HDFS Stateless Sink connector

The HDFS Stateless Sink connector is a Stateless NiFi dataflow developed by Cloudera that is running in the Kafka Connect Framework. Learn about the connector, its properties, and configuration.

The HDFS Stateless Sink Connector fetches messages from Kafka and stores them in HDFS. The topic this connector receives messages from is determined by the value of the topics parameter in the configuration. The messages can contain unstructured data (character or binary) or they can be in Avro or JSON format.

If the input is unstructured data, record processing is disabled. Multiple messages can be concatenated into a single output file on HDFS using a demarcator (for example newline for text messages). Merging is optional, large binary data can be forwarded to HDFS at a 1:1 ratio, one Kafka message equals a single HDFS file. However, in case of large files, this usage is encouraged only to avoid small file problems in HDFS.

If the input is either Avro or JSON, record processing is enabled. In a case like this, the schema of the records can be a predefined schema retrieved from Schema Registry (for both Avro and JSON data), it can be embedded in the Avro data, or inferred from the JSON data. The output can be Avro, JSON, CSV, or Parquet. Multiple records are typically merged into a single output file to store in HDFS.

The connector supports connecting to Kerberos enabled HDFS and Schema Registry services.

Properties and configuration

Configuration is passed to the connector in a JSON file during creation. The properties of the connector can be categorized into three groups. These are as follows:

Common connector properties

These are the properties of the Kafka Connect framework that are accepted by all connectors. For a comprehensive list of these properties, see the Apache Kafka documentation.

Stateless NiFi Sink properties

These are the properties that are specific to the Stateless NiFi Sink connector. All Stateless NiFi Sink connectors share and accept these properties. For a comprehensive list of these properties, see the Stateless NiFi Sink property reference.

Connector/dataflow-specific properties

These properties are unique to this specific connector. Or to be more precise, unique to the dataflow running within the connector. These properties use the following prefix:

parameter.[***CONNECTOR NAME***] Parameters:

For a comprehensive list of these properties, see the HDFS Stateless Sink properties reference.

Notes and limitations

Required properties must be assigned a valid value even if they are not used in the particular configuration. If a required property is not used, either leave its default value, or completely remove the property from the configuration JSON.
If a property that has a default value is completely removed from the configuration JSON, the system uses the default value.
Properties not marked as required must be completely removed from the configuration JSON if not set.
Schema Branch and Schema Version can not be specified at the same time.
If HDFS is running on a different cluster than the connector, the HDFS configuration resource files must be copied to the Kafka Connect node that the connector runs on.

Configuration example for unstructured data

In this example, the connector fetches unstructured messages from Kafka. The connector concatenates the messages using newline characters as the demarcator. The files that the connector creates and stores in HDFS are maximum 10 MB in size. The files are named according to the following pattern: messages_[***UUID***].

{
"connector.class": "org.apache.nifi.kafka.connect.StatelessNiFiSinkConnector",
"meta.smm.predefined.flow.name": "HDFS Sink",
"meta.smm.predefined.flow.version": "1.0.0",
"key.converter": "org.apache.kafka.connect.storage.StringConverter",
"value.converter": "org.apache.kafka.connect.converters.ByteArrayConverter",
"tasks.max": "1",
"nexus.url": "https://repository.cloudera.com/artifactory/repo/",
"working.directory": "/tmp/nifi-stateless-working",
"failure.ports": "",
"topics": "[***TOPIC NAME***]",
"parameter.HDFS Sink Parameters:Hadoop Configuration Resources": "/etc/hadoop/conf/core-site.xml,/etc/hadoop/conf/hdfs-site.xml",
"parameter.HDFS Sink Parameters:Kafka Message Data Format": "Raw",
"parameter.HDFS Sink Parameters:Output Directory Pattern": "[***OUTPUT DIRECTORY PATH***]",
"parameter.HDFS Sink Parameters:Output Filename Pattern": "messages_${filename.uuid}",
"parameter.HDFS Sink Parameters:Maximum File Size": "10 MB",
"parameter.HDFS Sink Parameters:Output File Demarcator": " \n"
}

The following list collects the properties from the configuration example that must be customized for this use case.

topics: The name of the Kafka topic the connector fetches messages from.
Hadoop Configuration Resources: A comma delimited list of Hadoop configuration files (core-site.xml, hdfs-site.xml, and so on). If HDFS is running on a different cluster than the connector, ensure that you copy the configuration files to the Kafka Connect node that the connector runs on.
note
In the majority of cases, these files can be found in/etc/hadoop/conf.
Output Directory Pattern: Specifies the path to the output directory in HDFS.
Maximum File Size: The maximum size of the output data file. In this example, the maximum size is 10 MB.
Kafka Message Data Format: The format of the messages the connector receives from Kafka. In this example, this property is set to Raw. This means that the connector expects unstructured text or binary data and that record processing is disabled.
Output File Demarcator: Specifies the character sequence for demarcating (delimiting) message boundaries when multiple Kafka messages are ingested into an output file as raw messages. In this example, the property is set to \n (newline). This means that the newline character is used to separate the single line character data of the Kafka messages.
Output Filename Pattern: Specifies the structure of the name of the output file. This property accepts string literals (fixed text) as well as various expressions. In this example, the property is set to messages_${filename.uuid}. The word messages and the underscore ( _ ) are string literals. ${filename.uuid} is an expression, which inserts a Universal Unique Identifier (UUID). As a result, the files are named according to the following pattern: messages_[***UUID***].

Configuration example for Avro input and JSON output

In this example, the connector fetches messages in Avro format from Kafka. The connector parses the Avro records and converts them to JSON format using the schema definition stored in Schema Registry. The Avro file does not contain any schema reference. This is the reason why the schema must be specified in the connector parameters. The connector merges multiple records (all messages that are available for a single execution of the connector) into a JSON file, generates a name for the file using the following pattern: messages_[***UUID***].json, and stores the file in HDFS.

{
"connector.class": "org.apache.nifi.kafka.connect.StatelessNiFiSinkConnector",
"meta.smm.predefined.flow.name": "HDFS Sink",
"meta.smm.predefined.flow.version": "1.0.0",
"key.converter": "org.apache.kafka.connect.storage.StringConverter",
"value.converter": "org.apache.kafka.connect.converters.ByteArrayConverter",
"tasks.max": "1",
"nexus.url": "https://repository.cloudera.com/artifactory/repo/",
"working.directory": "/tmp/nifi-stateless-working",
"failure.ports": "",
"topics": "[***TOPIC NAME***]",
"parameter.HDFS Sink Parameters:Hadoop Configuration Resources": "/etc/hadoop/conf/core-site.xml,/etc/hadoop/conf/hdfs-site.xml",
"parameter.HDFS Sink Parameters:Kafka Message Data Format": "Avro",
"parameter.HDFS Sink Parameters:Output Directory Pattern": "[***OUTPUT DIRECTORY NAME***]",
"parameter.HDFS Sink Parameters:Output File Data Format": "JSON",
"parameter.HDFS Sink Parameters:Output Filename Pattern": "messages_${filename.uuid}.json",
"parameter.HDFS Sink Parameters:Schema Access Strategy": "Schema Registry",
"parameter.HDFS Sink Parameters:Schema Name": "[***SCHEMA NAME***]"
"parameter.HDFS Sink Parameters:Schema Registry URL": "[***SCHEMA REGISTRY SERVER URL***]"
}

The following list collects the properties from the configuration example that must be customized for this use case.

topics

The name of the Kafka topic the connector fetches messages from.

Hadoop Configuration Resources

A comma delimited list of Hadoop configuration files (core-site.xml, hdfs-site.xml, and so on). If HDFS is running on a different cluster than the connector, ensure that you copy the configuration files to the Kafka Connect node that the connector runs on.

Output Directory Pattern

Specifies the path to the output directory in HDFS.

Kafka Message Data Format

The format of the messages the connector receives from Kafka. In this example, this property is set to Avro. This means that the connector expects Avro data and that record processing is enabled.

Output File Data Format

Specifies the format of the records written to the output file. In this example, this property is set to JSON, meaning that the output file format is JSON.

Output Filename Pattern

Specifies the structure of the name of the output file. This property accepts string literals (fixed text) as well as various expressions. In this example, the property is set to messages_${filename.uuid}.json. The word messages, the underscore ( _ ), and .json are string literals. ${filename.uuid} is an expression, which inserts a Universal Unique Identifier (UUID). As a result, the files are named according to the following pattern: messages_[***UUID***].json.

Schema Access Strategy

Specifies the strategy used for determining the schema of the Kafka record. In this example, this property is set to Schema Registry, meaning that the schema definition used to parse input data is taken from the Schema Registry.

Schema Name

Specifies the schema name to look up in Schema Registry.

Schema Registry URL

The URL of the Schema Registry server.

Configuration example for JSON input and Parquet output

In this example, the connector fetches messages in JSON format from Kafka. The connector parses the JSON records and converts them to Parquet format using the schema inferred from the JSON data. The connector merges multiple records (all messages that are available for a single execution of the connector) into a Parquet file, generates a name for the file using the following pattern: messages_[***TIMESTAMP***]_[***SEQUENCE***].parquet.gz, and stores the files in HDFS. The output directory path contains a fixed name and a timestamp appended to it: /outputDir/[***TIMESTAMP***]. The connector authenticates to HDFS using Kerberos.

{
"connector.class": "org.apache.nifi.kafka.connect.StatelessNiFiSinkConnector",
"meta.smm.predefined.flow.name": "HDFS Sink",
"meta.smm.predefined.flow.version": "1.0.0",
"key.converter": "org.apache.kafka.connect.storage.StringConverter",
"value.converter": "org.apache.kafka.connect.converters.ByteArrayConverter",
"tasks.max": "1",
"nexus.url": "https://repository.cloudera.com/artifactory/repo/",
"working.directory": "/tmp/nifi-stateless-working",
"failure.ports": "",
"topics": "[***TOPIC NAME***]",
"parameter.HDFS Sink Parameters:Hadoop Configuration Resources": "/etc/hadoop/conf/core-site.xml,/etc/hadoop/conf/hdfs-site.xml",
"parameter.HDFS Sink Parameters:Kafka Message Data Format": "JSON",
"parameter.HDFS Sink Parameters:Output Directory Pattern": "/outputDir/${directory.timestamp}",
"parameter.HDFS Sink Parameters:Output Directory Timestamp Format":"yyMMdd",
"parameter.HDFS Sink Parameters:Output File Data Format": "Parquet",
"parameter.HDFS Sink Parameters:Output Filename Pattern": "messages_${filename.timestamp}_${filename.sequence}.parquet.gz",
"parameter.HDFS Sink Parameters:Output Filename Timestamp Format”: “HHmmss”,
"parameter.HDFS Sink Parameters:Schema Access Strategy": "Infer Schema",
"parameter.HDFS Sink Parameters:Kerberos Keytab for HDFS”:”[***KEYTAB LOCATION***]”,
"parameter.HDFS Sink Parameters:Kerberos Principal for HDFS”:”[***PRINCIPAL NAME***]”,
"parameter.HDFS Sink Parameters:Compression Codec for Parquet”:”GZIP” 
}

The following list collects the properties from the configuration example that must be customized for this use case.

topics: The name of the Kafka topic the connector fetches messages from.
Hadoop Configuration Resources: A comma delimited list of Hadoop configuration files (core-site.xml, hdfs-site.xml, and so on). If HDFS is running on a different cluster than the connector, ensure that you copy the configuration files to the Kafka Connect node that the connector runs on.
note
In the majority of cases, these files can be found in/etc/hadoop/conf.
Output Directory Pattern: Specifies the path to the output directory in HDFS. In this example, the path is set to /outputDir/${directory.timestamp}. As a result, the directory path in this example resolves to /outputDir/[***TIMESTAMP***]. The timestamp format is configured in the Output Directory Timestamp Format property.
Output Directory Timestamp Format: Specifies the timestamp format used in the ${directory.timestamp} expression, which can be used in the Output Directory Pattern property’s value. Directory Time Based Partitioning can be implemented using this expression. For example, if Output Directory Pattern is set to /outputDir/${directory.timestamp} and the timestamp format specified is yyMMdd, the output files will be generated in the /outputDir/220101/ directory if the date is January 1, 2022.
Kafka Message Data Format: The format of the messages the connector receives from Kafka. In this example, this property is set to JSON. This means that the connector expects JSON data and that record processing is enabled.
Output File Data Format: Specifies the format of the records written to the output file. In this example, this property is set to Parquet, meaning that the output file format is Parquet.
Output Filename Pattern: Specifies the structure of the name of the output file. This property accepts string literals (fixed text) as well as various expressions. In this example, the property is set to messages_${filename.timestamp}_${filename.sequence}.parquet.gz.The word messages, the underscores ( _ ), and parquet.gz are string literals. ${filename.timestamp} and ${filename.sequence} are expressions. ${filename.timestamp} inserts the current timestamp in the filename. ${filename.sequence} inserts an incrementing sequence value in the filename. As a result, the files are named according to the following pattern: messages_[***TIMESTAMP***]_[***SEQUENCE***].parquet.gz.
Output Filename Timestamp Format: Specifies the timestamp format used in the ${filename.timestamp} expression. The expression is used when generating the output filename. In this example, the format is HHmmss.
Schema Access Strategy: Specifies the strategy used for determining the schema of the Kafka record. In this example, this property is set to Infer Schema, meaning that the schema is determined (inferred) from the JSON data.
Kerberos Keytab for HDFS: Specifies the keytab used to authenticate to HDFS with Kerberos.
Kerberos Principal for HDFS: Specifies the principal used to authenticate to HDFS with Kerberos.
Compression Codec: Specifies the codec used to compress files and store them in compressed format in HDFS, in the example the GZIP codec is used.