Restructure Kafka connector output files

Data streamed by a Kafka Connector will be converted to a RADAR-base oriented output directory, by organizing it by project, user and collection date. It supports data written by RADAR S3 sink connector is streamed to files based on topic name only. This package transforms that output to a local directory structure as follows: projectId/userId/topic/date_hour.csv. The date and hour are extracted from the time field of each record, and is formatted in UTC time. This package is included in the RADAR-Docker repository, in the dcompose/radar-cp-hadoop-stack/bin/hdfs-restructure script.

Upgrade instructions

Since version 2.0.0, HDFS is no longer supported, only AWS S3 or Azure Blob Storage, and local file system compatible. If HDFS is still needed, please implement a HDFS source storage factory with constructor org.radarbase.output.source.HdfsSourceStorageFactory(resourceConfig: ResourceConfig, tempPath: Path) with method createSourceStorage(): SourceStorage. This implementation may be added as a separate JAR in the lib/radar-output-plugins/ directory of where the distribution is installed.

When upgrading to version 1.2.0, please follow the following instructions:

• When using local target storage, ensure that:
  1. it is writable by the user 101, or change the runtime user using the docker command-line flag --user to a user that can write to the target storage and
  2. local storage properties userId and groupId are set to values that can write to the target storage.

When upgrading to version 1.0.0 or later from version 0.6.0 please follow the following instructions:

• This package now relies on Redis for locking and offset management. Please install Redis or use the docker-compose.yml file to start it.
• Write configuration file restructure.yml to match settings used with 0.6.0

  • HDFS settings have moved to source. Specify all name nodes in the nameNodes property. The name property is no longer used.

    source:
      type: hdfs
      hdfs:
        nameNodes: [hdfs-namenode]

  • Add a redis block:

    redis:
      uri: redis://localhost:6379

  • Offset accounting will automatically be migrated from a file-based storage to a Redis entry as radar-output processes the topic. Please do not remove the offsets directory until it is empty.

  • storage settings have moved to the target block. Using local output directory:

    target:
      type: local
      local:
        # User ID to write data as. This only works when explicitly setting
        # the runtime user to root.
        userId: 123
        # Group ID to write data as. This only works when explicitly setting
        # the runtime user to root.
        groupId: 123

    With the S3StorageDriver, use the following configuration instead:

    target:
      type: s3
      s3:
        endpoint: https://my-region.s3.aws.amazon.com  # or http://localhost:9000 for local minio
        accessToken: ABA...
        secretKey: CSD...
        bucket: myBucketName

When upgrading to version 0.6.0 from version 0.5.x or earlier, please follow the following instructions:

• Write configuration file restructure.yml to match command-line settings used with 0.5.x.
• If needed, move all entries of offsets.csv to their per-topic file in offsets/<topic>.csv. First go to the output directory, then run the bin/migrate-offsets-to-0.6.0.sh script.

Docker usage

This package is available as docker image radarbase/radar-output-restructure. The entrypoint of the image is the current application. So in all the commands listed in usage, replace radar-output-restructure with for example:

docker run --rm -t --network s3 -v "$PWD/output:/output" radarbase/radar-output-restructure:2.3.2 -o /output /myTopic

Command line usage

To display the usage and all available options you can use the help option as follows:

radar-output-restructure --help

Note that the options preceded by the * in the above output are required to run the app. Also note that there can be multiple input paths from which to read the files. Eg - /topicAndroidNew/topic1 /topicAndroidNew/topic2 .... Provide at least one input path.

Each argument, as well as much more, can be supplied in a config file. The default name of the config file is restructure.yml. Please refer to restructure.yml in the current directory for all available options. An alternative file can be specified with the -F flag.

File Format

By default, this will output the data in CSV format. If JSON format is preferred, use the following instead:

radar-output-restructure --format json --output-directory <output_folder>  <input_path_1> [<input_path_2> ...]

By default, files records are not deduplicated after writing. To enable this behaviour, specify the option --deduplicate or -d. This set to false by default because of an issue with Biovotion data. Please see - issue #16 before enabling it. Deduplication can also be enabled or disabled per topic using the config file. If lines should be deduplicated using a subset of fields, e.g. only sourceId and time define a unique record and only the last record with duplicate values should be kept, then specify topics: <topicName>: deduplication: distinctFields: [key.sourceId, value.time].

Compression

Another option is to output the data in compressed form. All files will get the gz suffix, and can be decompressed with a GZIP decoder. Note that for a very small number of records, this may actually increase the file size. Zip compression is also available.

radar-output-restructure --compression gzip --output-directory <output_folder> <input_path_1> [<input_path_2> ...]

Redis

This package assumes a Redis service running. See the example restructure.yml for configuration options.

Source and target

The source and target properties contain resource descriptions. The source can have two types, azure and s3:

source:
  type: s3  # azure or s3
  s3:
    endpoint: http://localhost:9000  # using AWS S3 endpoint is also possible.
    bucket: radar
    accessToken: minioadmin
    secretKey: minioadmin
  # only actually needed if source type is hdfs
  azure:
    # azure options

The target is similar, and in addition supports the local file system (local).

target:
  type: s3  # s3, local or azure
  s3:
    endpoint: http://localhost:9000
    bucket: out
    accessToken: minioadmin
    secretKey: minioadmin
  # only actually needed if target type is local
  local:
    userId: 1000  # write as regular user, use -1 to use current user (default).
    groupId: 100  # write as regular group, use -1 to use current user (default).

Secrets can be provided as environment variables as well:

Environment variable	Corresponding value
SOURCE_S3_ACCESS_TOKEN	source.s3.accessToken
SOURCE_S3_SECRET_KEY	source.s3.secretKey
SOURCE_AZURE_USERNAME	source.azure.username
SOURCE_AZURE_PASSWORD	source.azure.password
SOURCE_AZURE_ACCOUNT_NAME	source.azure.accountName
SOURCE_AZURE_ACCOUNT_KEY	source.azure.accountKey
SOURCE_AZURE_SAS_TOKEN	source.azure.sasToken
REDIS_URL	redis.url

Replace SOURCE with TARGET in the variables above to configure the target storage.

Path format

The output path at the target storage is determined by the path format. The class that handles path output by default is the org.radarbase.output.path.FormattedPathFactory. The default format is

${projectId}/${userId}/${topic}/${filename}

Each format parameter is enclosed by a dollar sign with curly brackets.

The full set of parameters is listed here:

paths:
  # Input directories in source storage
  inputs:
    - /testIn
  # Temporary directory for local file processing.
  temp: ./output/+tmp
  # Output directory in target storage
  output: /output
  # Output path construction factory
  factory: org.radarbase.output.path.FormattedPathFactory
  # Additional properties
  # properties:
  #   format: ${projectId}/${userId}/${topic}/${time:mm}/${time:YYYYmmDD_HH'00'}${attempt}${extension}
  #   plugins: fixed time key value org.example.plugin.MyPathPlugin

The FormattedPathFactory can use multiple plugins to format paths based on a given record. The fixed plugin has a number of fixed parameters that can be used:

Parameter	Description
projectId	record project ID
userId	record user ID
sourceId	record source ID
topic	Kafka topic
filename	default time binning with attempt suffix and file extension
attempt	attempt suffix for if a file with an incompatible format already exists
extension	file extension

At least filename should be used, or a combination of attempt and extension.

Then there are also plugins that take their own format. The time plugin formats a parameter according to the record time. It takes parameters with format time:<date format> where <date format> should be replaced by a Java date format, such as YYYY-mm-dd. The plugin tries to use the following time fields, in this order: a double time in the value struct, timeStart double or start long in the key struct, dateTime string in the value struct, date string in the value struct, timeReceived double in the value struct or timeCompleted double in the value struct. The first valid value used. If no valid time values are found, unknown-date is returned.

The key and value plugins read values from the key or value structs of a given record. For example, parameter value:color.red will attempt to read the value struct, finding first the color field and then the enclosed red field. If no such value exists, unknown-value will be used in the format.

Cleaner

Source files can be automatically be removed by a cleaner process. This checks whether the file has already been extracted and is older than a configured age. This feature is not enabled by default. It can be configured in the cleaner configuration section:

cleaner:
  # Enable cleaning up old source files
  enable: true
  # Interval in seconds to clean data
  interval: 1260  # 21 minutes
  # Number of days after which a source file is considered old
  age: 7

The cleaner can also be enabled with the --cleaner command-line flag. To run the cleaner as a separate process from output restructuring, start a process that has configuration property worker: enable: false or command-line argument --no-restructure.

Service

To run the output generator as a service that will regularly poll the source directory, add the --service flag and optionally the --interval flag to adjust the polling interval or use the corresponding configuration file parameters.

Local build

This package requires at least Java JDK 8. Build the distribution with

./gradlew build

and install the package into /usr/local with for example

sudo mkdir -p /usr/local
sudo tar -xzf build/distributions/radar-output-restructure-2.3.2.tar.gz -C /usr/local --strip-components=1

Now the radar-output-restructure command should be available.

Extending the connector

To implement alternative storage paths, storage drivers or storage formats, put your custom JAR in $APP_DIR/lib/radar-output-plugins. To load them, use the following options:

Parameter	Base class	Behaviour	Default
paths: factory: ...	org.radarbase.output.path.RecordPathFactory	Factory to create output path names with.	FormattedPathFactory
format: factory: ...	org.radarbase.output.format.FormatFactory	Factory for output formats.	FormatFactory
compression: factory: ...	org.radarbase.output.compression.CompressionFactory	Factory class to use for data compression.	CompressionFactory

The respective <type>: properties: {} configuration parameters can be used to provide custom configuration of the factory. This configuration will be passed to the Plugin#init(Map<String, String>) method.

By adding additional path format plugins to the classpath, the path format of FormattedPathFactory may be expanded with different parameters or lookup engines.