Streaming Data Summarization (Incremental Aggregation)

Introduction

In the previous tutorial, you looked at the Siddhi real time data summarization capabilities by calculating the total production in the past minute.Now let's consider a more advanced scenario where you need to calculate the total value for a specific time period.

In this scenario, the foreman of the Sweet Factory needs to know the total production of Sherbet Lemon during each hour in November 2017.

It is costly to do this by recalculating the total for each and every event. What you need is a time based aggregation of the events in real time and retrieval on demand. Siddhi supports this functionality through the Incremental Aggregation concept.

Incremental Aggregation calculates the aggregated values continuously and stores them. These values can be retrieved efficiently from the store on demand. Furthermore, Incremental Aggregators support out of order event arrival with in-memory buffers for higher accuracy.

Tutorial steps

Lets get started!

User Scenario 1: Defining incremental aggregation

In this scenario, lets define an incremental aggregation to calculate the total production in an incremental manner, and store the results.

1. Let's define an input stream as follows based on the data received from Sweet Bots. This is the same stream definition used in the previous tutorials to capture the name of the sweet category and the amount produced.

   define stream SweetProductionStream (name string, amount long);

2. Now, let's define an aggregation for the input data. Here, you can assume that the foreman would like to know the production per hour, month and year for each sweet.

   define aggregation SweetProductionAggregation 
   from SweetProductionStream
   select name, sum(amount) as totalAmount
   group by name
   aggregate every hour...year

   This calculates the total amount per hour, day, month and year by the arrival ime of each event. Incremental Aggregation can also be done for seconds, minutes, hours, days, months and years. However, in this sweet production scenario, aggregating by second holds no information value. Therefore, the sweet production is aggregated from hour to year.

3. Now, comes the question of when the production occurs. In the above aggregation, event arrival time is the time used in aggregation. The Sweet Bots send information directly from the factory floor to the server in the same network. Therefore, we can assume that the event arrival time is the production time.

   If you want you can be more accurate by appending the data sent by the SweetBots to include time as shown below.

   1. First define the input stream to include a timestamp:

      define stream SweetProductionStream (name string, amount long, timestamp long);

   2. Then use the timestamp for aggregation as shown below.

      aggregate by SweetProductionStream.timestamp every hour...year

   For this tutorial, let's continue to use the format mentioned first instead of the format in these substeps because the time differences are very slight in the hourly calculations.

   from SweetProductionStream
   select name, sum(amount) as totalAmount
   group by name

   This part of the aggregation specifies the following:

   • From where the information to be processed is taken (i.e., SweetProductionStream input stream)
   • The value you are aggregating. In this scenario, sum(amount) as totalAmount aggregates only the summation of values. The aggregation can also be minimum, maximum or average.
   • The group by clause is optional and can be ignored if all production must be aggregated.

   
   The completed Siddhi application looks as follows.

   • Here, you have defined a stream to get information and aggregated a value for minutes-year.
   • Aggregation can be stored in any type of store supported in Siddhi. For more information about the supported stores, see Defining Tables for Physical Stores.

   define stream SweetProductionStream(name string, amount long);
   
   @store(......)
   define aggregation SweetProductionAggregation 
   from SweetProductionStream
   select name, sum(amount) as totalAmount
   group by name
   aggregate every min...year

User Scenario 2: Retrieval of data on demand

In the previous scenario, you defined the aggregation. Now let's see how to retrieve from it. Siddhi supports this functionality through correlation of data. In this tutorial, you are retrieving data via aggregation joins. For more information on correlating data through joins see Siddhi Query Guide - Joins.

1. First, let's define a stream to retrieve data. The foreman needs to see the hourly production of Sherbet Lemon for November 2016. Therefore, the criteria to retrieve values are as follows.

   Sweet	Sherbet Lemon
   Interval	Hourly
   Duration	November 2017

   Therefore, the input stream needs to be defined as follows:

   define stream GetTotalSweetProductionStream (name string, start_duration string, end_duration string, interval string);

2. A possible output of this retrieval is the timestamp (beginning of each hour), the name of the sweet and the total amount. Therefore, let's define an output stream with these values as follows.

   define stream HourlyProductionStream(AGG_TIMESTAMP long, name string, totalAmount long);

   In the above definition, AGG_TIMESTAMP is the internal reference of the aggregation defining the start of the time interval.

3. Now, let's use the aggregation, retrieval stream, and the output stream to define data correlation from an aggregation.
   Aggregation for the selected period contains aggregation for all sweets. Therefore, let's join the aggregation, and the retrieval stream based on the sweet name to filter aggregations for Sherbet Lemon.

   from GetTotalSweetProductionStream as b join SweetProductionAggregation as a
on a.name == b.name

4. You need to retrieve data relevant only for November 2017. Therefore, let's add it in the retrieval stream as the duration. 

   from GetTotalSweetProductionStream as b join SweetProductionAggregation as a
on a.name == b.name
within b.duration

   In the output event, the duration for which the data is retrieved must be represented in a specific format. For example, November 2017 can be represented as 2017-11-** **:**:**. The supported date formats are <yyyy>-<MM>-<dd> <HH>:<mm>:<ss> (if time is in GMT) and <yyyy>-<MM>-<dd> <HH>:<mm>:<ss> <Z> (if the time is not in GMT), here the ISO 8601 UTC offset must be provided for <Z> (e.g., +05:30, -11:00).

   If the user needs a specific time duration, the query must be changed as follows. Both durations specified must adhere to the data formats required by Siddhi.

   from GetSweetProductionStream as b join SweetProductionAggregation as a
on a.name == b.name
within b.start_duration, b.end_duration

5. Let's add interval for the retrieval to specify for which intervals you want the data to be retrieved.

   from GetTotalSweetProductionStream as b join SweetProductionAggregation as a
on a.name == b.name
within b.duration
per b.interval

   Interval can be in the format of SECONDS, MINUTES, HOURS, DAYS, MONTHS or YEARS ( these values are not case sensitive).

   The completed statement including the output stream looks as follows:

   from GetTotalSweetProductionStream as b join SweetProductionAggregation as a
on a.name == b.name
within b.duration
per b.interval
select a.AGG_TIMESTAMP, a.name, a.totalAmount
insert into HourlyProductionStream;

   In the above definition, a.AGG_TIMESTAMP is the internal data of the aggregation defining the start of the time interval. For instance, in the  November 2017 duration, there is a 24*30 hourly production aggregation. The first output event has the timestamp of the date and time of 1st November 2017 00:00:00.
   
   The completed Siddhi application with the possible sink and source configurations is as follows.

   @App:name('TotalProductionHistoryApp')
   @source(type = 'http', @map(type = 'json'))
   define stream SweetProductionStream(name string, amount long);
   @source(type = 'http', @map(type = 'json'))
   define stream GetTotalSweetProductionStream (name string, duration string, interval string);
   
   
   @sink(type='log', prefix='Hourly Production Stream')
   define stream HourlyProductionStream(AGG_TIMESTAMP long, name string, totalAmount long);
   
   @index('name')
   @store(type='rdbms', jdbc.url="jdbc:mysql://localhost:3306/SweetFactoryDB", username="root", password="root" , jdbc.driver.name="com.mysql.jdbc.Driver")
   define aggregation SweetProductionAggregation 
   from SweetProductionStream
   select name, sum(amount) as totalAmount
   group by name
   aggregate every hour ... year;
   from GetTotalSweetProductionStream as b join SweetProductionAggregation as a
     on a.name == b.name 
     within b.duration
     per b.interval 
   select a.AGG_TIMESTAMP, a.name, a.totalAmount 
   insert into HourlyProductionStream;