gpmapreduce configuration file.
%YAML 1.1 --- VERSION: 126.96.36.199 DATABASE: dbname USER: db_username HOST: master_hostname PORT: master_port
DEFINE: - INPUT: NAME: input_name FILE: - hostname:/path/to/file GPFDIST: - hostname:port/file_pattern TABLE: table_name QUERY: SELECT_statement EXEC: command_string COLUMNS: - field_name data_type FORMAT: TEXT | CSV DELIMITER: delimiter_character ESCAPE: escape_character NULL: null_string QUOTE: csv_quote_character ERROR_LIMIT: integer ENCODING: database_encoding
- OUTPUT: NAME: output_name FILE: file_path_on_client TABLE: table_name KEYS: - column_name MODE: REPLACE | APPEND
- MAP: NAME: function_name FUNCTION: function_definition LANGUAGE: perl | python | c LIBRARY: /path/filename.so PARAMETERS: - nametype RETURNS: - nametype OPTIMIZE: STRICT IMMUTABLE MODE: SINGLE | MULTI
- TRANSITION | CONSOLIDATE | FINALIZE: NAME: function_name FUNCTION: function_definition LANGUAGE: perl | python | c LIBRARY: /path/filename.so PARAMETERS: - nametype RETURNS: - nametype OPTIMIZE: STRICT IMMUTABLE MODE: SINGLE | MULTI
- REDUCE: NAME: reduce_job_name TRANSITION: transition_function_name CONSOLIDATE: consolidate_function_name FINALIZE: finalize_function_name INITIALIZE: value KEYS: - key_name
- TASK: NAME: task_name SOURCE: input_name MAP: map_function_name REDUCE: reduce_function_name EXECUTE:
You specify the input, map and reduce tasks, and the output for the Greenplum MapReduce gpmapreduce program in a YAML-formatted configuration file. (This reference page uses the name gpmapreduce.yaml when referring to this file; you may choose your own name for the file.)
The gpmapreduce utility processes the YAML configuration file in order, using indentation (spaces) to determine the document hierarchy and the relationships between the sections. The use of white space in the file is significant.
Keys and Values
Required. The version of the Greenplum MapReduce YAML specification. Current supported versions are 188.8.131.52, 184.108.40.206, and 220.127.116.11.
- Optional. Specifies which database in Greenplum to connect to. If not specified, defaults to the default database or $PGDATABASE if set.
- Optional. Specifies which database role to use to connect. If not specified, defaults to the current user or $PGUSER if set. You must be a Greenplum superuser to run functions written in untrusted Python and Perl. Regular database users can run functions written in trusted Perl. You also must be a database superuser to run MapReduce jobs that contain FILE, GPFDIST and EXEC input types.
- Optional. Specifies Greenplum master host name. If not specified, defaults to localhost or $PGHOST if set.
- Optional. Specifies Greenplum master port. If not specified, defaults to 5432 or $PGPORT if set.
- Required. A sequence of definitions for this MapReduce document. The
DEFINE section must have at least one INPUT
- Required. Defines the input data. Every MapReduce document must have at least
one input defined. Multiple input definitions are allowed in a document, but each
input definition can specify only one of these access types: a file, a
gpfdist file reference, a table in the database, an
SQL command, or an operating system command. See gpfdist
for information about this reference.
A name for this input. Names must be unique with regards to the names of other objects in this MapReduce job (such as map function, task, reduce function and output names). Also, names cannot conflict with existing objects in the database (such as tables, functions or views).
- A sequence of one or more input files in the format: seghostname:/path/to/filename. You must be a Greenplum Database superuser to run MapReduce jobs with FILE input. The file must reside on a Greenplum segment host.
- A sequence identifying one or more running gpfdist file servers in the format: hostname[:port]/file_pattern. You must be a Greenplum Database superuser to run MapReduce jobs with GPFDIST input.
- The name of an existing table in the database.
- A SQL SELECT command to run within the database.
- An operating system command to run on the Greenplum segment hosts. The command is run by all segment instances in the system by default. For example, if you have four segment instances per segment host, the command will be run four times on each host. You must be a Greenplum Database superuser to run MapReduce jobs with EXEC input.
- Optional. Columns are specified as: column_name [data_type]. If not specified, the default is value text. The DELIMITER character is what separates two data value fields (columns). A row is determined by a line feed character (0x0a).
Optional. Specifies the format of the data - either delimited text (TEXT) or comma separated values (CSV) format. If the data format is not specified, defaults to TEXT.
Optional for FILE, GPFDIST and EXEC inputs. Specifies a single character that separates data values. The default is a tab character in TEXT mode, a comma in CSV mode. The delimiter character must only appear between any two data value fields. Do not place a delimiter at the beginning or end of a row.
- Optional for FILE, GPFDIST and EXEC inputs. Specifies the single character that is used for C escape sequences (such as \n,\t,\100, and so on) and for escaping data characters that might otherwise be taken as row or column delimiters. Make sure to choose an escape character that is not used anywhere in your actual column data. The default escape character is a \ (backslash) for text-formatted files and a " (double quote) for csv-formatted files, however it is possible to specify another character to represent an escape. It is also possible to disable escaping by specifying the value 'OFF' as the escape value. This is very useful for data such as text-formatted web log data that has many embedded backslashes that are not intended to be escapes.
- Optional for FILE, GPFDIST and EXEC inputs. Specifies the string that represents a null value. The default is \N in TEXT format, and an empty value with no quotations in CSV format. You might prefer an empty string even in TEXT mode for cases where you do not want to distinguish nulls from empty strings. Any input data item that matches this string will be considered a null value.
- Optional for FILE, GPFDIST and EXEC inputs. Specifies the quotation character for CSV formatted files. The default is a double quote ("). In CSV formatted files, data value fields must be enclosed in double quotes if they contain any commas or embedded new lines. Fields that contain double quote characters must be surrounded by double quotes, and the embedded double quotes must each be represented by a pair of consecutive double quotes. It is important to always open and close quotes correctly in order for data rows to be parsed correctly.
- If the input rows have format errors they will be discarded provided that the error limit count is not reached on any Greenplum segment instance during input processing. If the error limit is not reached, all good rows will be processed and any error rows discarded.
- Character set encoding to use for the data. Specify a string constant (such as 'SQL_ASCII'), an integer encoding number, or DEFAULT to use the default client encoding. See Character Set Support for more information.
- Optional. Defines where to output the formatted data of this MapReduce job. If
output is not defined, the default is STDOUT (standard output of
the client). You can send output to a file on the client host or to an existing
table in the database.
- A name for this output. The default output name is STDOUT. Names must be unique with regards to the names of other objects in this MapReduce job (such as map function, task, reduce function and input names). Also, names cannot conflict with existing objects in the database (such as tables, functions or views).
- Specifies a file location on the MapReduce client machine to output data in the format: /path/to/filename.
- Specifies the name of a table in the database to output data. If this table does not exist prior to running the MapReduce job, it will be created using the distribution policy specified with KEYS.
- Optional for TABLE output. Specifies the column(s) to use as the Greenplum Database distribution key. If the EXECUTE task contains a REDUCE definition, then the REDUCE keys will be used as the table distribution key by default. Otherwise, the first column of the table will be used as the distribution key.
- Optional for TABLE output. If not specified, the default is to create the table if it does not already exist, but error out if it does exist. Declaring APPEND adds output data to an existing table (provided the table schema matches the output format) without removing any existing data. Declaring REPLACE will drop the table if it exists and then recreate it. Both APPEND and REPLACE will create a new table if one does not exist.
- Required. Each MAP function takes data structured in (key, value) pairs, processes each pair, and generates zero or more output (key, value) pairs. The Greenplum MapReduce framework then collects all pairs with the same key from all output lists and groups them together. This output is then passed to the REDUCE task, which is comprised of TRANSITION | CONSOLIDATE | FINALIZE functions. There is one predefined MAP function named IDENTITY that returns (key, value) pairs unchanged. Although (key, value) are the default parameters, you can specify other prototypes as needed.
- TRANSITION | CONSOLIDATE | FINALIZE
- TRANSITION, CONSOLIDATE and FINALIZE are all component pieces of REDUCE. A TRANSITION function is required. CONSOLIDATE and FINALIZE functions are optional. By default, all take state as the first of their input PARAMETERS, but other prototypes can be defined as well.
A TRANSITION function iterates through each value of a given key and accumulates values in a state variable. When the transition function is called on the first value of a key, the state is set to the value specified by INITIALIZE of a REDUCE job (or the default state value for the data type). A transition takes two arguments as input; the current state of the key reduction, and the next value, which then produces a new state.
- If a CONSOLIDATE function is specified, TRANSITION processing is performed at the segment-level before redistributing the keys across the Greenplum interconnect for final aggregation (two-phase aggregation). Only the resulting state value for a given key is redistributed, resulting in lower interconnect traffic and greater parallelism. CONSOLIDATE is handled like a TRANSITION, except that instead of (state + value) => state, it is (state + state) => state.
- If a FINALIZE function is specified, it takes the final
state produced by CONSOLIDATE (if present) or
TRANSITION and does any final processing before emitting the
final result. TRANSITION and CONSOLIDATE
functions cannot return a set of values. If you need a
REDUCE job to return a set, then a FINALIZE is
necessary to transform the final state into a set of output values.
- Required. A name for the function. Names must be unique with regards to the names of other objects in this MapReduce job (such as function, task, input and output names). You can also specify the name of a function built-in to Greenplum Database. If using a built-in function, do not supply LANGUAGE or a FUNCTION body.
- Optional. Specifies the full body of the function using the specified LANGUAGE. If FUNCTION is not specified, then a built-in database function corresponding to NAME is used.
- Required when FUNCTION is used. Specifies the implementation language used to interpret the function. This release has language support for perl, python, and C. If calling a built-in database function, LANGUAGE should not be specified.
- Required when LANGUAGE is C (not allowed for other language functions). To use this attribute, VERSION must be 18.104.22.168. The specified library file must be installed prior to running the MapReduce job, and it must exist in the same file system location on all Greenplum hosts (master and segments).
- Optional. Function input parameters. The default type is
MAP default - key text, value text
TRANSITION default - state text, value text
CONSOLIDATE default - state1 text, state2 text (must have exactly two input parameters of the same data type)
FINALIZE default - state text (single parameter only)
- Optional. The default return type is
MAP default - key text, value text
TRANSITION default - state text (single return value only)
CONSOLIDATE default - state text (single return value only)
FINALIZE default - value text
- Optional optimization parameters for the
STRICT - function is not affected by NULL values
IMMUTABLE - function will always return the same value for a given input
- Optional. Specifies the number of rows returned by the function.
MULTI - returns 0 or more rows per input record. The return value of the function must be an array of rows to return, or the function must be written as an iterator using yield in Python or return_next in Perl. MULTI is the default mode for MAP and FINALIZE functions.
SINGLE - returns exactly one row per input record. SINGLE is the only mode supported for TRANSITION and CONSOLIDATE functions. When used with MAP and FINALIZE functions, SINGLE mode can provide modest performance improvement.
- Required. A REDUCE definition names the TRANSITION | CONSOLIDATE | FINALIZE
functions that comprise the reduction of (key,
value) pairs to the final result set. There are also several
predefined REDUCE jobs you can execute, which all operate over a
column named value:
IDENTITY - returns (key, value) pairs unchanged
SUM - calculates the sum of numeric data
AVG - calculates the average of numeric data
COUNT - calculates the count of input data
MIN - calculates minimum value of numeric data
MAX - calculates maximum value of numeric data
- Required. The name of this REDUCE job. Names must be unique with regards to the names of other objects in this MapReduce job (function, task, input and output names). Also, names cannot conflict with existing objects in the database (such as tables, functions or views).
- Required. The name of the TRANSITION function.
- Optional. The name of the CONSOLIDATE function.
- Optional. The name of the FINALIZE function.
- Optional for text and float data types. Required for all other data types. The default value for text is '' . The default value for float is 0.0 . Sets the initial state value of the TRANSITION function.
- Optional. Defaults to [key, *]. When using a multi-column reduce it may be necessary to specify which columns are key columns and which columns are value columns. By default, any input columns that are not passed to the TRANSITION function are key columns, and a column named key is always a key column even if it is passed to the TRANSITION function. The special indicator * indicates all columns not passed to the TRANSITION function. If this indicator is not present in the list of keys then any unmatched columns are discarded.
- Optional. A TASK defines a complete end-to-end
INPUT/MAP/REDUCE stage within
a Greenplum MapReduce job pipeline. It is similar to EXECUTE except it is not immediately executed. A task
object can be called as INPUT to
further processing stages.
- Required. The name of this task. Names must be unique with regards to the names of other objects in this MapReduce job (such as map function, reduce function, input and output names). Also, names cannot conflict with existing objects in the database (such as tables, functions or views).
- The name of an INPUT or another TASK.
Optional. The name of a MAP function. If not specified, defaults to IDENTITY.
Optional. The name of a REDUCE function. If not specified, defaults to IDENTITY.
- Required. EXECUTE defines the final INPUT/MAP/REDUCE stage within a Greenplum MapReduce job pipeline.