Examples

For information about Apache Flink SQL query settings, see Flink on Zeppelin Notebooks for Interactive Data Analysis.

To view your application in the Apache Flink dashboard, choose FLINK JOB in your application's Zeppelin Note page.

For more information about window queries, see Windows in the Apache Flink documentation.

For more examples of Apache Flink Streaming SQL queries, see Queries in the Apache Flink documentation.

Creating tables with Amazon MSK/Apache Kafka

You can use the Amazon MSK Flink connector with Managed Service for Apache Flink Studio to authenticate your connection with Plaintext, SSL, or IAM authentication. Create your tables using the specific properties per your requirements.

-- Plaintext connection

CREATE TABLE your_table (
  `column1` STRING,
  `column2` BIGINT
) WITH (
  'connector' = 'kafka',
  'topic' = 'your_topic',
  'properties.bootstrap.servers' = '<bootstrap servers>',
  'scan.startup.mode' = 'earliest-offset',
  'format' = 'json'
);

-- SSL connection

CREATE TABLE your_table (
  `column1` STRING,
  `column2` BIGINT
) WITH (
  'connector' = 'kafka',
   'topic' = 'your_topic',
  'properties.bootstrap.servers' = '<bootstrap servers>',
  'properties.security.protocol' = 'SSL',
  'properties.ssl.truststore.location' = '/usr/lib/jvm/java-11-amazon-corretto/lib/security/cacerts',
  'properties.ssl.truststore.password' = 'changeit',
  'properties.group.id' = 'myGroup',
  'scan.startup.mode' = 'earliest-offset',
  'format' = 'json'
);

-- IAM connection (or for MSK Serverless)

CREATE TABLE your_table (
  `column1` STRING,
  `column2` BIGINT
) WITH (
  'connector' = 'kafka',
  'topic' = 'your_topic',
  'properties.bootstrap.servers' = '<bootstrap servers>',
  'properties.security.protocol' = 'SASL_SSL',
  'properties.sasl.mechanism' = 'AWS_MSK_IAM',
  'properties.sasl.jaas.config' = 'software.amazon.msk.auth.iam.IAMLoginModule required;',
  'properties.sasl.client.callback.handler.class' = 'software.amazon.msk.auth.iam.IAMClientCallbackHandler',
  'properties.group.id' = 'myGroup',
  'scan.startup.mode' = 'earliest-offset',
  'format' = 'json'
);

You can combine these with other properties at Apache Kafka SQL Connector.

Creating tables with Kinesis

In the following example, you create a table using Kinesis:

CREATE TABLE KinesisTable (
  `column1` BIGINT,
  `column2` BIGINT,
  `column3` BIGINT,
  `column4` STRING,
  `ts` TIMESTAMP(3)
)
PARTITIONED BY (column1, column2)
WITH (
  'connector' = 'kinesis',
  'stream' = 'test_stream',
  'aws.region' = '<region>',
  'scan.stream.initpos' = 'LATEST',
  'format' = 'csv'
);

For more information on other properties you can use, see Amazon Kinesis Data Streams SQL Connector.

Tumbling window

The following Flink Streaming SQL query selects the highest price in each five-second tumbling window from the ZeppelinTopic table:

%flink.ssql(type=update)
SELECT TUMBLE_END(event_time, INTERVAL '5' SECOND) as winend, MAX(price) as five_second_high, ticker
FROM ZeppelinTopic
GROUP BY ticker, TUMBLE(event_time, INTERVAL '5' SECOND)

Sliding window

The following Apache Flink Streaming SQL query selects the highest price in each five-second sliding window from the ZeppelinTopic table:

%flink.ssql(type=update)
SELECT HOP_END(event_time, INTERVAL '3' SECOND, INTERVAL '5' SECOND) AS winend, MAX(price) AS sliding_five_second_max
FROM ZeppelinTopic//or your table name in AWS Glue
GROUP BY HOP(event_time, INTERVAL '3' SECOND, INTERVAL '5' SECOND)

Interactive SQL

This example prints the max of event time and processing time and the sum of values from the key-values table. Ensure that you have the sample data generation script from the Data generator running. To try other SQL queries such as filtering and joins in your Studio notebook, see the Apache Flink documentation: Queries in the Apache Flink documentation.

%flink.ssql(type=single, parallelism=4, refreshInterval=1000, template=<h1>{2}</h1> records seen until <h1>Processing Time: {1}</h1> and <h1>Event Time: {0}</h1>)

-- An interactive query prints how many records from the `key-value-stream` we have seen so far, along with the current processing and event time.
SELECT
  MAX(`et`) as `et`,
  MAX(`pt`) as `pt`,
  SUM(`value`) as `sum`
FROM
  `key-values`

%flink.ssql(type=update, parallelism=4, refreshInterval=1000)

-- An interactive tumbling window query that displays the number of records observed per (event time) second.
-- Browse through the chart views to see different visualizations of the streaming result.
SELECT
  TUMBLE_START(`et`, INTERVAL '1' SECONDS) as `window`,
  `key`,
  SUM(`value`) as `sum`
FROM
  `key-values`
GROUP BY
  TUMBLE(`et`, INTERVAL '1' SECONDS),
  `key`;

BlackHole SQL connector

The BlackHole SQL connector doesn't require that you create a Kinesis data stream or an Amazon MSK cluster to test your queries. For information about the BlackHole SQL connector, see BlackHole SQL Connector in the Apache Flink documentation. In this example, the default catalog is an in-memory catalog.

%flink.ssql

CREATE TABLE default_catalog.default_database.blackhole_table (
 `key` BIGINT,
 `value` BIGINT,
 `et` TIMESTAMP(3)
) WITH (
 'connector' = 'blackhole'
)

%flink.ssql(parallelism=1)

INSERT INTO `test-target`
SELECT
  `key`,
  `value`,
  `et`
FROM
  `test-source`
WHERE
  `key` > 3

%flink.ssql(parallelism=2)

INSERT INTO `default_catalog`.`default_database`.`blackhole_table`
SELECT
  `key`,
  `value`,
  `et`
FROM
  `test-target`
WHERE
  `key` > 7

Data generator

This example uses Scala to generate sample data. You can use this sample data to test various queries. Use the create table statement to create the key-values table.

import org.apache.flink.streaming.api.functions.source.datagen.DataGeneratorSource
import org.apache.flink.streaming.api.functions.source.datagen.RandomGenerator
import org.apache.flink.streaming.api.scala.DataStream

import java.sql.Timestamp

// ad-hoc convenience methods to be defined on Table 
implicit class TableOps[T](table: DataStream[T]) {
    def asView(name: String): DataStream[T] = {
      if (stenv.listTemporaryViews.contains(name)) {
        stenv.dropTemporaryView("`" + name + "`")
      }
      stenv.createTemporaryView("`" + name + "`", table)
      return table;
    }
}

%flink(parallelism=4)
val stream = senv
 .addSource(new DataGeneratorSource(RandomGenerator.intGenerator(1, 10), 1000))
 .map(key => (key, 1, new Timestamp(System.currentTimeMillis)))
 .asView("key-values-data-generator")

%flink.ssql(parallelism=4)
-- no need to define the paragraph type with explicit parallelism (such as "%flink.ssql(parallelism=2)")
-- in this case the INSERT query will inherit the parallelism of the of the above paragraph
INSERT INTO `key-values`
SELECT
 `_1` as `key`,
 `_2` as `value`,
 `_3` as `et`
FROM
 `key-values-data-generator`

Interactive Scala

This is the Scala translation of the Interactive SQL. For more Scala examples, see Table API in the Apache Flink documentation.

%flink
import org.apache.flink.api.scala._
import org.apache.flink.table.api._
import org.apache.flink.table.api.bridge.scala._

// ad-hoc convenience methods to be defined on Table
implicit class TableOps(table: Table) {
    def asView(name: String): Table = {
      if (stenv.listTemporaryViews.contains(name)) {
        stenv.dropTemporaryView(name)
      }
      stenv.createTemporaryView(name, table)
      return table;
    }
}

%flink(parallelism=4)

// A view that computes many records from the `key-values` we have seen so far, along with the current processing and event time.
val query01 = stenv
  .from("`key-values`")
  .select(
    $"et".max().as("et"),
    $"pt".max().as("pt"),
    $"value".sum().as("sum")
  ).asView("query01")

%flink.ssql(type=single, parallelism=16, refreshInterval=1000, template=<h1>{2}</h1> records seen until <h1>Processing Time: {1}</h1> and <h1>Event Time: {0}</h1>)

-- An interactive query prints the query01 output.
SELECT * FROM query01

%flink(parallelism=4)

// An tumbling window view that displays the number of records observed per (event time) second.
val query02 = stenv
  .from("`key-values`")
  .window(Tumble over 1.seconds on $"et" as $"w")
  .groupBy($"w", $"key")
  .select(
    $"w".start.as("window"),
    $"key",
    $"value".sum().as("sum")
  ).asView("query02")

%flink.ssql(type=update, parallelism=4, refreshInterval=1000)

-- An interactive query prints the query02 output.
-- Browse through the chart views to see different visualizations of the streaming result.
SELECT * FROM `query02`

Interactive Python

This is the Python translation of the Interactive SQL. For more Python examples, see Table API in the Apache Flink documentation.

%flink.pyflink
from pyflink.table.table import Table

def as_view(table, name):
  if (name in st_env.list_temporary_views()):
    st_env.drop_temporary_view(name)
  st_env.create_temporary_view(name, table)
  return table

Table.as_view = as_view

%flink.pyflink(parallelism=16)

# A view that computes many records from the `key-values` we have seen so far, along with the current processing and event time
st_env \
  .from_path("`keyvalues`") \
  .select(", ".join([
    "max(et) as et",
    "max(pt) as pt",
    "sum(value) as sum"
  ])) \
  .as_view("query01")

%flink.ssql(type=single, parallelism=16, refreshInterval=1000, template=<h1>{2}</h1> records seen until <h1>Processing Time: {1}</h1> and <h1>Event Time: {0}</h1>)

-- An interactive query prints the query01 output.
SELECT * FROM query01

%flink.pyflink(parallelism=16)

# A view that computes many records from the `key-values` we have seen so far, along with the current processing and event time
st_env \
  .from_path("`key-values`") \
  .window(Tumble.over("1.seconds").on("et").alias("w")) \
  .group_by("w, key") \
  .select(", ".join([
    "w.start as window",
    "key",
    "sum(value) as sum"
  ])) \
  .as_view("query02")

%flink.ssql(type=update, parallelism=16, refreshInterval=1000)

-- An interactive query prints the query02 output.
-- Browse through the chart views to see different visualizations of the streaming result.
SELECT * FROM `query02`

Interactive Python, SQL, and Scala

You can use any combination of SQL, Python, and Scala in your notebook for interactive analysis. In a Studio notebook that you plan to deploy as an application with durable state, you can use a combination of SQL and Scala. This example shows you the sections that are ignored and those that get deployed in the application with durable state.

%flink.ssql
CREATE TABLE `default_catalog`.`default_database`.`my-test-source` (
  `key` BIGINT NOT NULL,
  `value` BIGINT NOT NULL,
  `et` TIMESTAMP(3) NOT NULL,
  `pt` AS PROCTIME(),
  WATERMARK FOR `et` AS `et` - INTERVAL '5' SECOND
)
WITH (
  'connector' = 'kinesis',
  'stream' = 'kda-notebook-example-test-source-stream',
  'aws.region' = 'eu-west-1',
  'scan.stream.initpos' = 'LATEST',
  'format' = 'json',
  'json.timestamp-format.standard' = 'ISO-8601'
)

%flink.ssql
CREATE TABLE `default_catalog`.`default_database`.`my-test-target` (
  `key` BIGINT NOT NULL,
  `value` BIGINT NOT NULL,
  `et` TIMESTAMP(3) NOT NULL,
  `pt` AS PROCTIME(),
  WATERMARK FOR `et` AS `et` - INTERVAL '5' SECOND
)
WITH (
  'connector' = 'kinesis',
  'stream' = 'kda-notebook-example-test-target-stream',
  'aws.region' = 'eu-west-1',
  'scan.stream.initpos' = 'LATEST',
  'format' = 'json',
  'json.timestamp-format.standard' = 'ISO-8601'
)

%flink()

// ad-hoc convenience methods to be defined on Table
implicit class TableOps(table: Table) {
  def asView(name: String): Table = {
    if (stenv.listTemporaryViews.contains(name)) {
      stenv.dropTemporaryView(name)
    }
    stenv.createTemporaryView(name, table)
    return table;
  }
}

%flink(parallelism=1)
val table = stenv
  .from("`default_catalog`.`default_database`.`my-test-source`")
  .select($"key", $"value", $"et")
  .filter($"key" > 10)
  .asView("query01")

%flink.ssql(parallelism=1)

-- forward data
INSERT INTO `default_catalog`.`default_database`.`my-test-target`
SELECT * FROM `query01`

%flink.ssql(type=update, parallelism=1, refreshInterval=1000)

-- forward data to local stream (ignored when deployed as application)
SELECT * FROM `query01`

%flink

// tell me the meaning of life (ignored when deployed as application!)
print("42!")

Cross-account Kinesis data stream

To use a Kinesis data stream that's in an account other than the account that has your Studio notebook, create a service execution role in the account where your Studio notebook is running and a role trust policy in the account that has the data stream. Use aws.credentials.provider, aws.credentials.role.arn, and aws.credentials.role.sessionName in the Kinesis connector in your create table DDL statement to create a table against the data stream.

Use the following service execution role for the Studio notebook account.

{
 "Sid": "AllowNotebookToAssumeRole",
 "Effect": "Allow",
 "Action": "sts:AssumeRole"
 "Resource": "*"
}

Use the AmazonKinesisFullAccess policy and the following role trust policy for the data stream account.

{
 "Version": "2012-10-17",
 "Statement": [
 {
 "Effect": "Allow",
 "Principal": {
 "AWS": "arn:aws:iam::<accountID>:root"
 },
 "Action": "sts:AssumeRole",
 "Condition": {}
 }
 ]
}

Use the following paragraph for the create table statement.

%flink.ssql
CREATE TABLE test1 (
name VARCHAR,
age BIGINT
) WITH (
'connector' = 'kinesis',
'stream' = 'stream-assume-role-test',
'aws.region' = 'us-east-1',
'aws.credentials.provider' = 'ASSUME_ROLE',
'aws.credentials.role.arn' = 'arn:aws:iam::<accountID>:role/stream-assume-role-test-role',
'aws.credentials.role.sessionName' = 'stream-assume-role-test-session',
'scan.stream.initpos' = 'TRIM_HORIZON',
'format' = 'json'
)