public void Deploy(ref TaskBase task, ref OperatorsToplogy operatorsTopology, ref OperatorTransforms operandTransforms)
{
GenerateProduceTask(ref operatorsTopology);
var isRightOperandInput = task.IsRightOperandInput;
if (isRightOperandInput)
{
task.InputIds.SetInputId2(_shardedDatasetId);
task.NextInputIds.SetInputId2(_shardedDatasetId);
task.OperationTypes.SetSecondaryKeyType(typeof(TKey));
task.OperationTypes.SetSecondaryPayloadType(typeof(TPayload));
task.OperationTypes.SetSecondaryDatasetType(typeof(TDataset));
}
else
{
task.InputIds.SetInputId1(_shardedDatasetId);
task.NextInputIds.SetInputId1(_shardedDatasetId);
task.OperationTypes.SetInputKeyType(typeof(TKey));
task.OperationTypes.SetInputPayloadType(typeof(TPayload));
task.OperationTypes.SetInputDatasetType(typeof(TDataset));
}
task.IsRightOperandInput = false;
}
public static bool DeployClientTerminal(CRAClientLibrary client, ClientTerminalTask task,
ref DetachedVertex clientTerminal, OperatorsToplogy topology)
{
try
{
foreach (string fromInputId in task.EndpointsDescriptor.FromInputs.Keys)
{
string[] inputEndpoints = OperatorUtils.PrepareInputEndpointsIdsForOperator(fromInputId, task.EndpointsDescriptor);
string[] outputEndpoints = OperatorUtils.PrepareOutputEndpointsIdsForOperator(
task.OutputId, topology.OperatorsEndpointsDescriptors[fromInputId]);
int shardsCount = client.CountVertexShards(task.DeployDescriptor.InstancesMap());
for (int i = 0; i < shardsCount; i++)
{
for (int j = 0; j < inputEndpoints.Length; j++)
{
clientTerminal.FromRemoteOutputEndpointStream(inputEndpoints[j] + i, fromInputId + "$" + i, outputEndpoints[j]);
}
}
}
return(true);
}
catch (Exception e)
{
Console.WriteLine("Error in deploying a CRA client terminal vertex. Please, double check your task configurations: " + e.ToString());
return(false);
}
}
private void DeployUnaryTransform(ref TaskBase task, ref OperatorsToplogy topology, ref OperatorTransforms parentTransforms)
{
var isRightOperandInput = task.IsRightOperandInput;
OperatorInputs temporaryInputs = new OperatorInputs();
// Prepare the transformation of the operation unary (left) operand
task.IsRightOperandInput = false;
OperatorTransforms unaryTransforms = new OperatorTransforms();
(_input1 as IDeployable).Deploy(ref task, ref topology, ref unaryTransforms);
temporaryInputs.InputId1 = task.NextInputIds.InputId1;
// Update the inputs for the next operation
if (isRightOperandInput)
{
task.NextInputIds.SetInputId2(temporaryInputs.InputId1);
}
else
{
task.NextInputIds.SetInputId1(temporaryInputs.InputId1);
}
// Add the transformations from this operand to the parent operand
unaryTransforms.AddTransform(SerializationHelper.Serialize(_unaryTransformer),
OperatorType.UnaryTransform.ToString(),
TransformUtils.FillUnaryTransformTypes(typeof(TKeyI1), typeof(TPayloadI1), typeof(TDataSetI1),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO)).ToString(), temporaryInputs);
parentTransforms = unaryTransforms;
}
public static OperatorsToplogy GetInstance()
{
if (_instance == null)
{
_instance = new OperatorsToplogy();
}
return(_instance);
}
public static async Task <bool> DeployOperators(CRAClientLibrary client, OperatorsToplogy topology)
{
topology.PrepareFinalOperatorsTasks();
string[] tasksIds = topology.OperatorsIds.ToArray();
TaskBase[] tasks = topology.OperatorsTasks.ToArray();
for (int i = 0; i < tasks.Length; i++)
{
if (tasks[i].OperationType == OperatorType.Move)
{
ShuffleTask shuffleTask = (ShuffleTask)tasks[i];
shuffleTask.SecondaryEndpointsDescriptor = new OperatorEndpointsDescriptor();
shuffleTask.SecondaryEndpointsDescriptor.FromInputs = topology.OperatorsEndpointsDescriptors[shuffleTask.ReducerVertexName].FromInputs;
topology.OperatorsEndpointsDescriptors[shuffleTask.ReducerVertexName].FromInputs = new ConcurrentDictionary <string, int>();
int shardsCount = client.CountVertexShards(shuffleTask.DeployDescriptor.InstancesMap());
topology.UpdateShuffleInputs(shuffleTask.MapperVertexName, shuffleTask.ReducerVertexName, shardsCount);
topology.UpdateShuffleOperatorTask(shuffleTask.ReducerVertexName, shuffleTask);
}
}
var tasksDictionary = PrepareTasksDictionary(tasks);
var connectionsMap = PrepareVerticesConnectionsMap(client, tasks, tasksIds, tasksDictionary, topology);
var tasksDeploymentStatus = new Dictionary <string, bool>();
for (int i = 0; i < tasks.Length; i++)
{
tasks[i].EndpointsDescriptor = topology.OperatorsEndpointsDescriptors[tasksIds[i]];
tasks[i].VerticesConnectionsMap = connectionsMap;
tasksDeploymentStatus.Add(tasksIds[i], false);
}
bool isSuccessful = true;
for (int i = 0; i < tasks.Length; i++)
{
if (tasks[i].OperationType == OperatorType.Produce && tasks[i].EndpointsDescriptor.FromInputs.Count == 0 &&
tasks[i].EndpointsDescriptor.SecondaryFromInputs.Count == 0)
{
isSuccessful = await DeployProduceTask(client, (ProduceTask)tasks[i], topology);
if (isSuccessful)
{
tasksDeploymentStatus[tasksIds[i]] = true;
}
}
}
for (int i = 0; i < tasks.Length; i++)
{
isSuccessful = isSuccessful & await DeployTask(i, tasks, tasksIds, tasksDeploymentStatus, client, topology);
}
return(isSuccessful);
}
public static bool DeployOperators(CRAClientLibrary client, OperatorsToplogy topology)
{
topology.PrepareFinalOperatorsTasks();
string[] tasksIds = topology.OperatorsIds.ToArray();
TaskBase[] tasks = topology.OperatorsTasks.ToArray();
for (int i = 0; i < tasks.Length; i++)
{
if (tasks[i].OperationType == OperatorType.Move)
{
ShuffleTask shuffleTask = (ShuffleTask)tasks[i];
shuffleTask.SecondaryEndpointsDescriptor = new OperatorEndpointsDescriptor();
shuffleTask.SecondaryEndpointsDescriptor.FromInputs = topology.OperatorsEndpointsDescriptors[shuffleTask.ReducerVertexName].FromInputs;
topology.OperatorsEndpointsDescriptors[shuffleTask.ReducerVertexName].FromInputs = new ConcurrentDictionary <string, int>();
int shardsCount = client.CountVertexShards(shuffleTask.DeployDescriptor.InstancesMap());
topology.UpdateShuffleInputs(shuffleTask.MapperVertexName, shuffleTask.ReducerVertexName, shardsCount);
topology.UpdateShuffleOperatorTask(shuffleTask.ReducerVertexName, shuffleTask);
}
}
var tasksDictionary = PrepareTasksDictionary(tasks);
var connectionsMap = PrepareVerticesConnectionsMap(client, tasks, tasksIds, tasksDictionary, topology);
bool isSuccessful = true;
for (int i = 0; i < tasks.Length; i++)
{
tasks[i].EndpointsDescriptor = topology.OperatorsEndpointsDescriptors[tasksIds[i]];
tasks[i].VerticesConnectionsMap = connectionsMap;
if (tasks[i].OperationType == OperatorType.Produce)
{
isSuccessful = DeployProduceTask(client, (ProduceTask)tasks[i]);
}
else if (tasks[i].OperationType == OperatorType.Subscribe)
{
isSuccessful = DeploySubscribeTask(client, (SubscribeTask)tasks[i], topology);
}
else if (tasks[i].OperationType == OperatorType.Move)
{
isSuccessful = DeployShuffleReduceTask(client, (ShuffleTask)tasks[i], topology);
}
if (!isSuccessful)
{
break;
}
}
return(isSuccessful);
}
public static async Task <bool> DeployClientTerminal(
CRAClientLibrary client,
string workerName,
ClientTerminalTask task,
OperatorsToplogy topology)
{
try
{
bool result = true;
client.DisableArtifactUploading();
if (!_isSubscribeClientOperatorDefined)
{
await client.DefineVertexAsync(typeof(ShardedSubscribeClientOperator).Name.ToLower(), () => new ShardedSubscribeClientOperator());
_isSubscribeClientOperatorDefined = true;
}
var status = await client.InstantiateVertexAsync(new string[] { workerName }, task.OutputId, typeof(ShardedSubscribeClientOperator).Name.ToLower(), task, 1);
if (status == CRAErrorCode.Success)
{
foreach (string fromInputId in task.EndpointsDescriptor.FromInputs.Keys)
{
string outputEndpoint = OperatorUtils.PrepareOutputEndpointIdForOperator(
task.OutputId, topology.OperatorsEndpointsDescriptors[fromInputId]);
string inputEndpoint = OperatorUtils.PrepareInputEndpointIdForOperator(fromInputId, task.EndpointsDescriptor, false);
await client.ConnectAsync(fromInputId, outputEndpoint, task.OutputId, inputEndpoint);
}
result = true;
}
else
{
result = false;
}
client.EnableArtifactUploading();
return(result);
}
catch (Exception e)
{
Console.WriteLine("Error in deploying a CRA client terminal vertex. Please, double check your task configurations: " + e.ToString());
return(false);
}
}
public void Deploy(ref TaskBase task, ref OperatorsToplogy topology, ref OperatorTransforms parentTransforms)
{
switch (OperationType)
{
case OperatorType.Move:
DeployMove(ref task, ref topology);
break;
case OperatorType.BinaryTransform:
DeployBinaryTransform(ref task, ref topology, ref parentTransforms);
break;
default:
DeployUnaryTransform(ref task, ref topology, ref parentTransforms);
break;
}
}
public override IShardedDataset <TKey, TPayload, TDataset> Deploy()
{
if (!_isDeployed)
{
OperatorsToplogy operatorsTopology = OperatorsToplogy.GetInstance();
GenerateProduceTask(ref operatorsTopology);
_craClient = new CRAClientLibrary();
_isDeployed = DeploymentUtils.DeployOperators(_craClient, operatorsTopology);
if (!_isDeployed)
{
return(null);
}
}
return(this);
}
private static async Task <bool> DeployShuffleReduceTask(
CRAClientLibrary client,
ShuffleTask task,
OperatorsToplogy topology)
{
try
{
if (!_isShuffleOperatorDefined)
{
await client.DefineVertexAsync(typeof(ShardedShuffleOperator).Name.ToLower(), () => new ShardedShuffleOperator());
_isShuffleOperatorDefined = true;
}
var status = await client.InstantiateVertexAsync(CreateInstancesNames(task.DeployDescriptor.InstancesMap()), task.ReducerVertexName, typeof(ShardedShuffleOperator).Name.ToLower(), task, 1);
if (status == CRAErrorCode.Success)
{
foreach (string fromInputId in task.EndpointsDescriptor.FromInputs.Keys)
{
var fromToConnection = task.VerticesConnectionsMap[fromInputId + task.ReducerVertexName][0];
await client.ConnectAsync(fromToConnection.FromVertex, fromToConnection.FromEndpoint, fromToConnection.ToVertex, fromToConnection.ToEndpoint);
}
foreach (string fromSecondaryInputId in task.EndpointsDescriptor.SecondaryFromInputs.Keys)
{
var fromToConnection = task.VerticesConnectionsMap[fromSecondaryInputId + task.ReducerVertexName][0];
await client.ConnectAsync(fromToConnection.FromVertex, fromToConnection.FromEndpoint, fromToConnection.ToVertex, fromToConnection.ToEndpoint);
}
return(true);
}
else
{
return(false);
}
}
catch (Exception e)
{
Console.WriteLine("Error in deploying a sharded CRA shuffle mapper task. Please, double check your task configurations: " + e.ToString());
return(false);
}
}
private void GenerateProduceTask(ref OperatorsToplogy operatorsTopology)
{
_shardedDatasetId = typeof(ShardedProducerOperator).Name.ToLower() + Guid.NewGuid().ToString();
TaskBase produceTask = new ProduceTask(SerializationHelper.Serialize(_producer));
produceTask.OperationTypes = TransformUtils.FillBinaryTransformTypes(
typeof(TKey), typeof(TPayload), typeof(TDataset),
typeof(TKey), typeof(TPayload), typeof(TDataset),
typeof(TKey), typeof(TPayload), typeof(TDataset));
produceTask.IsRightOperandInput = false;
produceTask.InputIds.SetInputId1(_shardedDatasetId);
produceTask.InputIds.SetInputId2(_shardedDatasetId);
produceTask.OutputId = _shardedDatasetId;
produceTask.NextInputIds.SetInputId1(_shardedDatasetId);
produceTask.NextInputIds.SetInputId2(_shardedDatasetId);
produceTask.PrepareTaskTransformations(new OperatorTransforms());
operatorsTopology.AddOperatorBase(produceTask.OutputId, produceTask);
}
private void DeployBinaryTransform(ref TaskBase task, ref OperatorsToplogy topology, ref OperatorTransforms parentTransforms)
{
var isRightOperandInput = task.IsRightOperandInput;
OperatorInputs temporaryInputs = new OperatorInputs();
// Prepare the transformations of the operation left operand
task.IsRightOperandInput = false;
OperatorTransforms leftTransforms = new OperatorTransforms();
(_input1 as IDeployable).Deploy(ref task, ref topology, ref leftTransforms);
temporaryInputs.InputId1 = task.NextInputIds.InputId1;
// Prepare the transformations of the operation right operand
task.IsRightOperandInput = true;
OperatorTransforms rightTransforms = new OperatorTransforms();
(_input2 as IDeployable).Deploy(ref task, ref topology, ref rightTransforms);
temporaryInputs.InputId2 = task.NextInputIds.InputId2;
// Update the inputs for the next operation
if (isRightOperandInput)
{
task.NextInputIds.SetInputId2(temporaryInputs.InputId1);
}
else
{
task.NextInputIds.SetInputId1(temporaryInputs.InputId1);
}
// Merge the tranformations from the two operands, and assign them to the parent operation
parentTransforms = TransformUtils.MergeTwoSetsOfTransforms(leftTransforms, rightTransforms);
parentTransforms.AddTransform(SerializationHelper.Serialize(_binaryTransformer),
OperatorType.BinaryTransform.ToString(),
TransformUtils.FillBinaryTransformTypes(typeof(TKeyI1), typeof(TPayloadI1), typeof(TDataSetI1),
typeof(TKeyI2), typeof(TPayloadI2), typeof(TDataSetI2),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO)).ToString(), temporaryInputs);
}
public override async Task <IShardedDataset <TKeyO, TPayloadO, TDataSetO> > Deploy()
{
if (!_isDeployed)
{
OperatorsToplogy toplogy = OperatorsToplogy.GetInstance();
TaskBase subscribeTask = new SubscribeTask();
subscribeTask.OperationTypes = TransformUtils.FillBinaryTransformTypes(
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO));
subscribeTask.IsRightOperandInput = false;
OperatorTransforms subscribeInputTransforms = new OperatorTransforms();
Deploy(ref subscribeTask, ref toplogy, ref subscribeInputTransforms);
subscribeTask.InputIds.SetInputId1(subscribeTask.NextInputIds.InputId1);
subscribeTask.InputIds.SetInputId2(subscribeTask.NextInputIds.InputId2);
subscribeTask.OutputId = typeof(ShardedSubscribeOperator).Name.ToLower() + Guid.NewGuid().ToString();
subscribeTask.PrepareTaskTransformations(subscribeInputTransforms);
toplogy.AddOperatorBase(subscribeTask.OutputId, subscribeTask);
toplogy.AddOperatorInput(subscribeTask.OutputId, subscribeTask.InputIds.InputId1);
toplogy.AddOperatorSecondaryInput(subscribeTask.OutputId, subscribeTask.InputIds.InputId2);
toplogy.AddOperatorOutput(subscribeTask.InputIds.InputId1, subscribeTask.OutputId);
toplogy.AddOperatorOutput(subscribeTask.InputIds.InputId2, subscribeTask.OutputId);
if (subscribeTask.Transforms != null)
{
foreach (OperatorInputs inputs in subscribeTask.TransformsInputs)
{
toplogy.AddOperatorSecondaryInput(subscribeTask.OutputId, inputs.InputId2);
toplogy.AddOperatorOutput(inputs.InputId2, subscribeTask.OutputId);
}
foreach (OperatorInputs inputs in subscribeTask.TransformsInputs)
{
if (!toplogy.ContainsSecondaryOperatorInput(subscribeTask.OutputId, inputs.InputId1))
{
toplogy.AddOperatorInput(subscribeTask.OutputId, inputs.InputId1);
toplogy.AddOperatorOutput(inputs.InputId1, subscribeTask.OutputId);
}
}
}
_clientTerminalTask = new ClientTerminalTask();
_clientTerminalTask.InputIds.SetInputId1(subscribeTask.OutputId);
_clientTerminalTask.OutputId = typeof(ShardedSubscribeClientOperator).Name.ToLower() + Guid.NewGuid().ToString();
_clientTerminalTask.OperationTypes = TransformUtils.FillBinaryTransformTypes(
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO));
_craClient = new CRAClientLibrary(_dataProvider);
toplogy.AddOperatorBase(_clientTerminalTask.OutputId, _clientTerminalTask);
toplogy.AddOperatorInput(_clientTerminalTask.OutputId, _clientTerminalTask.InputIds.InputId1);
toplogy.AddOperatorInput(_clientTerminalTask.OutputId, _clientTerminalTask.InputIds.InputId2);
toplogy.AddOperatorOutput(_clientTerminalTask.InputIds.InputId1, _clientTerminalTask.OutputId);
toplogy.AddOperatorOutput(_clientTerminalTask.InputIds.InputId2, _clientTerminalTask.OutputId);
_isDeployed = await DeploymentUtils.DeployOperators(_craClient, toplogy);
if (_isDeployed)
{
string craWorkerName = typeof(ShardedSubscribeClientOperator).Name.ToLower() + "worker" + Guid.NewGuid().ToString();
_craWorker = new CRAWorker(craWorkerName, "127.0.0.1", NetworkUtils.GetAvailablePort(), _craClient.DataProvider, null, 1000);
_craWorker.DisableDynamicLoading();
_craWorker.SideloadVertex(new ShardedSubscribeClientOperator(), typeof(ShardedSubscribeClientOperator).Name.ToLower());
new Thread(() => _craWorker.Start()).Start();
Thread.Sleep(1000);
_isDeployed = await DeploymentUtils.DeployClientTerminal(_craClient, craWorkerName, _clientTerminalTask, toplogy);
}
else
{
return(null);
}
}
return(this);
}
public static bool DeploySubscribeTask(CRAClientLibrary client, SubscribeTask task, OperatorsToplogy topology)
{
try
{
client.DefineVertex(typeof(SubscribeOperator).Name.ToLower(), () => new SubscribeOperator());
CRAErrorCode status = client.InstantiateShardedVertex(task.OutputId, typeof(SubscribeOperator).Name.ToLower(),
task, task.DeployDescriptor.InstancesMap());
if (status == CRAErrorCode.Success)
{
foreach (string fromInputId in task.EndpointsDescriptor.FromInputs.Keys)
{
client.ConnectShardedVertices(task.VerticesConnectionsMap[fromInputId + task.OutputId]);
}
return(true);
}
else
{
return(false);
}
}
catch (Exception e)
{
Console.WriteLine("Error in deploying a sharded CRA Subscribe task. Please, double check your task configurations: " + e.ToString());
return(false);
}
}
private static async Task <bool> DeployTask(int taskIndex, TaskBase[] tasks, string[] tasksIds, Dictionary <string, bool> tasksDeploymentStatus, CRAClientLibrary client, OperatorsToplogy topology)
{
if (!tasksDeploymentStatus[tasksIds[taskIndex]])
{
bool isSuccessful = true;
foreach (var fromInput in tasks[taskIndex].EndpointsDescriptor.FromInputs.Keys)
{
int fromInputIndex = RetrieveTaskIndexOfOperator(fromInput, tasksIds);
isSuccessful = isSuccessful & await DeployTask(fromInputIndex, tasks, tasksIds, tasksDeploymentStatus, client, topology);
}
foreach (var fromSecondaryInput in tasks[taskIndex].EndpointsDescriptor.SecondaryFromInputs.Keys)
{
int fromSecondaryInputIndex = RetrieveTaskIndexOfOperator(fromSecondaryInput, tasksIds);
isSuccessful = isSuccessful & await DeployTask(fromSecondaryInputIndex, tasks, tasksIds, tasksDeploymentStatus, client, topology);
}
if (isSuccessful)
{
if (tasks[taskIndex].OperationType == OperatorType.Produce)
{
isSuccessful = isSuccessful & await DeployProduceTask(client, (ProduceTask)tasks[taskIndex], topology);
if (isSuccessful)
{
tasksDeploymentStatus[tasksIds[taskIndex]] = true;
}
}
else if (tasks[taskIndex].OperationType == OperatorType.Subscribe)
{
isSuccessful = isSuccessful & await DeploySubscribeTask(client, (SubscribeTask)tasks[taskIndex], topology);
if (isSuccessful)
{
tasksDeploymentStatus[tasksIds[taskIndex]] = true;
}
}
else if (tasks[taskIndex].OperationType == OperatorType.Move)
{
isSuccessful = isSuccessful & await DeployShuffleReduceTask(client, (ShuffleTask)tasks[taskIndex], topology);
if (isSuccessful)
{
tasksDeploymentStatus[tasksIds[taskIndex]] = true;
}
}
}
return(isSuccessful);
}
else
{
return(true);
}
}
private static bool DeployShuffleReduceTask(CRAClientLibrary client, ShuffleTask task, OperatorsToplogy topology)
{
try
{
client.DefineVertex(typeof(ShuffleOperator).Name.ToLower(), () => new ShuffleOperator());
CRAErrorCode status = client.InstantiateShardedVertex(task.ReducerVertexName, typeof(ShuffleOperator).Name.ToLower(),
task, task.DeployDescriptor.InstancesMap());
if (status == CRAErrorCode.Success)
{
foreach (string fromInputId in task.SecondaryEndpointsDescriptor.FromInputs.Keys)
{
client.ConnectShardedVertices(task.VerticesConnectionsMap[fromInputId + task.OutputId]);
}
var fromInput = task.EndpointsDescriptor.FromInputs.First().Key;
var newFromInputId = fromInput.Substring(0, fromInput.Length - 1);
client.ConnectShardedVertices(task.VerticesConnectionsMap[newFromInputId + task.ReducerVertexName]);
return(true);
}
else
{
return(false);
}
}
catch (Exception)
{
Console.WriteLine("Error in deploying a sharded CRA shuffle mapper task. Please, double check your task configurations");
return(false);
}
}
public override IShardedDataset <TKeyO, TPayloadO, TDataSetO> Deploy()
{
if (!_isDeployed)
{
OperatorsToplogy toplogy = OperatorsToplogy.GetInstance();
TaskBase subscribeTask = new SubscribeTask();
subscribeTask.OperationTypes = TransformUtils.FillBinaryTransformTypes(
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO));
subscribeTask.IsRightOperandInput = false;
OperatorTransforms subscribeInputTransforms = new OperatorTransforms();
Deploy(ref subscribeTask, ref toplogy, ref subscribeInputTransforms);
subscribeTask.InputIds.SetInputId1(subscribeTask.NextInputIds.InputId1);
subscribeTask.InputIds.SetInputId2(subscribeTask.NextInputIds.InputId2);
subscribeTask.OutputId = typeof(SubscribeOperator).Name.ToLower() + Guid.NewGuid().ToString();
subscribeTask.PrepareTaskTransformations(subscribeInputTransforms);
toplogy.AddOperatorBase(subscribeTask.OutputId, subscribeTask);
toplogy.AddOperatorInput(subscribeTask.OutputId, subscribeTask.InputIds.InputId1);
toplogy.AddOperatorSecondaryInput(subscribeTask.OutputId, subscribeTask.InputIds.InputId2);
toplogy.AddOperatorOutput(subscribeTask.InputIds.InputId1, subscribeTask.OutputId);
toplogy.AddOperatorOutput(subscribeTask.InputIds.InputId2, subscribeTask.OutputId);
if (subscribeTask.Transforms != null)
{
foreach (OperatorInputs inputs in subscribeTask.TransformsInputs)
{
toplogy.AddOperatorInput(subscribeTask.OutputId, inputs.InputId1);
toplogy.AddOperatorSecondaryInput(subscribeTask.OutputId, inputs.InputId2);
toplogy.AddOperatorOutput(inputs.InputId1, subscribeTask.OutputId);
toplogy.AddOperatorOutput(inputs.InputId2, subscribeTask.OutputId);
}
}
ClientTerminalTask clientTerminalTask = new ClientTerminalTask();
clientTerminalTask.InputIds.SetInputId1(subscribeTask.OutputId);
clientTerminalTask.OutputId = typeof(DetachedVertex).Name.ToLower() + Guid.NewGuid().ToString();
_craClient = new CRAClientLibrary();
_clientTerminal = _craClient.RegisterAsVertex(clientTerminalTask.OutputId);
toplogy.AddOperatorBase(clientTerminalTask.OutputId, clientTerminalTask);
toplogy.AddOperatorInput(clientTerminalTask.OutputId, clientTerminalTask.InputIds.InputId1);
toplogy.AddOperatorInput(clientTerminalTask.OutputId, clientTerminalTask.InputIds.InputId2);
toplogy.AddOperatorOutput(clientTerminalTask.InputIds.InputId1, clientTerminalTask.OutputId);
toplogy.AddOperatorOutput(clientTerminalTask.InputIds.InputId2, clientTerminalTask.OutputId);
_isDeployed = DeploymentUtils.DeployOperators(_craClient, toplogy);
if (_isDeployed)
{
_isDeployed = DeploymentUtils.DeployClientTerminal(_craClient, clientTerminalTask, ref _clientTerminal, toplogy);
}
else
{
return(null);
}
}
return(this);
}
private static ConcurrentDictionary <string, List <ConnectionInfoWithLocality> > PrepareVerticesConnectionsMap(
CRAClientLibrary client, TaskBase[] tasks, string[] tasksIds, Dictionary <string, TaskBase> tasksDictionary, OperatorsToplogy topology)
{
ConcurrentDictionary <string, List <ConnectionInfoWithLocality> > verticesConnectionsMap = new ConcurrentDictionary <string, List <ConnectionInfoWithLocality> >();
for (int i = 0; i < tasks.Length; i++)
{
int shardsCount = client.CountVertexShards(tasks[i].DeployDescriptor.InstancesMap());
tasks[i].EndpointsDescriptor = topology.OperatorsEndpointsDescriptors[tasksIds[i]];
foreach (string fromInputId in tasks[i].EndpointsDescriptor.FromInputs.Keys)
{
var flatFromToConnections = PrepareFlatConnectionsMap(client, shardsCount,
fromInputId, topology.OperatorsEndpointsDescriptors[fromInputId], tasksDictionary[fromInputId].DeployDescriptor.InstancesMap(),
tasks[i].OutputId, tasks[i].EndpointsDescriptor, tasks[i].DeployDescriptor.InstancesMap(), false);
verticesConnectionsMap.AddOrUpdate(fromInputId + tasks[i].OutputId,
flatFromToConnections, (key, value) => flatFromToConnections);
}
foreach (string secondaryFromInputId in tasks[i].EndpointsDescriptor.SecondaryFromInputs.Keys)
{
var flatFromToConnections = PrepareFlatConnectionsMap(client, shardsCount,
secondaryFromInputId, topology.OperatorsEndpointsDescriptors[secondaryFromInputId], tasksDictionary[secondaryFromInputId].DeployDescriptor.InstancesMap(),
tasks[i].OutputId, tasks[i].EndpointsDescriptor, tasks[i].DeployDescriptor.InstancesMap(), true);
verticesConnectionsMap.AddOrUpdate(secondaryFromInputId + tasks[i].OutputId,
flatFromToConnections, (key, value) => flatFromToConnections);
}
}
return(verticesConnectionsMap);
}
private void DeployMove(ref TaskBase task, ref OperatorsToplogy topology)
{
var isRightOperandInput = task.IsRightOperandInput;
OperatorInputs temporaryInputs = new OperatorInputs();
TaskBase shuffleTask = new ShuffleTask(_moveDescriptor);
shuffleTask.OperationTypes = TransformUtils.FillBinaryTransformTypes(
typeof(TKeyI1), typeof(TPayloadI1), typeof(TDataSetI1),
typeof(TKeyI2), typeof(TPayloadI2), typeof(TDataSetI2),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO));
shuffleTask.IsRightOperandInput = false;
OperatorTransforms shuffleInputTransforms = new OperatorTransforms();
(_input1 as IDeployable).Deploy(ref shuffleTask, ref topology, ref shuffleInputTransforms);
shuffleTask.PrepareTaskTransformations(shuffleInputTransforms);
(shuffleTask as ShuffleTask).MapperVertexName = "shufflemapper" + Guid.NewGuid().ToString();
(shuffleTask as ShuffleTask).ReducerVertexName = typeof(ShuffleOperator).Name.ToLower() + Guid.NewGuid().ToString();
shuffleTask.InputIds.SetInputId1(shuffleTask.NextInputIds.InputId1);
shuffleTask.InputIds.SetInputId2(shuffleTask.NextInputIds.InputId2);
shuffleTask.OutputId = (shuffleTask as ShuffleTask).ReducerVertexName;
OperatorTransforms shuffleTransforms = new OperatorTransforms();
shuffleTransforms.AddTransform(SerializationHelper.Serialize(_splitter),
OperatorType.MoveSplit.ToString(),
TransformUtils.FillBinaryTransformTypes(typeof(TKeyI1), typeof(TPayloadI1), typeof(TDataSetI1),
typeof(TKeyI2), typeof(TPayloadI2), typeof(TDataSetI2),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO)).ToString(),
shuffleTask.InputIds);
shuffleTransforms.AddTransform(SerializationHelper.Serialize(_merger),
OperatorType.MoveMerge.ToString(),
TransformUtils.FillBinaryTransformTypes(typeof(TKeyI1), typeof(TPayloadI1), typeof(TDataSetI1),
typeof(TKeyI2), typeof(TPayloadI2), typeof(TDataSetI2),
typeof(TKeyO), typeof(TPayloadO), typeof(TDataSetO)).ToString(),
shuffleTask.InputIds);
((ShuffleTask)shuffleTask).PrepareShuffleTransformations(shuffleTransforms);
topology.AddShuffleOperator((shuffleTask as ShuffleTask).MapperVertexName, (shuffleTask as ShuffleTask).ReducerVertexName, shuffleTask as ShuffleTask);
topology.AddOperatorInput((shuffleTask as ShuffleTask).MapperVertexName, shuffleTask.InputIds.InputId1);
topology.AddOperatorSecondaryInput((shuffleTask as ShuffleTask).MapperVertexName, shuffleTask.InputIds.InputId2);
topology.AddOperatorOutput(shuffleTask.InputIds.InputId1, (shuffleTask as ShuffleTask).MapperVertexName);
topology.AddOperatorOutput(shuffleTask.InputIds.InputId2, (shuffleTask as ShuffleTask).MapperVertexName);
if (shuffleTask.Transforms != null)
{
foreach (OperatorInputs inputs in shuffleTask.TransformsInputs)
{
topology.AddOperatorInput((shuffleTask as ShuffleTask).MapperVertexName, inputs.InputId1);
topology.AddOperatorSecondaryInput((shuffleTask as ShuffleTask).MapperVertexName, inputs.InputId2);
topology.AddOperatorOutput(inputs.InputId1, (shuffleTask as ShuffleTask).MapperVertexName);
topology.AddOperatorOutput(inputs.InputId2, (shuffleTask as ShuffleTask).MapperVertexName);
}
}
// Update the inputs and types for the next operation
task.InputIds.SetInputId1(shuffleTask.OutputId);
task.OperationTypes.SetInputKeyType(typeof(TKeyO));
task.OperationTypes.SetInputPayloadType(typeof(TPayloadO));
task.OperationTypes.SetInputDatasetType(typeof(TDataSetO));
if (isRightOperandInput)
{
temporaryInputs.InputId2 = shuffleTask.OutputId;
}
else
{
temporaryInputs.InputId1 = shuffleTask.OutputId;
}
task.NextInputIds = temporaryInputs;
}
private static ConcurrentDictionary <string, List <ConnectionInfoWithLocality> > PrepareVerticesConnectionsMap(
CRAClientLibrary client, TaskBase[] tasks, string[] tasksIds, Dictionary <string, TaskBase> tasksDictionary, OperatorsToplogy topology)
{
ConcurrentDictionary <string, List <ConnectionInfoWithLocality> > verticesConnectionsMap = new ConcurrentDictionary <string, List <ConnectionInfoWithLocality> >();
for (int i = 0; i < tasks.Length; i++)
{
int shardsCount = client.CountVertexShards(tasks[i].DeployDescriptor.InstancesMap());
tasks[i].EndpointsDescriptor = topology.OperatorsEndpointsDescriptors[tasksIds[i]];
if (tasks[i].OperationType == OperatorType.Move)
{
var shuffleTask = (ShuffleTask)tasks[i];
foreach (string fromInputId in shuffleTask.SecondaryEndpointsDescriptor.FromInputs.Keys)
{
var flatFromToConnections = PrepareFlatConnectionsMap(client, shardsCount,
fromInputId, topology.OperatorsEndpointsDescriptors[fromInputId], tasksDictionary[fromInputId].DeployDescriptor.InstancesMap(),
shuffleTask.ReducerVertexName, shuffleTask.EndpointsDescriptor, shuffleTask.DeployDescriptor.InstancesMap());
verticesConnectionsMap.AddOrUpdate(fromInputId + shuffleTask.OutputId,
flatFromToConnections, (key, value) => flatFromToConnections);
}
var fromInput = shuffleTask.EndpointsDescriptor.FromInputs.First().Key;
var newFromInputId = fromInput.Substring(0, fromInput.Length - 1);
var shuffleFromToConnections = PrepareShuffleConnectionsMap(client, shardsCount,
newFromInputId, topology.OperatorsEndpointsDescriptors[newFromInputId], tasksDictionary[newFromInputId].DeployDescriptor.InstancesMap(),
shuffleTask.ReducerVertexName, shuffleTask.EndpointsDescriptor, shuffleTask.DeployDescriptor.InstancesMap());
verticesConnectionsMap.AddOrUpdate(newFromInputId + shuffleTask.ReducerVertexName,
shuffleFromToConnections, (key, value) => shuffleFromToConnections);
}
else
{
foreach (string fromInputId in tasks[i].EndpointsDescriptor.FromInputs.Keys)
{
var flatFromToConnections = PrepareFlatConnectionsMap(client, shardsCount,
fromInputId, topology.OperatorsEndpointsDescriptors[fromInputId], tasksDictionary[fromInputId].DeployDescriptor.InstancesMap(),
tasks[i].OutputId, tasks[i].EndpointsDescriptor, tasks[i].DeployDescriptor.InstancesMap());
verticesConnectionsMap.AddOrUpdate(fromInputId + tasks[i].OutputId,
flatFromToConnections, (key, value) => flatFromToConnections);
}
}
}
return(verticesConnectionsMap);
}
