use of edu.iu.dsc.tws.api.config.Config in project twister2 by DSC-SPIDAL.
the class KafkaExample method main.
public static void main(String[] args) throws ParseException {
Options options = new Options();
options.addOption(CLI_SERVER, true, "Kafka bootstrap server in the format host:port");
options.addOption(CLI_TOPICS, true, "Set of topics in the format topic1,topic2");
CommandLineParser cliParser = new DefaultParser();
CommandLine cli = cliParser.parse(options, args);
HashMap<String, Object> configs = new HashMap<>();
configs.put(CLI_SERVER, "localhost:9092");
configs.put(CLI_TOPICS, Collections.singleton("test2"));
if (cli.hasOption(CLI_SERVER)) {
configs.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, cli.getOptionValue(CLI_SERVER));
}
if (cli.hasOption(CLI_TOPICS)) {
String topics = cli.getOptionValue(CLI_TOPICS);
Set<String> topicsSet = Arrays.stream(topics.split(",")).map(String::trim).collect(Collectors.toSet());
configs.put(CLI_TOPICS, topicsSet);
}
Config config = ResourceAllocator.loadConfig(new HashMap<>());
JobConfig jobConfig = new JobConfig();
jobConfig.putAll(configs);
Twister2Job twister2Job;
twister2Job = Twister2Job.newBuilder().setJobName(KafkaExample.class.getName()).setWorkerClass(KafkaExample.class).addComputeResource(1, 1024, 1).setConfig(jobConfig).build();
// now submit the job
Twister2Submitter.submitJob(twister2Job, config);
}
use of edu.iu.dsc.tws.api.config.Config in project twister2 by DSC-SPIDAL.
the class TeraSort method main.
public static void main(String[] args) throws ParseException {
Config config = ResourceAllocator.loadConfig(new HashMap<>());
JobConfig jobConfig = new JobConfig();
Options options = new Options();
// file based mode configuration
options.addOption(createOption(ARG_INPUT_FILE, true, "Path to the file containing input tuples. " + "Path can be specified with %d, where it will be replaced by task index. For example," + "input-%d, will be considered as input-0 in source task having index 0.", false));
// non-file based mode configurations
options.addOption(createOption(ARG_SIZE, true, "Total Data Size in GigaBytes for all workers.", true));
options.addOption(createOption(ARG_KEY_SIZE, true, "Size of the key in bytes of a single Tuple", true));
options.addOption(createOption(ARG_KEY_SEED, true, "Size of the key in bytes of a single Tuple", false));
options.addOption(createOption(ARG_VALUE_SIZE, true, "Size of the value in bytes of a single Tuple", true));
// resources
options.addOption(createOption(ARG_RESOURCE_CPU, true, "Amount of CPUs to allocate per instance", true));
options.addOption(createOption(ARG_RESOURCE_MEMORY, true, "Amount of Memory in mega bytes to allocate per instance", true));
options.addOption(createOption(ARG_RESOURCE_INSTANCES, true, "No. of instances", true));
options.addOption(createOption(ARG_RESOURCE_VOLATILE_DISK, true, "Volatile Disk for each worker at K8s", false));
options.addOption(createOption(ARG_WORKERS_PER_POD, true, "Workers per pod in Kubernetes", false));
// tasks and sources counts
options.addOption(createOption(ARG_TASKS_SOURCES, true, "No of source tasks", true));
options.addOption(createOption(ARG_TASKS_SINKS, true, "No of sink tasks", true));
// optional configurations (tune performance)
options.addOption(createOption(ARG_TUNE_MAX_BYTES_IN_MEMORY, true, "Maximum bytes to keep in memory", false));
options.addOption(createOption(ARG_TUNE_MAX_SHUFFLE_FILE_SIZE, true, "Maximum records to keep in memory", false));
options.addOption(createOption(ARG_BENCHMARK_METADATA, true, "Auto generated argument by benchmark suite", false));
// output folder
options.addOption(createOption(ARG_OUTPUT_FOLDER, true, "Folder to save output files", false));
// fixed schema
options.addOption(createOption(ARG_FIXED_SCHEMA, false, "Use fixed schema feature", false));
// verify option
options.addOption(createOption(VERIFY, false, "Verify whether the results are sorted.", false));
CommandLineParser commandLineParser = new DefaultParser();
CommandLine cmd = commandLineParser.parse(options, args);
if (cmd.hasOption(ARG_INPUT_FILE)) {
jobConfig.put(ARG_INPUT_FILE, cmd.getOptionValue(ARG_INPUT_FILE));
} else {
jobConfig.put(ARG_SIZE, Double.valueOf(cmd.getOptionValue(ARG_SIZE)));
jobConfig.put(ARG_VALUE_SIZE, Integer.valueOf(cmd.getOptionValue(ARG_VALUE_SIZE)));
jobConfig.put(ARG_KEY_SIZE, Integer.valueOf(cmd.getOptionValue(ARG_KEY_SIZE)));
}
// in GB, default value is 4GB
double volatileDisk = 0.0;
if (cmd.hasOption(ARG_RESOURCE_VOLATILE_DISK)) {
volatileDisk = Double.valueOf(cmd.getOptionValue(ARG_RESOURCE_VOLATILE_DISK));
}
// default value is 1
int workersPerPod = 1;
if (cmd.hasOption(ARG_WORKERS_PER_POD)) {
workersPerPod = Integer.valueOf(cmd.getOptionValue(ARG_WORKERS_PER_POD));
}
jobConfig.put(ARG_TASKS_SOURCES, Integer.valueOf(cmd.getOptionValue(ARG_TASKS_SOURCES)));
jobConfig.put(ARG_TASKS_SINKS, Integer.valueOf(cmd.getOptionValue(ARG_TASKS_SINKS)));
jobConfig.put(ARG_RESOURCE_INSTANCES, Integer.valueOf(cmd.getOptionValue(ARG_RESOURCE_INSTANCES)) * workersPerPod);
if (cmd.hasOption(ARG_TUNE_MAX_BYTES_IN_MEMORY)) {
long maxBytesInMemory = Long.valueOf(cmd.getOptionValue(ARG_TUNE_MAX_BYTES_IN_MEMORY));
jobConfig.put(SHUFFLE_MAX_BYTES_IN_MEMORY, maxBytesInMemory);
// for benchmark service
jobConfig.put(ARG_TUNE_MAX_BYTES_IN_MEMORY, maxBytesInMemory);
}
if (cmd.hasOption(ARG_TUNE_MAX_SHUFFLE_FILE_SIZE)) {
long maxRecordsInMemory = Long.valueOf(cmd.getOptionValue(ARG_TUNE_MAX_SHUFFLE_FILE_SIZE));
jobConfig.put(SHUFFLE_MAX_FILE_SIZE, maxRecordsInMemory);
jobConfig.put(ARG_TUNE_MAX_SHUFFLE_FILE_SIZE, maxRecordsInMemory);
}
if (cmd.hasOption(ARG_BENCHMARK_METADATA)) {
jobConfig.put(ARG_BENCHMARK_METADATA, cmd.getOptionValue(ARG_BENCHMARK_METADATA));
jobConfig.put(ARG_RUN_BENCHMARK, true);
}
if (cmd.hasOption(ARG_OUTPUT_FOLDER)) {
jobConfig.put(ARG_OUTPUT_FOLDER, cmd.getOptionValue(ARG_OUTPUT_FOLDER));
}
if (cmd.hasOption(ARG_FIXED_SCHEMA)) {
jobConfig.put(ARG_FIXED_SCHEMA, true);
}
if (cmd.hasOption(VERIFY)) {
jobConfig.put(VERIFY, true);
}
Twister2Job twister2Job;
twister2Job = Twister2Job.newBuilder().setJobName("terasort").setWorkerClass(TeraSort.class.getName()).addComputeResource(Double.valueOf(cmd.getOptionValue(ARG_RESOURCE_CPU)), Integer.valueOf(cmd.getOptionValue(ARG_RESOURCE_MEMORY)), volatileDisk, Integer.valueOf(cmd.getOptionValue(ARG_RESOURCE_INSTANCES)), workersPerPod).setConfig(jobConfig).build();
Twister2Submitter.submitJob(twister2Job, config);
}
use of edu.iu.dsc.tws.api.config.Config in project twister2 by DSC-SPIDAL.
the class TeraSort method execute.
@Override
public void execute(WorkerEnvironment workerEnv) {
int workerID = workerEnv.getWorkerId();
ComputeEnvironment cEnv = ComputeEnvironment.init(workerEnv);
Config config = workerEnv.getConfig();
resultsRecorder = new BenchmarkResultsRecorder(config, workerID == 0);
Timing.setDefaultTimingUnit(TimingUnit.MILLI_SECONDS);
final String filePath = config.getStringValue(ARG_INPUT_FILE, null);
final int keySize = config.getIntegerValue(ARG_KEY_SIZE, 10);
final int valueSize = config.getIntegerValue(ARG_VALUE_SIZE, 90);
// Sampling Graph : if file based only
TaskPartitioner taskPartitioner;
if (filePath != null) {
ComputeGraphBuilder samplingGraph = ComputeGraphBuilder.newBuilder(config);
samplingGraph.setMode(OperationMode.BATCH);
Sampler samplerTask = new Sampler();
samplingGraph.addSource(TASK_SAMPLER, samplerTask, config.getIntegerValue(ARG_TASKS_SOURCES, 4));
SamplerReduce samplerReduce = new SamplerReduce();
samplingGraph.addCompute(TASK_SAMPLER_REDUCE, samplerReduce, config.getIntegerValue(ARG_RESOURCE_INSTANCES, 4)).allreduce(TASK_SAMPLER).viaEdge(EDGE).withReductionFunction(byte[].class, (minMax1, minMax2) -> {
byte[] min1 = Arrays.copyOfRange(minMax1, 0, keySize);
byte[] max1 = Arrays.copyOfRange(minMax1, keySize, minMax1.length);
byte[] min2 = Arrays.copyOfRange(minMax2, 0, keySize);
byte[] max2 = Arrays.copyOfRange(minMax2, keySize, minMax2.length);
byte[] newMinMax = new byte[keySize * 2];
byte[] min = min1;
byte[] max = max1;
if (ByteArrayComparator.getInstance().compare(min1, min2) > 0) {
min = min2;
}
if (ByteArrayComparator.getInstance().compare(max1, max2) < 0) {
max = max2;
}
System.arraycopy(min, 0, newMinMax, 0, keySize);
System.arraycopy(max, 0, newMinMax, keySize, keySize);
return newMinMax;
});
ComputeGraph sampleGraphBuild = samplingGraph.build();
ExecutionPlan sampleTaskPlan = cEnv.getTaskExecutor().plan(sampleGraphBuild);
cEnv.getTaskExecutor().execute(sampleGraphBuild, sampleTaskPlan);
DataObject<byte[]> output = cEnv.getTaskExecutor().getOutput("sample-reduce");
LOG.info("Sample output received");
taskPartitioner = new TaskPartitionerForSampledData(output.getPartitions()[0].getConsumer().next(), keySize);
} else {
taskPartitioner = new TaskPartitionerForRandom();
}
// Sort Graph
ComputeGraphBuilder teraSortTaskGraph = ComputeGraphBuilder.newBuilder(config);
teraSortTaskGraph.setMode(OperationMode.BATCH);
BaseSource dataSource;
if (filePath == null) {
dataSource = new RandomDataSource();
} else {
dataSource = new FileDataSource();
}
teraSortTaskGraph.addSource(TASK_SOURCE, dataSource, config.getIntegerValue(ARG_TASKS_SOURCES, 4));
Receiver receiver = new Receiver();
KeyedGatherConfig keyedGatherConfig = teraSortTaskGraph.addCompute(TASK_RECV, receiver, config.getIntegerValue(ARG_TASKS_SINKS, 4)).keyedGather(TASK_SOURCE).viaEdge(EDGE).withDataType(MessageTypes.BYTE_ARRAY).withKeyType(MessageTypes.BYTE_ARRAY).withTaskPartitioner(taskPartitioner).useDisk(true).sortBatchByKey(ByteArrayComparator.getInstance()).groupBatchByKey(false);
if (config.getBooleanValue(ARG_FIXED_SCHEMA, false)) {
LOG.info("Using fixed schema feature with message size : " + (keySize + valueSize) + " and key size : " + keySize);
keyedGatherConfig.withMessageSchema(MessageSchema.ofSize(keySize + valueSize, keySize));
}
ComputeGraph computeGraph = teraSortTaskGraph.build();
ExecutionPlan executionPlan = cEnv.getTaskExecutor().plan(computeGraph);
cEnv.getTaskExecutor().execute(computeGraph, executionPlan);
cEnv.close();
LOG.info("Finished Sorting...");
}
use of edu.iu.dsc.tws.api.config.Config in project twister2 by DSC-SPIDAL.
the class WordCountWorker method execute.
@Override
public void execute(WorkerEnvironment wEnv) {
this.workerEnv = wEnv;
this.workerId = workerEnv.getWorkerId();
taskStages.add(NO_OF_TASKS);
taskStages.add(NO_OF_TASKS);
// lets create the task plan
this.logicalPlan = Utils.createStageLogicalPlan(workerEnv, taskStages);
setupTasks();
// create the communication
wordAggregator = new WordAggregator();
keyGather = new BKeyedReduce(workerEnv.getCommunicator(), logicalPlan, sources, destinations, new ReduceFunction() {
@Override
public void init(Config cfg, DataFlowOperation op, Map<Integer, List<Integer>> expectedIds) {
}
@Override
public Object reduce(Object t1, Object t2) {
return (Integer) t1 + (Integer) t2;
}
}, wordAggregator, MessageTypes.OBJECT, MessageTypes.INTEGER, new HashingSelector());
// assign the task ids to the workers, and run them using threads
scheduleTasks();
// progress the communication
progress();
// close communication
workerEnv.close();
}
use of edu.iu.dsc.tws.api.config.Config in project twister2 by DSC-SPIDAL.
the class CheckpointingTaskExample method main.
public static void main(String[] args) {
int numberOfWorkers = 4;
if (args.length == 1) {
numberOfWorkers = Integer.valueOf(args[0]);
}
// first load the configurations from command line and config files
Config config = ResourceAllocator.loadConfig(new HashMap<>());
// lets put a configuration here
JobConfig jobConfig = new JobConfig();
jobConfig.put("parallelism", numberOfWorkers);
Twister2Job twister2Job = Twister2Job.newBuilder().setJobName("hello-checkpointing-job").setWorkerClass(CheckpointingTaskExample.class).addComputeResource(1, 1024, numberOfWorkers).setConfig(jobConfig).build();
// now submit the job
Twister2Submitter.submitJob(twister2Job, config);
}
Aggregations