcOptions = ArrowIpcOptions();
ArrowStreamWriter writer = null;
foreach (RecordBatch input in GetInputIterator(inputStream))
{
FxDataFrame dataFrame = FxDataFrame.FromArrowRecordBatch(input);
FxDataFrame resultDataFrame = worker.Func(dataFrame);
IEnumerable <RecordBatch> recordBatches = resultDataFrame.ToArrowRecordBatches();
foreach (RecordBatch result in recordBatches)
{
stat.NumEntriesProcessed += result.Length;
if (writer == null)
{
writer =
new ArrowStreamWriter(outputStream, result.Schema, leaveOpen: true, ipcOptions);
}
// TODO: Remove sync-over-async once WriteRecordBatch exists.
writer.WriteRecordBatchAsync(result).GetAwaiter().GetResult();
}
}
WriteEnd(outputStream, ipcOptions);
writer?.Dispose();
return(stat);
}
private CommandExecutorStat ExecuteArrowSqlCommand(
Stream inputStream,
Stream outputStream,
SqlCommand[] commands)
{
var stat = new CommandExecutorStat();
ICommandRunner commandRunner = CreateCommandRunner(commands);
SerDe.Write(outputStream, (int)SpecialLengths.START_ARROW_STREAM);
IpcOptions ipcOptions = ArrowIpcOptions();
ArrowStreamWriter writer = null;
Schema resultSchema = null;
foreach (ReadOnlyMemory <IArrowArray> input in GetArrowInputIterator(inputStream))
{
IArrowArray[] results = commandRunner.Run(input);
// Assumes all columns have the same length, so uses 0th for num entries.
int numEntries = results[0].Length;
stat.NumEntriesProcessed += numEntries;
if (writer == null)
{
Debug.Assert(resultSchema == null);
resultSchema = BuildSchema(results);
writer =
new ArrowStreamWriter(outputStream, resultSchema, leaveOpen: true, ipcOptions);
}
var recordBatch = new RecordBatch(resultSchema, results, numEntries);
// TODO: Remove sync-over-async once WriteRecordBatch exists.
writer.WriteRecordBatchAsync(recordBatch).GetAwaiter().GetResult();
}
WriteEnd(outputStream, ipcOptions);
writer?.Dispose();
return(stat);
}
public void TestPicklingSqlCommandExecutorWithEmptyInput(
Version sparkVersion,
IpcOptions ipcOptions)
{
_ = ipcOptions;
var udfWrapper = new Sql.PicklingUdfWrapper <string, string>((str) => $"udf: {str}");
var command = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.PicklingWorkerFunction(udfWrapper.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var commandPayload = new Worker.CommandPayload()
{
EvalType = UdfUtils.PythonEvalType.SQL_BATCHED_UDF,
Commands = new[] { command }
};
using var inputStream = new MemoryStream();
using var outputStream = new MemoryStream();
// Write test data to the input stream. For the empty input scenario,
// only send SpecialLengths.END_OF_DATA_SECTION.
SerDe.Write(inputStream, (int)SpecialLengths.END_OF_DATA_SECTION);
inputStream.Seek(0, SeekOrigin.Begin);
CommandExecutorStat stat = new CommandExecutor(sparkVersion).Execute(
inputStream,
outputStream,
0,
commandPayload);
// Validate that all the data on the stream is read.
Assert.Equal(inputStream.Length, inputStream.Position);
Assert.Equal(0, stat.NumEntriesProcessed);
// Validate the output stream.
Assert.Equal(0, outputStream.Length);
}
public async Task WriteLegacyIpcFormatAsync(bool writeLegacyIpcFormat)
{
RecordBatch originalBatch = TestData.CreateSampleRecordBatch(length: 100);
var options = new IpcOptions()
{
WriteLegacyIpcFormat = writeLegacyIpcFormat
};
using (MemoryStream stream = new MemoryStream())
{
using (var writer = new ArrowStreamWriter(stream, originalBatch.Schema, leaveOpen: true, options))
{
await writer.WriteRecordBatchAsync(originalBatch);
await writer.WriteEndAsync();
}
stream.Position = 0;
// ensure the continuation is written correctly
byte[] buffer = stream.ToArray();
int messageLength = BinaryPrimitives.ReadInt32LittleEndian(buffer);
int endOfBuffer1 = BinaryPrimitives.ReadInt32LittleEndian(buffer.AsSpan(buffer.Length - 8));
int endOfBuffer2 = BinaryPrimitives.ReadInt32LittleEndian(buffer.AsSpan(buffer.Length - 4));
if (writeLegacyIpcFormat)
{
// the legacy IPC format doesn't have a continuation token at the start
Assert.NotEqual(-1, messageLength);
Assert.NotEqual(-1, endOfBuffer1);
}
else
{
// the latest IPC format has a continuation token at the start
Assert.Equal(-1, messageLength);
Assert.Equal(-1, endOfBuffer1);
}
Assert.Equal(0, endOfBuffer2);
}
}
public async Task TestDataFrameGroupedMapCommandExecutor(
Version sparkVersion,
IpcOptions ipcOptions)
{
public async Task TestArrowGroupedMapCommandExecutor(
Version sparkVersion,
IpcOptions ipcOptions)
{
StringArray ConvertStrings(StringArray strings)
{
return((StringArray)ToArrowArray(
Enumerable.Range(0, strings.Length)
.Select(i => $"udf: {strings.GetString(i)}")
.ToArray()));
}
Int64Array ConvertInt64s(Int64Array int64s)
{
return((Int64Array)ToArrowArray(
Enumerable.Range(0, int64s.Length)
.Select(i => int64s.Values[i] + 100)
.ToArray()));
}
Schema resultSchema = new Schema.Builder()
.Field(b => b.Name("arg1").DataType(StringType.Default))
.Field(b => b.Name("arg2").DataType(Int64Type.Default))
.Build();
var udfWrapper = new Sql.ArrowGroupedMapUdfWrapper(
(batch) => new RecordBatch(
resultSchema,
new IArrowArray[]
{
ConvertStrings((StringArray)batch.Column(0)),
ConvertInt64s((Int64Array)batch.Column(1)),
},
batch.Length));
var command = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.ArrowGroupedMapWorkerFunction(udfWrapper.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var commandPayload = new Worker.CommandPayload()
{
EvalType = UdfUtils.PythonEvalType.SQL_GROUPED_MAP_PANDAS_UDF,
Commands = new[] { command }
};
using var inputStream = new MemoryStream();
using var outputStream = new MemoryStream();
int numRows = 10;
// Write test data to the input stream.
Schema schema = new Schema.Builder()
.Field(b => b.Name("arg1").DataType(StringType.Default))
.Field(b => b.Name("arg2").DataType(Int64Type.Default))
.Build();
var arrowWriter =
new ArrowStreamWriter(inputStream, schema, leaveOpen: false, ipcOptions);
await arrowWriter.WriteRecordBatchAsync(
new RecordBatch(
schema,
new[]
{
ToArrowArray(
Enumerable.Range(0, numRows)
.Select(i => i.ToString())
.ToArray()),
ToArrowArray(
Enumerable.Range(0, numRows)
.Select(i => (long)i)
.ToArray())
},
numRows));
inputStream.Seek(0, SeekOrigin.Begin);
CommandExecutorStat stat = new CommandExecutor(sparkVersion).Execute(
inputStream,
outputStream,
0,
commandPayload);
// Validate that all the data on the stream is read.
Assert.Equal(inputStream.Length, inputStream.Position);
Assert.Equal(numRows, stat.NumEntriesProcessed);
// Validate the output stream.
outputStream.Seek(0, SeekOrigin.Begin);
int arrowLength = SerDe.ReadInt32(outputStream);
Assert.Equal((int)SpecialLengths.START_ARROW_STREAM, arrowLength);
var arrowReader = new ArrowStreamReader(outputStream);
RecordBatch outputBatch = await arrowReader.ReadNextRecordBatchAsync();
Assert.Equal(numRows, outputBatch.Length);
StringArray stringArray;
Int64Array longArray;
if (sparkVersion < new Version(Versions.V3_0_0))
{
Assert.Equal(2, outputBatch.ColumnCount);
stringArray = (StringArray)outputBatch.Column(0);
longArray = (Int64Array)outputBatch.Column(1);
}
else
{
Assert.Equal(1, outputBatch.ColumnCount);
var structArray = (StructArray)outputBatch.Column(0);
Assert.Equal(2, structArray.Fields.Count);
stringArray = (StringArray)structArray.Fields[0];
longArray = (Int64Array)structArray.Fields[1];
}
for (int i = 0; i < numRows; ++i)
{
Assert.Equal($"udf: {i}", stringArray.GetString(i));
}
for (int i = 0; i < numRows; ++i)
{
Assert.Equal(100 + i, longArray.Values[i]);
}
CheckEOS(outputStream, ipcOptions);
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
public void TestDataFrameSqlCommandExecutorWithEmptyInput(
Version sparkVersion,
IpcOptions ipcOptions)
{
var udfWrapper = new Sql.DataFrameUdfWrapper <ArrowStringDataFrameColumn, ArrowStringDataFrameColumn>(
(strings) => strings.Apply(cur => $"udf: {cur}"));
var command = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.DataFrameWorkerFunction(udfWrapper.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var commandPayload = new Worker.CommandPayload()
{
EvalType = UdfUtils.PythonEvalType.SQL_SCALAR_PANDAS_UDF,
Commands = new[] { command }
};
using var inputStream = new MemoryStream();
using var outputStream = new MemoryStream();
// Write test data to the input stream.
Schema schema = new Schema.Builder()
.Field(b => b.Name("arg1").DataType(StringType.Default))
.Build();
var arrowWriter = new ArrowStreamWriter(inputStream, schema, false, ipcOptions);
// The .NET ArrowStreamWriter doesn't currently support writing just a
// schema with no batches - but Java does. We use Reflection to simulate
// the request Spark sends.
MethodInfo writeSchemaMethod = arrowWriter.GetType().GetMethod(
"WriteSchemaAsync",
BindingFlags.NonPublic | BindingFlags.Instance);
writeSchemaMethod.Invoke(
arrowWriter,
new object[] { schema, CancellationToken.None });
SerDe.Write(inputStream, 0);
inputStream.Seek(0, SeekOrigin.Begin);
CommandExecutorStat stat = new CommandExecutor(sparkVersion).Execute(
inputStream,
outputStream,
0,
commandPayload);
// Validate that all the data on the stream is read.
Assert.Equal(inputStream.Length, inputStream.Position);
Assert.Equal(0, stat.NumEntriesProcessed);
// Validate the output stream.
outputStream.Seek(0, SeekOrigin.Begin);
int arrowLength = SerDe.ReadInt32(outputStream);
Assert.Equal((int)SpecialLengths.START_ARROW_STREAM, arrowLength);
var arrowReader = new ArrowStreamReader(outputStream);
RecordBatch outputBatch = arrowReader.ReadNextRecordBatch();
Assert.Equal(1, outputBatch.Schema.Fields.Count);
Assert.IsType <StringType>(outputBatch.Schema.GetFieldByIndex(0).DataType);
Assert.Equal(0, outputBatch.Length);
Assert.Single(outputBatch.Arrays);
var array = (StringArray)outputBatch.Arrays.ElementAt(0);
Assert.Equal(0, array.Length);
CheckEOS(outputStream, ipcOptions);
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
public async Task TestDataFrameSqlCommandExecutorWithMultiCommands(
Version sparkVersion,
IpcOptions ipcOptions)
{
var udfWrapper1 = new Sql.DataFrameUdfWrapper <ArrowStringDataFrameColumn, ArrowStringDataFrameColumn>(
(strings) => strings.Apply(cur => $"udf: {cur}"));
var udfWrapper2 = new Sql.DataFrameUdfWrapper <Int32DataFrameColumn, Int32DataFrameColumn, Int32DataFrameColumn>(
(arg1, arg2) => arg1 * arg2);
var command1 = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.DataFrameWorkerFunction(udfWrapper1.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var command2 = new SqlCommand()
{
ArgOffsets = new[] { 1, 2 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.DataFrameWorkerFunction(udfWrapper2.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var commandPayload = new Worker.CommandPayload()
{
EvalType = UdfUtils.PythonEvalType.SQL_SCALAR_PANDAS_UDF,
Commands = new[] { command1, command2 }
};
using var inputStream = new MemoryStream();
using var outputStream = new MemoryStream();
int numRows = 10;
// Write test data to the input stream.
Schema schema = new Schema.Builder()
.Field(b => b.Name("arg1").DataType(StringType.Default))
.Field(b => b.Name("arg2").DataType(Int32Type.Default))
.Field(b => b.Name("arg3").DataType(Int32Type.Default))
.Build();
var arrowWriter =
new ArrowStreamWriter(inputStream, schema, leaveOpen: false, ipcOptions);
await arrowWriter.WriteRecordBatchAsync(
new RecordBatch(
schema,
new[]
{
ToArrowArray(
Enumerable.Range(0, numRows)
.Select(i => i.ToString())
.ToArray()),
ToArrowArray(Enumerable.Range(0, numRows).ToArray()),
ToArrowArray(Enumerable.Range(0, numRows).ToArray()),
},
numRows));
inputStream.Seek(0, SeekOrigin.Begin);
CommandExecutorStat stat = new CommandExecutor(sparkVersion).Execute(
inputStream,
outputStream,
0,
commandPayload);
// Validate all the data on the stream is read.
Assert.Equal(inputStream.Length, inputStream.Position);
Assert.Equal(numRows, stat.NumEntriesProcessed);
// Validate the output stream.
outputStream.Seek(0, SeekOrigin.Begin);
var arrowLength = SerDe.ReadInt32(outputStream);
Assert.Equal((int)SpecialLengths.START_ARROW_STREAM, arrowLength);
var arrowReader = new ArrowStreamReader(outputStream);
RecordBatch outputBatch = await arrowReader.ReadNextRecordBatchAsync();
Assert.Equal(numRows, outputBatch.Length);
Assert.Equal(2, outputBatch.Arrays.Count());
var array1 = (StringArray)outputBatch.Arrays.ElementAt(0);
var array2 = (Int32Array)outputBatch.Arrays.ElementAt(1);
for (int i = 0; i < numRows; ++i)
{
Assert.Equal($"udf: {i}", array1.GetString(i));
Assert.Equal(i * i, array2.Values[i]);
}
CheckEOS(outputStream, ipcOptions);
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
public void TestPicklingSqlCommandExecutorWithSingleCommand(
Version sparkVersion,
IpcOptions ipcOptions)
{
_ = ipcOptions;
var udfWrapper = new Sql.PicklingUdfWrapper <string, string>(
(str) => "udf: " + ((str is null) ? "NULL" : str));
var command = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.PicklingWorkerFunction(udfWrapper.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var commandPayload = new Worker.CommandPayload()
{
EvalType = UdfUtils.PythonEvalType.SQL_BATCHED_UDF,
Commands = new[] { command }
};
using var inputStream = new MemoryStream();
using var outputStream = new MemoryStream();
int numRows = 10;
// Write test data to the input stream.
var pickler = new Pickler();
for (int i = 0; i < numRows; ++i)
{
byte[] pickled = pickler.dumps(
new[] { new object[] { (i % 2 == 0) ? null : i.ToString() } });
SerDe.Write(inputStream, pickled.Length);
SerDe.Write(inputStream, pickled);
}
SerDe.Write(inputStream, (int)SpecialLengths.END_OF_DATA_SECTION);
inputStream.Seek(0, SeekOrigin.Begin);
CommandExecutorStat stat = new CommandExecutor(sparkVersion).Execute(
inputStream,
outputStream,
0,
commandPayload);
// Validate that all the data on the stream is read.
Assert.Equal(inputStream.Length, inputStream.Position);
Assert.Equal(10, stat.NumEntriesProcessed);
// Validate the output stream.
outputStream.Seek(0, SeekOrigin.Begin);
var unpickler = new Unpickler();
// One row was written as a batch above, thus need to read 'numRows' batches.
List <object> rows = new List <object>();
for (int i = 0; i < numRows; ++i)
{
int length = SerDe.ReadInt32(outputStream);
byte[] pickledBytes = SerDe.ReadBytes(outputStream, length);
rows.Add((unpickler.loads(pickledBytes) as ArrayList)[0] as object);
}
Assert.Equal(numRows, rows.Count);
// Validate the single command.
for (int i = 0; i < numRows; ++i)
{
Assert.Equal(
"udf: " + ((i % 2 == 0) ? "NULL" : i.ToString()),
(string)rows[i]);
}
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
public void TestPicklingSqlCommandExecutorWithMultiCommands(
Version sparkVersion,
IpcOptions ipcOptions)
{
_ = ipcOptions;
var udfWrapper1 = new Sql.PicklingUdfWrapper <string, string>((str) => $"udf: {str}");
var udfWrapper2 = new Sql.PicklingUdfWrapper <int, int, int>(
(arg1, arg2) => arg1 * arg2);
var command1 = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.PicklingWorkerFunction(udfWrapper1.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var command2 = new SqlCommand()
{
ArgOffsets = new[] { 1, 2 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.PicklingWorkerFunction(udfWrapper2.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var commandPayload = new Worker.CommandPayload()
{
EvalType = UdfUtils.PythonEvalType.SQL_BATCHED_UDF,
Commands = new[] { command1, command2 }
};
using var inputStream = new MemoryStream();
using var outputStream = new MemoryStream();
int numRows = 10;
// Write test data to the input stream.
var pickler = new Pickler();
for (int i = 0; i < numRows; ++i)
{
byte[] pickled = pickler.dumps(
new[] { new object[] { i.ToString(), i, i } });
SerDe.Write(inputStream, pickled.Length);
SerDe.Write(inputStream, pickled);
}
SerDe.Write(inputStream, (int)SpecialLengths.END_OF_DATA_SECTION);
inputStream.Seek(0, SeekOrigin.Begin);
CommandExecutorStat stat = new CommandExecutor(sparkVersion).Execute(
inputStream,
outputStream,
0,
commandPayload);
// Validate all the data on the stream is read.
Assert.Equal(inputStream.Length, inputStream.Position);
Assert.Equal(10, stat.NumEntriesProcessed);
// Validate the output stream.
outputStream.Seek(0, SeekOrigin.Begin);
var unpickler = new Unpickler();
// One row was written as a batch above, thus need to read 'numRows' batches.
List <object[]> rows = new List <object[]>();
for (int i = 0; i < numRows; ++i)
{
int length = SerDe.ReadInt32(outputStream);
byte[] pickledBytes = SerDe.ReadBytes(outputStream, length);
rows.Add((unpickler.loads(pickledBytes) as ArrayList)[0] as object[]);
}
Assert.Equal(numRows, rows.Count);
for (int i = 0; i < numRows; ++i)
{
// There were two UDFs each of which produces one column.
object[] columns = rows[i];
Assert.Equal($"udf: {i}", (string)columns[0]);
Assert.Equal(i * i, (int)columns[1]);
}
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
private static async Task TestRoundTripRecordBatchAsync(RecordBatch originalBatch, IpcOptions options = null)
{
await TestRoundTripRecordBatchesAsync(new List <RecordBatch> {
originalBatch
}, options);
}
private static void TestRoundTripRecordBatch(RecordBatch originalBatch, IpcOptions options = null)
{
TestRoundTripRecordBatches(new List <RecordBatch> {
originalBatch
}, options);
}
private static async Task TestRoundTripRecordBatchesAsync(List <RecordBatch> originalBatches, IpcOptions options = null)
{
using (MemoryStream stream = new MemoryStream())
{
using (var writer = new ArrowStreamWriter(stream, originalBatches[0].Schema, leaveOpen: true, options))
{
foreach (RecordBatch originalBatch in originalBatches)
{
await writer.WriteRecordBatchAsync(originalBatch);
}
await writer.WriteEndAsync();
}
stream.Position = 0;
using (var reader = new ArrowStreamReader(stream))
{
foreach (RecordBatch originalBatch in originalBatches)
{
RecordBatch newBatch = reader.ReadNextRecordBatch();
ArrowReaderVerifier.CompareBatches(originalBatch, newBatch);
}
}
}
}
