Tigon Developer Guide

Tigon Flows

Flows are developer-implemented, real-time stream processors. They are comprised of one or more Flowlets that are wired together into a directed acyclic graph or DAG.

Flowlets pass DataObjects between one another. Each Flowlet is able to perform custom logic and execute data operations for each individual data object processed. All data operations happen in a consistent and durable way.

When processing a single input object, all operations, including the removal of the object from the input, and emission of data to the outputs, are executed in a transaction. This provides us with Atomicity, Consistency, Isolation, and Durability (ACID) properties, and helps assure a unique and core property of the Flow system: it guarantees atomic and “exactly-once” processing of each input object by each Flowlet in the DAG.

A Flow runs as a YARN application and each Flowlet instance runs in its own container. Each Flowlet in the DAG can have multiple concurrent instances, each consuming a partition of the Flowlet’s inputs.

To put data into your Flow, you implement a Flowlet to generate or pull the data from an external source.

The Flow interface allows you to specify the Flow’s metadata, Flowlets, and Flowlet connections.

To create a Flow, implement Flow via a configure method that returns a FlowSpecification using FlowSpecification.Builder():

class MyExampleFlow implements Flow {
  public FlowSpecification configure() {
    return FlowSpecification.Builder.with()
      .setDescription("Flow for showing examples")
        .add("flowlet1", new MyExampleFlowlet(), 1)
        .add("flowlet2", new MyExampleFlowlet2(), 3)

In this example, the name, description, with (or without) Flowlets, and connections are specified before building the Flow. The code above creates a Flow with two Flowlets; the first named flowlet1, a MyExampleFlowlet with 1 instance, and the second flowlet2, a MyExampleFlowlet2 with 3 instances.

Tigon Flowlets

Flowlets, the basic building blocks of a Flow, represent each individual processing node within a Flow. Flowlets consume data objects from their inputs and execute custom logic on each data object, allowing you to perform data operations as well as emit data objects to the Flowlet’s outputs. Flowlets specify an initialize() method, which is executed at the startup of each instance of a Flowlet before it receives any data.

The example below shows a Flowlet that reads Double values, rounds them, and emits the results. It has a simple configuration method and doesn’t do anything for initialization or destruction:

class RoundingFlowlet implements Flowlet {

  public FlowletSpecification configure() {
    return FlowletSpecification.Builder.with().
      setDescription("A rounding Flowlet").

    public void initialize(FlowletContext context) throws Exception {

  public void destroy() {

  OutputEmitter<Long> output;
  public void round(Double number) {

The most interesting method of this Flowlet is round(), the method that does the actual processing. It uses an output emitter to send data to its output. This is the only way that a Flowlet can emit output to another connected Flowlet:

OutputEmitter<Long> output;
public void round(Double number) {

Note that the Flowlet declares the output emitter but does not initialize it. The Flow system initializes and injects its implementation at runtime.

The method is annotated with @ProcessInput—this tells the Flow system that this method can process input data.

You can overload the process method of a Flowlet by adding multiple methods with different input types. When an input object comes in, the Flowlet will call the method that matches the object’s type:

OutputEmitter<Long> output;

public void round(Double number) {
public void round(Float number) {

If you define multiple process methods, a method will be selected based on the input object’s origin; that is, the name of a Stream or the name of an output of a Flowlet.

A Flowlet that emits data can specify this name using an annotation on the output emitter. In the absence of this annotation, the name of the output defaults to “out”:

OutputEmitter<String> out;

Data objects emitted through this output can then be directed to a process method of a receiving Flowlet by annotating the method with the origin name:

public void tokenizeCode(String text) {
  ... // perform fancy code tokenization

Input Context

A process method can have an additional parameter, the InputContext. The input context provides information about the input object, such as its origin and the number of times the object has been retried. For example, this Flowlet tokenizes text in a smart way and uses the input context to decide which tokenizer to use:

public void tokenize(String text, InputContext context) throws Exception {
  Tokenizer tokenizer;
  // If this failed before, fall back to simple white space
  if (context.getRetryCount() > 0) {
    tokenizer = new WhiteSpaceTokenizer();
  // Is this code? If its origin is named "code", then assume yes
  else if ("code".equals(context.getOrigin())) {
    tokenizer = new CodeTokenizer();
  else {
    // Use the smarter tokenizer
    tokenizer = new NaturalLanguageTokenizer();
  for (String token : tokenizer.tokenize(text)) {

Type Projection

Flowlets perform an implicit projection on the input objects if they do not match exactly what the process method accepts as arguments. This allows you to write a single process method that can accept multiple compatible types. For example, if you have a process method:

count(String word) {

and you send data of type Long to this Flowlet, then that type does not exactly match what the process method expects. You could now write another process method for Long numbers:

@ProcessInput count(Long number) {

and you could do that for every type that you might possibly want to count, but that would be rather tedious. Type projection does this for you automatically. If no process method is found that matches the type of an object exactly, it picks a method that is compatible with the object.

In this case, because Long can be converted into a String, it is compatible with the original process method. Other compatible conversions are:

  • Every primitive type that can be converted to a String is compatible with String.

  • Any numeric type is compatible with numeric types that can represent it. For example, int is compatible with long, float and double, and long is compatible with float and double, but long is not compatible with int because int cannot represent every long value.

  • A byte array is compatible with a ByteBuffer and vice versa.

  • A collection of type A is compatible with a collection of type B, if type A is compatible with type B. Here, a collection can be an array or any Java Collection. Hence, a List<Integer> is compatible with a String[] array.

  • Two maps are compatible if their underlying types are compatible. For example, a TreeMap<Integer, Boolean> is compatible with a HashMap<String, String>.

  • Other Java objects can be compatible if their fields are compatible. For example, in the following class Point is compatible with Coordinate, because all common fields between the two classes are compatible. When projecting from Point to Coordinate, the color field is dropped, whereas the projection from Coordinate to Point will leave the color field as null:

    class Point {
      private int x;
      private int y;
      private String color;
    class Coordinates {
      int x;
      int y;

Type projections help you keep your code generic and reusable. They also interact well with inheritance. If a Flowlet can process a specific object class, then it can also process any subclass of that class.

Flowlet Method and @Tick Annotation

A Flowlet’s method can be annotated with @Tick. Instead of processing data objects from a Flowlet input, this method is invoked periodically, without arguments. This can be used, for example, to generate data, or pull data from an external data source periodically on a fixed cadence.

In this code snippet from the CountRandom example, the @Tick method in the Flowlet emits random numbers:

public class RandomSource extends AbstractFlowlet {

  private OutputEmitter<Integer> randomOutput;

  private final Random random = new Random();

  @Tick(delay = 1L, unit = TimeUnit.MILLISECONDS)
  public void generate() throws InterruptedException {

Note: @Tick method calls are serialized; subsequent calls to the tick method will be made only after the previous @Tick method call has returned.


There are multiple ways to connect the Flowlets of a Flow. The most common form is to use the Flowlet name. Because the name of each Flowlet defaults to its class name, when building the Flow specification you can simply write:

  .add(new RandomGenerator())
  .add(new RoundingFlowlet())

If you have multiple Flowlets of the same class, you can give them explicit names:

  .add("random", new RandomGenerator())
  .add("generator", new RandomGenerator())
  .add("rounding", new RoundingFlowlet())

Batch Execution

By default, a Flowlet processes a single data object at a time within a single transaction. To increase throughput, you can also process a batch of data objects within the same transaction:

public void process(String words) {

For the above batch example, the process method will be called up to 100 times per transaction, with different data objects read from the input each time it is called.

If you are interested in knowing when a batch begins and ends, you can use an Iterator as the method argument:

public void process(Iterator<String> words) {

In this case, the process will be called once per transaction and the Iterator will contain up to 100 data objects read from the input.

Flowlets and Instances

You can have one or more instances of any given Flowlet, each consuming a disjoint partition of each input. You can control the number of instances through the command-line interface (CLI) in Distributed Mode.. This enables you to scale your application to meet capacity at runtime.

In Tigon Standalone, multiple Flowlet instances are run in threads, so in some cases actual performance may not be improved. However, in the Tigon Distributed, each Flowlet instance runs in its own Java Virtual Machine (JVM) with independent compute resources. Scaling the number of Flowlets can improve performance and have a major impact depending on your implementation.

Partitioning Strategies

As mentioned above, if you have multiple instances of a Flowlet the input queue is partitioned among the Flowlets. The partitioning can occur in different ways, and each Flowlet can specify one of these three partitioning strategies:

  • First-in first-out (FIFO): Default mode. In this mode, every Flowlet instance receives the next available data object in the queue. However, since multiple consumers may compete for the same data object, access to the queue must be synchronized. This may not always be the most efficient strategy.
  • Round-robin: With this strategy, the number of items is distributed evenly among the instances. In general, round-robin is the most efficient partitioning. Though more efficient than FIFO, it is not ideal when the application needs to group objects into buckets according to business logic. In those cases, hash-based partitioning is preferable.
  • Hash-based: If the emitting Flowlet annotates each data object with a hash key, this partitioning ensures that all objects of a given key are received by the same consumer instance. This can be useful for aggregating by key, and can help reduce write conflicts when writing to HBase in distributed mode.

Let’s look at a case where a Hash Partition is required. Suppose we have a Flowlet that counts words:

public class Counter extends AbstractFlowlet {

  private Map<String, Integer> wordCount = Maps.newHashMap();

  public void process(String word) {
    int count = wordCount.containsKey(word) ? (wordCount.get(word) + 1) : 1;
    wordCount.put(word, count);

This Flowlet uses the default strategy of FIFO. To increase the throughput when this Flowlet has many instances, we can specify round-robin partitioning:

public void process(String word) {
  int count = wordCount.containsKey(word) ? (wordCount.get(word) + 1) : 1;
  wordCount.put(word, count);

Now, if we have three instances of this Flowlet, every instance will receive every third word. For example, for the sequence of words in the sentence, “I scream, you scream, we all scream for ice cream”:

  • The first instance receives the words: I scream scream cream
  • The second instance receives the words: scream we for
  • The third instance receives the words: you all ice

The potential problem with this is that the first two instances might both attempt to increment the counter for the word scream and thus lead to an incorrect count (since the count is stored in-memory in different flowlets). To avoid conflicts, we can use hash-based partitioning:

public void process(String word) {
  int count = wordCount.containsKey(word) ? (wordCount.get(word) + 1) : 1;
  wordCount.put(word, count);

Now only one of the Flowlet instances will receive the word scream, and there can be no more incorrect counts. Note that in order to use hash-based partitioning, the emitting Flowlet must annotate each data object with the partitioning key:

private OutputEmitter<String> wordOutput;
public void process(StreamEvent event) {
  // emit the word with the partitioning key name "wordHash"
  wordOutput.emit(word, "wordHash", word.hashCode());

Note that the emitter must use the same name (“wordHash”) for the key that the consuming Flowlet specifies as the partitioning key. If the output is connected to more than one Flowlet, you can also annotate a data object with multiple hash keys—each consuming Flowlet can then use different partitioning. This is useful if you want to aggregate by multiple keys, such as counting purchases by product ID as well as by customer ID.

Partitioning can be combined with batch execution:

public void process(Iterator<String> words) {


The data flows between Flowlets are implemented through Queues. In the Standalone Mode, this is implemented through in-memory data structures. In Distributed Mode, it is implemented using HBase Tables. This provides reliability and fault-tolerance to the Flow system such that when a Flowlet instances dies, it is respawned and it starts reading from the next event in the queue.

Flow Transaction System

The Need for Transactions

A Flowlet processes the data objects received on its inputs one at a time. While processing a single input object, all operations, including the removal of the data from the input, and emission of data to the outputs, are executed in a transaction. This provides us with ACID—atomicity, consistency, isolation, and durability properties:

  • The process method runs under read isolation to ensure that it does not see dirty writes (uncommitted writes from concurrent processing) in any of its reads. It does see, however, its own writes.
  • A failed attempt to process an input object leaves the data in a consistent state; it does not leave partial writes behind.
  • All writes and emission of data are committed atomically; either all of them or none of them are persisted.
  • After processing completes successfully, all its writes are persisted in a durable way.

In case of failure, the state of the data is unchanged and processing of the input object can be reattempted. This ensures “exactly-once” processing of each object.

OCC: Optimistic Concurrency Control

Tigon uses Optimistic Concurrency Control (OCC) to implement transactions. Unlike most relational databases that use locks to prevent conflicting operations between transactions, under OCC we allow these conflicting writes to happen. When the transaction is committed, we can detect whether it has any conflicts: namely, if during the lifetime of the transaction, another transaction committed a write for one of the same keys that the transaction has written. In that case, the transaction is aborted and all of its writes are rolled back.

In other words: If two overlapping transactions modify the same row, then the transaction that commits first will succeed, but the transaction that commits last is rolled back due to a write conflict.

Optimistic Concurrency Control is lockless and therefore avoids problems such as idle processes waiting for locks, or even worse, deadlocks. However, it comes at the cost of rollback in case of write conflicts. We can only achieve high throughput with OCC if the number of conflicts is small. It is therefore a good practice to reduce the probability of conflicts wherever possible.

Here are some rules to follow for Flows, Flowlets and Procedures:

  • Keep transactions short. Tigon attempts to delay the beginning of each transaction as long as possible. For instance, if your Flowlet only performs write operations, but no read operations, then all writes are deferred until the process method returns. They are then performed and transacted, together with the removal of the processed object from the input, in a single batch execution. This minimizes the duration of the transaction.
  • However, if your Flowlet performs a read, then the transaction must begin at the time of the read. If your Flowlet performs long-running computations after that read, then the transaction runs longer, too, and the risk of conflicts increases. It is therefore a good practice to perform reads as late in the process method as possible.
  • There are two ways to perform an increment: As a write operation that returns nothing, or as a read-write operation that returns the incremented value. If you perform the read-write operation, then that forces the transaction to begin, and the chance of conflict increases. Unless you depend on that return value, you should always perform an increment only as a write operation.
  • Use hash-based partitioning for the inputs of highly concurrent Flowlets that perform writes. This helps reduce concurrent writes to the same key from different instances of the Flowlet.

Keeping these guidelines in mind will help you write more efficient and faster-performing code.

Writing to HBase Transactionally From a Flowlet

Tigon internally uses Tephra extensively to complete transactional operations. Tephra can also be leveraged by developers to write to HBase transactionally, and in so doing obtain Tephra’s ACID properties of transactions. To do this, wrap an HTable instance (the variable htable in the example below) with Tephra’s TransactionAwareHTable and add it to the Flowlet’s context:

public static final class TransactionalFlowlet extends AbstractFlowlet {

  private OutputEmitter<Integer> intEmitter;
  private int i = 0;

  public void initialize(FlowletContext context) throws Exception {
    // Acquire HTable instance
    TransactionAwareHTable txAwareHTable = new TransactionAwareHTable(htable);

  @Tick(delay = 1L, unit = TimeUnit.SECONDS)
  public void process() throws Exception {
    Put put = new Put(Bytes.toBytes(“testRow”));
    put.add(Bytes.toBytes(“testFamily”), Bytes.toBytes(“testQualifier”), Bytes.toBytes(i));

    Integer value = ++i;
    intEmitter.emit(value, "integer", value.hashCode());

Operations performed on the TransactionAwareHTable instance inside the initialize, destroy, and each of the process methods are committed as a single transaction. Exceptions thrown in any of these methods will result in a rollback of the entire transaction.

Using TigonSQL

TigonSQL provides an in-memory SQL streaming engine and can perform filtering, aggregation, and joins of Streams. This can be highly useful for use cases where a large ingestion rate is required.

However, it must be noted that the data in TigonSQL is held in-memory and thus there is a possibility of data loss if the Flowlet container or the Stream Engine fails. The transaction guarantees and the persistence of data comes into play only after the results of the AbstractInputFlowlet is emitted and is persisted in HBase Tables through Queues. A further consideration is that in the current implementation, the instance count of AbstractInputFlowlet is limited to a single instance.

In order to use the TigonSQL library in your flow, you need a Flowlet that extends AbstractInputFlowlet. To use the StreamEngine, implement the create method. The building blocks of the StreamEngine are the StreamSchema objects, and the addJSONInput and addQuery methods.

StreamSchema objects are constructed using the StreamSchema Builder. These objects represent the input schema of a Stream, with these fields allowed to be part of the input schema:

  • BOOL
  • INT
  • LONG

The Builder’s addField method takes the name of the field, the field type and the SlidingWindowAttribute. The sliding window attribute is used to annotate that a field is monotonically increasing or decreasing. A field with this attribute set to increasing or decreasing might be required for certain SQL queries; for example, “GROUP BY increasingField”.

Once one or more StreamSchemas are created, they are added as an input using the addJSONInput method. This method takes the name of the input stream and the schema of the stream. Once the inputs streams have been added, one or more SQL queries can be defined using an addQuery method. The addQuery method takes the name of the query and the SQL statement.

The output of the SQL queries will be POJOs, whose output class you can define. The names of the members of the output class should match the names used in the SQL query statement. In the example given below, DataPacket is one such POJO class.

In order to process the output of SQL queries, you’ll need to annotate the methods with @QueryOutput(<QueryName>). You can then choose to process the objects in that method or emit the object to a subsequent Flowlet. In the example given below, emitData is a method which is annotated with QueryOutput and it emits the DataPacket object to the next Flowlet:

public class SQLFlowlet extends AbstractInputFlowlet {
    private OutputEmitter<DataPacket> dataEmitter;
    private final Logger LOG = LoggerFactory.getLogger(SQLFlowlet.class);

    public void create() {
      setDescription("Sums up the input value over a timewindow");
      StreamSchema schema = new StreamSchema.Builder()
        .addField("timestamp", GDATFieldType.LONG, GDATSlidingWindowAttribute.INCREASING)
        .addField("intStream", GDATFieldType.INT)
      addJSONInput("intInput", schema);
      addQuery("sumOut", "SELECT timestamp, SUM(intStream) AS sumValue FROM intInput GROUP BY timestamp");

    public void emitData(DataPacket dataPacket) {
      LOG.info("Emitting data to next flowlet");
      // Each data packet is forwarded to the next flowlet

class DataPacket {
    // Using the same data type and variable name as specified in the query output
    long timestamp;
    int sumValue;

Ingesting Data into an AbstractInputFlowlet

In order to ingest data into the flowlet, the AbstractInputFlowlet gives two options. One is a HTTP ingestion endpoint; the other is a TCP endpoint. If you run the Flow in Standalone Mode, the ingestion endpoints are printed out in the log messages on the console (wrapped for formatting):

2014-10-02 16:54:40,401 - INFO  [executor-13:c.c.t.s.f.AbstractInputFlowlet@322]
  - Announced Data Port tcpPort_intInput - 63537
2014-10-02 16:54:40,402 - INFO  [executor-13:c.c.t.s.f.AbstractInputFlowlet@322]
  - Announced Data Port httpPort - 63541

You can ingest data through the HTTP Port using a curl command such as:

curl -v -X POST http://localhost:<port>/v1/tigon/<InputName> -d '{ "data" : [ “12495”, “233“ ] }’

For the example given above, it would then be:

curl -v -X POST http://localhost:63541/v1/tigon/intInput -d '{ "data" : [ “12495”, “233“ ] }’

You can choose to ingest data through either HTTP or TCP endpoints; in the case above, the TCP server is running on 63537. There is one TCP endpoint for each input stream.

If the Flow is running in Distributed Mode on a cluster, you can use the serviceinfo and discover commands to find out the endpoints.

In the above example, if we execute serviceinfo <flow-name> as described in the Distributed Command-Line Intreface, we should see a list of available services:


Now we can discover a specific service’s endpoint (either HTTP or TCP) by executing:

discover <flow-name> httpPort


discover <flow-name> tcpPort_intInput

This is will display the hostname and port on which those services are running.

Optionally, you can provide a runtime arg when you start (--httpPort=1433) to give a port number for the HTTP service. The AbstractInputFlowlet will attempt to start the HTTP server on that port; it will fail if it can’t bind to that port. This option may be useful only in Standalone Mode; in Distributed Mode, you might also need to know the hostname where the service is running.

TigonSQL, The Query Language

TigonSQL refers both to a library (the In-memory Stream Processing engine that can perform filtering, aggregation, and joins of data streams) and the language used by that library.

The Tigon query language, TigonSQL, is a pure stream query language with a SQL-like syntax (being mostly a restriction of SQL).

TigonSQL is presented in the Tigon Architecture Guide, including the basic concepts with examples.

Details of the language, its theory of operation, quick start guide and complete reference can be found in the Tigon SQL User Manual.

For developers who are writing extensions to Tigon SQL, please refer to the Tigon SQL Contributor Manual.

Best Practices for Developing Applications

Initializing Instance Fields

There are three ways to initialize instance fields used in Flowlets:

  1. Using the default constructor;
  2. Using the initialize() method of the Flowlets; and
  3. Using @Property annotations.

To initialize using an Property annotation, simply annotate the field definition with @Property.

The following example demonstrates the convenience of using @Property in a WordFilter flowlet that filters out specific words:

public static class WordFilter extends AbstractFlowlet {

  private OutputEmitter<String> out;

  private final String toFilterOut;

  public CountByField(String toFilterOut) {
    this.toFilterOut = toFilterOut;

  public void process(String word) {
    if (!toFilterOut.equals(word)) {

The Flowlet constructor is called with the parameter when the Flow is configured:

public static class WordCountFlow implements Flow {
  public FlowSpecification configure() {
    return FlowSpecification.Builder.with()
      .setDescription("Flow for counting words")
      .withFlowlets().add(new Tokenizer())
                     .add(new WordsFilter("the"))
                     .add(new WordsCounter())

At run-time, when the Flowlet is started, a value is injected into the toFilterOut field.

Field types that are supported using the @Property annotation are primitives, boxed types (e.g. Integer), String and enum.

Where to Go Next

Now that you’re familiar with the components and concepts of Tigon, take a look at:

  • Examples, with a series of examples demonstrating Tigon.