HdfsFile

HDFS file sink connector

Description

Output data to hdfs file

提示

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.

Key features

• exactly-once

By default, we use 2PC commit to ensure exactly-once

• file format type
  • text
  • csv
  • parquet
  • orc
  • json
• compress codec
  • lzo

Options

name	type	required	default value	remarks
fs.defaultFS	string	yes	-
path	string	yes	-
hdfs_site_path	string	no	-
custom_filename	boolean	no	false	Whether you need custom the filename
file_name_expression	string	no	"${transactionId}"	Only used when custom_filename is true
filename_time_format	string	no	"yyyy.MM.dd"	Only used when custom_filename is true
file_format_type	string	no	"csv"
field_delimiter	string	no	'\001'	Only used when file_format is text
row_delimiter	string	no	"\n"	Only used when file_format is text
have_partition	boolean	no	false	Whether you need processing partitions.
partition_by	array	no	-	Only used then have_partition is true
partition_dir_expression	string	no	"${k0}=${v0}/${k1}=${v1}/.../${kn}=${vn}/"	Only used then have_partition is true
is_partition_field_write_in_file	boolean	no	false	Only used then have_partition is true
sink_columns	array	no		When this parameter is empty, all fields are sink columns
is_enable_transaction	boolean	no	true
batch_size	int	no	1000000
compress_codec	string	no	none
kerberos_principal	string	no	-
kerberos_keytab_path	string	no	-
compress_codec	string	no	none
common-options	object	no	-

fs.defaultFS [string]

The hadoop cluster address that start with hdfs://, for example: hdfs://hadoopcluster

path [string]

The target dir path is required.

hdfs_site_path [string]

The path of hdfs-site.xml, used to load ha configuration of namenodes

custom_filename [boolean]

Whether custom the filename

file_name_expression [string]

Only used when custom_filename is true

file_name_expression describes the file expression which will be created into the path. We can add the variable ${now} or ${uuid} in the file_name_expression, like test_${uuid}_${now}, ${now} represents the current time, and its format can be defined by specifying the option filename_time_format.

Please note that, If is_enable_transaction is true, we will auto add ${transactionId}_ in the head of the file.

filename_time_format [string]

Only used when custom_filename is true

When the format in the file_name_expression parameter is xxxx-${now} , filename_time_format can specify the time format of the path, and the default value is yyyy.MM.dd . The commonly used time formats are listed as follows:

Symbol	Description
y	Year
M	Month
d	Day of month
H	Hour in day (0-23)
m	Minute in hour
s	Second in minute

file_format_type [string]

We supported as the following file types:

text json csv orc parquet

Please note that, The final file name will end with the file_format's suffix, the suffix of the text file is txt.

field_delimiter [string]

The separator between columns in a row of data. Only needed by text file format.

row_delimiter [string]

The separator between rows in a file. Only needed by text file format.

have_partition [boolean]

Whether you need processing partitions.

partition_by [array]

Only used when have_partition is true.

Partition data based on selected fields.

partition_dir_expression [string]

Only used when have_partition is true.

If the partition_by is specified, we will generate the corresponding partition directory based on the partition information, and the final file will be placed in the partition directory.

Default partition_dir_expression is ${k0}=${v0}/${k1}=${v1}/.../${kn}=${vn}/. k0 is the first partition field and v0 is the value of the first partition field.

is_partition_field_write_in_file [boolean]

Only used when have_partition is true.

If is_partition_field_write_in_file is true, the partition field and the value of it will be write into data file.

For example, if you want to write a Hive Data File, Its value should be false.

sink_columns [array]

Which columns need be write to file, default value is all of the columns get from Transform or Source. The order of the fields determines the order in which the file is actually written.

is_enable_transaction [boolean]

If is_enable_transaction is true, we will ensure that data will not be lost or duplicated when it is written to the target directory.

Please note that, If is_enable_transaction is true, we will auto add ${transactionId}_ in the head of the file.

Only support true now.

batch_size [int]

The maximum number of rows in a file. For SeaTunnel Engine, the number of lines in the file is determined by batch_size and checkpoint.interval jointly decide. If the value of checkpoint.interval is large enough, sink writer will write rows in a file until the rows in the file larger than batch_size. If checkpoint.interval is small, the sink writer will create a new file when a new checkpoint trigger.

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: lzo snappy lz4 zlib none
• parquet: lzo snappy lz4 gzip brotli zstd none

kerberos_principal [string]

The principal of kerberos

kerberos_keytab_path [string]

The keytab path of kerberos

common options

Sink plugin common parameters, please refer to Sink Common Options for details

Example

For orc file format simple config

HdfsFile {
    fs.defaultFS = "hdfs://hadoopcluster"
    path = "/tmp/hive/warehouse/test2"
    file_format = "orc"
}

For text file format with have_partition and custom_filename and sink_columns

HdfsFile {
    fs.defaultFS = "hdfs://hadoopcluster"
    path = "/tmp/hive/warehouse/test2"
    file_format_type = "text"
    field_delimiter = "\t"
    row_delimiter = "\n"
    have_partition = true
    partition_by = ["age"]
    partition_dir_expression = "${k0}=${v0}"
    is_partition_field_write_in_file = true
    custom_filename = true
    file_name_expression = "${transactionId}_${now}"
    filename_time_format = "yyyy.MM.dd"
    sink_columns = ["name","age"]
    is_enable_transaction = true
}

For parquet file format with have_partition and custom_filename and sink_columns

HdfsFile {
    fs.defaultFS = "hdfs://hadoopcluster"
    path = "/tmp/hive/warehouse/test2"
    have_partition = true
    partition_by = ["age"]
    partition_dir_expression = "${k0}=${v0}"
    is_partition_field_write_in_file = true
    custom_filename = true
    file_name_expression = "${transactionId}_${now}"
    filename_time_format = "yyyy.MM.dd"
    file_format_type = "parquet"
    sink_columns = ["name","age"]
    is_enable_transaction = true
}

Changelog

2.2.0-beta 2022-09-26

• Add HDFS File Sink Connector

2.3.0-beta 2022-10-20

• [BugFix] Fix the bug of incorrect path in windows environment (2980)
• [BugFix] Fix filesystem get error (3117)
• [BugFix] Solved the bug of can not parse '\t' as delimiter from config file (3083)

2.3.0 2022-12-30

• [BugFix] Fixed the following bugs that failed to write data to files (3258)
  • When field from upstream is null it will throw NullPointerException
  • Sink columns mapping failed
  • When restore writer from states getting transaction directly failed

Next version

• [Improve] Support setting batch size for every file (3625)
• [Improve] Support lzo compression for text in file format (3782)
• [Improve] Support kerberos authentication (3840)
• [Improve] Support file compress (3899)