Hdfs File Source Connector

Support Those Engines

Spark
Flink
SeaTunnel Zeta

Key Features

Read all the data in a split in a pollNext call. What splits are read will be saved in snapshot.

Description

Read data from hdfs file system.

Supported DataSource Info

Datasource Supported Versions
HdfsFile hadoop 2.x and 3.x

Source Options

Name Type Required Default Description
path string yes - The source file path.
file_format_type string yes - We supported as the following file types:text json csv orc parquet excel xml.Please note that, The final file name will end with the file_format’s suffix, the suffix of the text file is txt.
fs.defaultFS string yes - The hadoop cluster address that start with hdfs://, for example: hdfs://hadoopcluster
read_columns list yes - The read column list of the data source, user can use it to implement field projection.The file type supported column projection as the following shown:[text,json,csv,orc,parquet,excel,xml].Tips: If the user wants to use this feature when reading text json csv files, the schema option must be configured.
hdfs_site_path string no - The path of hdfs-site.xml, used to load ha configuration of namenodes
delimiter/field_delimiter string no \001 Field delimiter, used to tell connector how to slice and dice fields when reading text files. default \001, the same as hive’s default delimiter
parse_partition_from_path boolean no true Control whether parse the partition keys and values from file path. For example if you read a file from path hdfs://hadoop-cluster/tmp/seatunnel/parquet/name=tyrantlucifer/age=26. Every record data from file will be added these two fields:[name:tyrantlucifer,age:26].Tips:Do not define partition fields in schema option.
date_format string no yyyy-MM-dd Date type format, used to tell connector how to convert string to date, supported as the following formats:yyyy-MM-dd yyyy.MM.dd yyyy/MM/dd default yyyy-MM-dd.Date type format, used to tell connector how to convert string to date, supported as the following formats:yyyy-MM-dd yyyy.MM.dd yyyy/MM/dd default yyyy-MM-dd
datetime_format string no yyyy-MM-dd HH:mm:ss Datetime type format, used to tell connector how to convert string to datetime, supported as the following formats:yyyy-MM-dd HH:mm:ss yyyy.MM.dd HH:mm:ss yyyy/MM/dd HH:mm:ss yyyyMMddHHmmss .default yyyy-MM-dd HH:mm:ss
time_format string no HH:mm:ss Time type format, used to tell connector how to convert string to time, supported as the following formats:HH:mm:ss HH:mm:ss.SSS.default HH:mm:ss
remote_user string no - The login user used to connect to hadoop login name. It is intended to be used for remote users in RPC, it won’t have any credentials.
krb5_path string no /etc/krb5.conf The krb5 path of kerberos
kerberos_principal string no - The principal of kerberos
kerberos_keytab_path string no - The keytab path of kerberos
skip_header_row_number long no 0 Skip the first few lines, but only for the txt and csv.For example, set like following:skip_header_row_number = 2.then Seatunnel will skip the first 2 lines from source files
schema config no - the schema fields of upstream data
sheet_name string no - Reader the sheet of the workbook,Only used when file_format is excel.
xml_row_tag string no - Specifies the tag name of the data rows within the XML file, only used when file_format is xml.
xml_use_attr_format boolean no - Specifies whether to process data using the tag attribute format, only used when file_format is xml.
compress_codec string no none The compress codec of files
encoding string no UTF-8
common-options no - Source plugin common parameters, please refer to Source Common Options for details.

delimiter/field_delimiter [string]

delimiter parameter will deprecate after version 2.3.5, please use field_delimiter instead.

compress_codec [string]

The compress codec of files and the details that supported as the following shown:

  • txt: lzo none
  • json: lzo none
  • csv: lzo none
  • orc/parquet:
    automatically recognizes the compression type, no additional settings required.

encoding [string]

Only used when file_format_type is json,text,csv,xml. The encoding of the file to read. This param will be parsed by Charset.forName(encoding).

Tips

If you use spark/flink, In order to use this connector, You must ensure your spark/flink cluster already integrated hadoop. The tested hadoop version is 2.x. If you use SeaTunnel Engine, It automatically integrated the hadoop jar when you download and install SeaTunnel Engine. You can check the jar package under ${SEATUNNEL_HOME}/lib to confirm this.

Task Example

Simple:

This example defines a SeaTunnel synchronization task that read data from Hdfs and sends it to Hdfs.

  1. # Defining the runtime environment
  2. env {
  3. parallelism = 1
  4. job.mode = "BATCH"
  5. }
  6. source {
  7. HdfsFile {
  8. schema {
  9. fields {
  10. name = string
  11. age = int
  12. }
  13. }
  14. path = "/apps/hive/demo/student"
  15. type = "json"
  16. fs.defaultFS = "hdfs://namenode001"
  17. }
  18. # If you would like to get more information about how to configure seatunnel and see full list of source plugins,
  19. # please go to https://seatunnel.apache.org/docs/category/source-v2
  20. }
  21. transform {
  22. # If you would like to get more information about how to configure seatunnel and see full list of transform plugins,
  23. # please go to https://seatunnel.apache.org/docs/category/transform-v2
  24. }
  25. sink {
  26. HdfsFile {
  27. fs.defaultFS = "hdfs://hadoopcluster"
  28. path = "/tmp/hive/warehouse/test2"
  29. file_format = "orc"
  30. }
  31. # If you would like to get more information about how to configure seatunnel and see full list of sink plugins,
  32. # please go to https://seatunnel.apache.org/docs/category/sink-v2
  33. }