Skip to content

treasure-data/embulk-filter-add_time

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

63 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Add time filter plugin for Embulk

The embulk-filter-add_time plugin allows users to add a new time-based column to the existing schema either by copying the value from another existing column in the schema or by specifying the value. The latter use case is particularly useful to tag all events bulk loaded with one Embulk run with a specific timestamp or time range.

Overview

  • Plugin type: filter

Configuration

to_column configuration

This configuration is required.

The to_column configuration specifies the name, type, and optional format of a new column to be added to the Embulk schema.

This configuration specifies the name, type, and format for the added time column as a set of key-value pairs:

  • name
    a string specifying the name of the new column (required).
  • type
    a string specifying the data type of the column: long or timestamp are the supported values for this parameter since they need to express a valid timestamp (required).
  • unix_timestamp_unit
    a string specifying the unit for the unix timestamp - it is required if the column type is long. sec, milli (for milliseconds), micro (for microseconds), or nano (for nanoseconds) are the supported values for this option (default is sec).

For example, the to_column configuration can be used to add a 'time' column to the schema:

filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
...

Note that if a column with the same name as name already exists in the schema, the existing column name is changed by appending '_' to its name. For example, if 'created_at' is specified as to_column but the column name already exists, the existing column name is changed to 'created_at_'.

When long type is specified for the type parameter, unix_timestamp_unit is required to convert from the unit timestamp value to long and tell the plugin how to exactly parse the timestamp number; sec for seconds, milli for milliseconds, micro for microseconds, or nano for nanoseconds are the supported values for this option:

filters:
- type: add_time
  to_column:
    name: time
    type: long
    unix_timestamp_unit: sec
...

from_value configuration

This configuration is mutually exclusive with from_column below. One amongst from_column and from_value configurations is required.

The from_value configuration specifies a specific value or range to be used as value of the column added by the to_column configuration. This configuration supports different modes and the behavior will vary accordingly. Refer to the examples below.

This configuration specifies the mode, the timestamp_format, and one amongst value, from and to, or nothing depending on the mode of choice as a set of key-value pairs. These parameters are required:

  • mode
    a string specifying the mode. There are 3 valid modes (default is fixed_time):
    • fixed_time
      In this mode, all values of the added to_column column are set with the fixed value specified by the value parameter, which becomes a required parameter, see below;
    • incremental_time
      In this mode, all values of the added to_column column are set to a value that increments by 1 second for each record, starting at from timestamp and up to to timestamp, after which it wrap around and starts from from again and so on - hence the additional from and to parameters are required in this mode.
    • upload_time
      In this mode, all values of the added to_column column are set with the fixed value corresponding to the time the Embulk upload was started. This mode does not require additional parameters.
  • timestamp_format
    a string specifying how to parse the string value provided as either value or from/to parameters depending on the mode in use (default is "%Y-%m-%d %H:%M:%S %z").
    It follow the Ruby's strptime format. Note that at the moment also Unix timestamps and epoch times need to be specified as string, therefore the timestamp_format parameter needs to specify how to parse it.
  • unix_timestamp_unit
    a string specifying the expected unit of the unix timestamp for the long value provides as either value or from/to parameters sec, milli (for milliseconds), micro (for microseconds), or nano (for nanoseconds) are the supported values for this option (default is sec).

These parameters are mutually exclusive and the usage depend on the mode. The type of them is string or long. The format of the string value used here needs to match the rule provided in the timestamp_format parameter. The long value is converted to the unix timestamp with the unit provided in the unix_timestamp_unit parameter:

  • value
    a string or long values specifying a fixed value for the added time column. This options is required if the mode is fixed_time.
  • from and to
    two strings or long values specifying the value for the beginning and end of the range of time for mode: incremental_time. The format needs to be consistent between these two parameters.

Example: mode: fixed_time

filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
  from_value:
    mode: fixed_time
    value: "2016-01-01 00:00:00 UTC"
    timestamp_format: "%Y-%m-%d %H:%M:%S %z"

The mode fixed_time is default and can be omitted:

filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
  from_value:
    value: "2016-01-01 00:00:00 UTC"
filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
  from_value:
    value: 1453939479

where in this example the format of the value is also corresponding to the default, therefore the timestamp_format option is also omitted.

Example: mode: incremental_time

filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
  from_value:
    mode: incremental_time
    from: "2016-01-01 00:00:00 UTC"
    to: "2016-01-01 01:00:00 UTC"

Example: mode: upload_time

filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
  from_value:
    mode: upload_time

where neither of the additional parameters value, to, and from is specified. Embulk will associate the fixed bulk upload start time to every record.

from_column configuration

This configuration is mutually exclusive with from_value above. One amongst from_column and from_value configurations is required.

The from_column configuration specifies the name of one of the columns found in the Embulk schema and the format to be used to parse and feed values in the column added by the to_column configuration. This configuration makes a copy of the values from the column specified by name, instead of renaming the source column itself. The parameters for this configuration are expressed as a set of key-value pairs.

  • name
    a string specifying the name of the source column from the Embulk schema. The column type must be one of long, timestamp or string (required).
  • unix_timestamp_unit
    a string specifying the expected unit of the unix timestamp for the values of the source column: it is required only if the type of the source column is long (see above for supported types). The supported values are sec for seconds, milli for milliseconds, micro for microseconds, and nano for nanoseconds (default is sec).
  • timestamp_format
    a string specifying the expected format of the values in the source column: it is required only if the type of the column is string (see above for supported types) (default is "%Y-%m-%d %H:%M:%S %z"). It follow the Ruby's strptime format.

Note that if neither timestamp_format or unix_timestamp_unit parameters are specified, the column type is expected to be timestamp.

Example: created_at column with timestamp data type

filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
  from_column:
    name: created_at

In this example the expected type for the source column created_at is timestamp. If the created_at column type is string, the timestamp_format parameter is required to provide Embulk with a way to parse the values and convert them to a timestamp.

filters:
- type: add_time
  to_column:
    name: time
    type: timestamp
  from_column:
    name: created_at
    timestamp_format: "%Y-%m-%d %H:%M:%S"
    timezone: UTC

If the type of the source column is long, the unix_timestamp_unit parameter is required. The additional parameter provides Embulk the information to properly parse the long values of the source column and convert them to type specified in the to_column configuration.

filters:
- type: add_time
  to_column:
    name: time
    type: long
    unix_timestamp_unit: sec
  from_column:
    name: created_at
    unix_timestamp_unit: milli

Install

$ embulk gem install embulk-filter-add_time

Build

Build by Gradle

$ git clone https://github.com/treasure-data/embulk-filter-add_time.git
$ cd embulk-filter-add_time
$ ./gradlew gem classpath

Run on Embulk

$ bin/embulk run -I embulk-filter-add_time/lib/ config.yml