private CommandExecutorStat ExecuteArrowGroupedMapCommand(
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 = (ArrowGroupedMapWorkerFunction)commands[0].WorkerFunction;
SerDe.Write(outputStream, (int)SpecialLengths.START_ARROW_STREAM);
IpcOptions ipcOptions = ArrowIpcOptions();
ArrowStreamWriter writer = null;
foreach (RecordBatch input in GetInputIterator(inputStream))
{
RecordBatch result = worker.Func(input);
int numEntries = result.Length;
stat.NumEntriesProcessed += numEntries;
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 ExecuteArrowGroupedMapCommand(
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 = (ArrowGroupedMapWorkerFunction)commands[0].WorkerFunction;
SerDe.Write(outputStream, (int)SpecialLengths.START_ARROW_STREAM);
IpcOptions ipcOptions = ArrowIpcOptions();
ArrowStreamWriter writer = null;
foreach (RecordBatch input in GetInputIterator(inputStream))
{
RecordBatch batch = worker.Func(input);
RecordBatch final = WrapColumnsInStructIfApplicable(batch);
int numEntries = final.Length;
stat.NumEntriesProcessed += numEntries;
if (writer == null)
{
writer =
new ArrowStreamWriter(outputStream, final.Schema, leaveOpen: true, ipcOptions);
}
writer.WriteRecordBatch(final);
}
WriteEnd(outputStream, ipcOptions);
writer?.Dispose();
return(stat);
}
private static async Task TestReaderFromMemory(
Func <ArrowStreamReader, RecordBatch, Task> verificationFunc,
bool writeEnd)
{
RecordBatch originalBatch = TestData.CreateSampleRecordBatch(length: 100);
byte[] buffer;
using (MemoryStream stream = new MemoryStream())
{
ArrowStreamWriter writer = new ArrowStreamWriter(stream, originalBatch.Schema);
await writer.WriteRecordBatchAsync(originalBatch);
if (writeEnd)
{
await writer.WriteEndAsync();
}
buffer = stream.GetBuffer();
}
ArrowStreamReader reader = new ArrowStreamReader(buffer);
await verificationFunc(reader, originalBatch);
}
public async Task TestArrowGroupedMapCommandExecutor()
{
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);
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().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);
Assert.Equal(2, outputBatch.ColumnCount);
var stringArray = (StringArray)outputBatch.Column(0);
for (int i = 0; i < numRows; ++i)
{
Assert.Equal($"udf: {i}", stringArray.GetString(i));
}
var longArray = (Int64Array)outputBatch.Column(1);
for (int i = 0; i < numRows; ++i)
{
Assert.Equal(100 + i, longArray.Values[i]);
}
int end = SerDe.ReadInt32(outputStream);
Assert.Equal(0, end);
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
}
public void TestArrowSqlCommandExecutorWithEmptyInput()
{
var udfWrapper = new Sql.ArrowUdfWrapper <StringArray, StringArray>(
(strings) => (StringArray)ToArrowArray(
Enumerable.Range(0, strings.Length)
.Select(i => $"udf: {strings.GetString(i)}")
.ToArray()));
var command = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.ArrowWorkerFunction(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);
// 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().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);
int end = SerDe.ReadInt32(outputStream);
Assert.Equal(0, end);
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
}
public async Task TestArrowSqlCommandExecutorWithMultiCommands()
{
var udfWrapper1 = new Sql.ArrowUdfWrapper <StringArray, StringArray>(
(strings) => (StringArray)ToArrowArray(
Enumerable.Range(0, strings.Length)
.Select(i => $"udf: {strings.GetString(i)}")
.ToArray()));
var udfWrapper2 = new Sql.ArrowUdfWrapper <Int32Array, Int32Array, Int32Array>(
(arg1, arg2) => (Int32Array)ToArrowArray(
Enumerable.Range(0, arg1.Length)
.Select(i => arg1.Values[i] * arg2.Values[i])
.ToArray()));
var command1 = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.ArrowWorkerFunction(udfWrapper1.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var command2 = new SqlCommand()
{
ArgOffsets = new[] { 1, 2 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.ArrowWorkerFunction(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);
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().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]);
}
int end = SerDe.ReadInt32(outputStream);
Assert.Equal(0, end);
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
}
public async Task WriteBatch()
{
ArrowStreamWriter writer = new ArrowStreamWriter(_memoryStream, _batch.Schema);
await writer.WriteRecordBatchAsync(_batch);
}
public async Task TestDataFrameSqlCommandExecutorWithSingleCommand(
Version sparkVersion,
IpcOptions ipcOptions)
{
var udfWrapper = new Sql.DataFrameUdfWrapper <ArrowStringDataFrameColumn, ArrowStringDataFrameColumn>(
(strings) => strings.Apply(cur => $"udf: {cur}"));
var command = new SqlCommand()
{
ArgOffsets = new[] { 0 },
NumChainedFunctions = 1,
WorkerFunction = new Sql.DataFrameWorkerFunction(udfWrapper.Execute),
SerializerMode = CommandSerDe.SerializedMode.Row,
DeserializerMode = CommandSerDe.SerializedMode.Row
};
var commandPayload = new Worker.CommandPayload()
{
EvalType = UdfUtils.PythonEvalType.SQL_SCALAR_PANDAS_UDF,
Commands = new[] { command }
};
using var inputStream = new MemoryStream();
using var outputStream = new MemoryStream();
int numRows = 10;
// Write test data to the input stream.
Schema schema = new Schema.Builder()
.Field(b => b.Name("arg1").DataType(StringType.Default))
.Build();
var arrowWriter =
new ArrowStreamWriter(inputStream, schema, leaveOpen: false, ipcOptions);
await arrowWriter.WriteRecordBatchAsync(
new RecordBatch(
schema,
new[]
{
ToArrowArray(
Enumerable.Range(0, numRows)
.Select(i => i.ToString())
.ToArray())
},
numRows));
inputStream.Seek(0, SeekOrigin.Begin);
CommandExecutorStat stat = new CommandExecutor(sparkVersion).Execute(
inputStream,
outputStream,
0,
commandPayload);
// Validate that all the data on the stream is read.
Assert.Equal(inputStream.Length, inputStream.Position);
Assert.Equal(numRows, stat.NumEntriesProcessed);
// Validate the output stream.
outputStream.Seek(0, SeekOrigin.Begin);
int arrowLength = SerDe.ReadInt32(outputStream);
Assert.Equal((int)SpecialLengths.START_ARROW_STREAM, arrowLength);
var arrowReader = new ArrowStreamReader(outputStream);
RecordBatch outputBatch = await arrowReader.ReadNextRecordBatchAsync();
Assert.Equal(numRows, outputBatch.Length);
Assert.Single(outputBatch.Arrays);
var array = (StringArray)outputBatch.Arrays.ElementAt(0);
// Validate the single command.
for (int i = 0; i < numRows; ++i)
{
Assert.Equal($"udf: {i}", array.GetString(i));
}
CheckEOS(outputStream, ipcOptions);
// Validate all the data on the stream is read.
Assert.Equal(outputStream.Length, outputStream.Position);
}
public async Task WriteBatchWithCorrectPaddingAsync()
{
byte value1 = 0x04;
byte value2 = 0x14;
var batch = new RecordBatch(
new Schema.Builder()
.Field(f => f.Name("age").DataType(Int32Type.Default))
.Field(f => f.Name("characterCount").DataType(Int32Type.Default))
.Build(),
new IArrowArray[]
{
new Int32Array(
new ArrowBuffer(new byte[] { value1, value1, 0x00, 0x00 }),
ArrowBuffer.Empty,
length: 1,
nullCount: 0,
offset: 0),
new Int32Array(
new ArrowBuffer(new byte[] { value2, value2, 0x00, 0x00 }),
ArrowBuffer.Empty,
length: 1,
nullCount: 0,
offset: 0)
},
length: 1);
await TestRoundTripRecordBatchAsync(batch);
using (MemoryStream stream = new MemoryStream())
{
using (var writer = new ArrowStreamWriter(stream, batch.Schema, leaveOpen: true))
{
await writer.WriteRecordBatchAsync(batch);
await writer.WriteEndAsync();
}
byte[] writtenBytes = stream.ToArray();
// ensure that the data buffers at the end are 8-byte aligned
Assert.Equal(value1, writtenBytes[writtenBytes.Length - 24]);
Assert.Equal(value1, writtenBytes[writtenBytes.Length - 23]);
for (int i = 22; i > 16; i--)
{
Assert.Equal(0, writtenBytes[writtenBytes.Length - i]);
}
Assert.Equal(value2, writtenBytes[writtenBytes.Length - 16]);
Assert.Equal(value2, writtenBytes[writtenBytes.Length - 15]);
for (int i = 14; i > 8; i--)
{
Assert.Equal(0, writtenBytes[writtenBytes.Length - i]);
}
// verify the EOS is written correctly
for (int i = 8; i > 4; i--)
{
Assert.Equal(0xFF, writtenBytes[writtenBytes.Length - i]);
}
for (int i = 4; i > 0; i--)
{
Assert.Equal(0x00, writtenBytes[writtenBytes.Length - i]);
}
}
}