public static async Task VerifyReaderAsync(ArrowStreamReader reader, RecordBatch originalBatch)
{
RecordBatch readBatch = await reader.ReadNextRecordBatchAsync();
CompareBatches(originalBatch, readBatch);
// There should only be one batch - calling ReadNextRecordBatchAsync again should return null.
Assert.Null(await reader.ReadNextRecordBatchAsync());
Assert.Null(await reader.ReadNextRecordBatchAsync());
}
public async Task <double> ArrowReaderWithMemory()
{
double sum = 0;
var reader = new ArrowStreamReader(_memoryStream.GetBuffer());
RecordBatch recordBatch;
while ((recordBatch = await reader.ReadNextRecordBatchAsync()) != null)
{
sum += SumAllNumbers(recordBatch);
}
return(sum);
}
public async Task <double> ArrowReaderWithMemoryStream_ManagedMemory()
{
double sum = 0;
var reader = new ArrowStreamReader(_memoryStream, s_allocator);
RecordBatch recordBatch;
while ((recordBatch = await reader.ReadNextRecordBatchAsync()) != null)
{
using (recordBatch)
{
sum += SumAllNumbers(recordBatch);
}
}
return(sum);
}
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 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 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);
}