HdfsFile

HDFS file sink connector

Description

Output data to hdfs file

Key features

• exactly-once

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

• schema projection
• file format
  • text
  • csv
  • parquet
  • orc
  • json

Options

In order to use this connector, You must ensure your spark/flink cluster already integrated hadoop. The tested hadoop version is 2.x.

name	type	required	default value
fs.defaultFS	string	yes	-
path	string	yes	-
file_name_expression	string	no	"${transactionId}"
file_format	string	no	"text"
filename_time_format	string	no	"yyyy.MM.dd"
field_delimiter	string	no	'\001'
row_delimiter	string	no	"\n"
partition_by	array	no	-
partition_dir_expression	string	no	"${k0}=${v0}/${k1}=${v1}/.../${kn}=${vn}/"
is_partition_field_write_in_file	boolean	no	false
sink_columns	array	no	When this parameter is empty, all fields are sink columns
is_enable_transaction	boolean	no	true
save_mode	string	no	"error"

fs.defaultFS [string]

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

path [string]

The target dir path is required.

file_name_expression [string]

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.

file_format [string]

We supported as the following file types:

text csv parquet orc json

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

filename_time_format [string]

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

See Java SimpleDateFormat for detailed time format syntax.

field_delimiter [string]

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

row_delimiter [string]

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

partition_by [array]

Partition data based on selected fields

partition_dir_expression [string]

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]

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.

save_mode [string]

Storage mode, currently supports overwrite. This means we will delete the old file when a new file have a same name with it.

If is_enable_transaction is true, Basically, we won't encounter the same file name. Because we will add the transaction id to file name.

For the specific meaning of each mode, see save-modes

Example

For text file format

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

For parquet file format

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

For orc file format

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