org.apache.pig

Class StoreFunc

java.lang.Object

org.apache.pig.StoreFunc

All Implemented Interfaces:

StoreFuncInterface

Direct Known Subclasses:

DBStorage, FixedWidthStorer, JsonStorage, MultiStorage

@InterfaceAudience.Public
@InterfaceStability.Stable
public abstract class StoreFunc
extends Object
implements StoreFuncInterface

StoreFuncs take records from Pig's processing and store them into a data store. Most frequently this is an HDFS file, but it could also be an HBase instance, RDBMS, etc.

Constructor Summary

Constructors

Constructor and Description
StoreFunc()

Method Summary

Methods

Modifier and Type	Method and Description
void	checkSchema(ResourceSchema s) Set the schema for data to be stored.
void	cleanupOnFailure(String location, org.apache.hadoop.mapreduce.Job job) This method will be called by Pig if the job which contains this store fails.
static void	cleanupOnFailureImpl(String location, org.apache.hadoop.mapreduce.Job job) Default implementation for cleanupOnFailure(String, Job) and cleanupOnSuccess(String, Job).
void	cleanupOnSuccess(String location, org.apache.hadoop.mapreduce.Job job) This method will be called by Pig if the job which contains this store is successful, and some cleanup of intermediate resources is required.
abstract org.apache.hadoop.mapreduce.OutputFormat	getOutputFormat() Return the OutputFormat associated with StoreFunc.
abstract void	prepareToWrite(org.apache.hadoop.mapreduce.RecordWriter writer) Initialize StoreFunc to write data.
abstract void	putNext(Tuple t) Write a tuple to the data store.
String	relToAbsPathForStoreLocation(String location, org.apache.hadoop.fs.Path curDir) This method is called by the Pig runtime in the front end to convert the output location to an absolute path if the location is relative.
void	setStoreFuncUDFContextSignature(String signature) This method will be called by Pig both in the front end and back end to pass a unique signature to the StoreFunc which it can use to store information in the UDFContext which it needs to store between various method invocations in the front end and back end.
abstract void	setStoreLocation(String location, org.apache.hadoop.mapreduce.Job job) Communicate to the storer the location where the data needs to be stored.
Boolean	supportsParallelWriteToStoreLocation() DAG execution engines like Tez support optimizing union by writing to output location in parallel from tasks of different vertices.
void	warn(String msg, Enum warningEnum) Issue a warning.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

StoreFunc

public StoreFunc()

Method Detail

relToAbsPathForStoreLocation

public String relToAbsPathForStoreLocation(String location,
                                  org.apache.hadoop.fs.Path curDir)
                                    throws IOException

This method is called by the Pig runtime in the front end to convert the output location to an absolute path if the location is relative. The StoreFunc implementation is free to choose how it converts a relative location to an absolute location since this may depend on what the location string represent (hdfs path or some other data source).

Specified by:

relToAbsPathForStoreLocation in interface StoreFuncInterface

Parameters:

location - location as provided in the "store" statement of the script

curDir - the current working direction based on any "cd" statements in the script before the "store" statement. If there are no "cd" statements in the script, this would be the home directory -

/user/ 

Returns:

the absolute location based on the arguments passed

Throws:

IOException - if the conversion is not possible

getOutputFormat

public abstract org.apache.hadoop.mapreduce.OutputFormat getOutputFormat()
                                                                  throws IOException

Return the OutputFormat associated with StoreFunc. This will be called on the front end during planning and on the backend during execution.

Specified by:

getOutputFormat in interface StoreFuncInterface

Returns:

the OutputFormat associated with StoreFunc

Throws:

IOException - if an exception occurs while constructing the OutputFormat

setStoreLocation

public abstract void setStoreLocation(String location,
                    org.apache.hadoop.mapreduce.Job job)
                               throws IOException

Communicate to the storer the location where the data needs to be stored. The location string passed to the StoreFunc here is the return value of relToAbsPathForStoreLocation(String, Path) This method will be called in the frontend and backend multiple times. Implementations should bear in mind that this method is called multiple times and should ensure there are no inconsistent side effects due to the multiple calls. checkSchema(ResourceSchema) will be called before any call to setStoreLocation(String, Job).

Specified by:

setStoreLocation in interface StoreFuncInterface

Parameters:

location - Location returned by relToAbsPathForStoreLocation(String, Path)

job - The Job object

Throws:

IOException - if the location is not valid.

checkSchema

public void checkSchema(ResourceSchema s)
                 throws IOException

Set the schema for data to be stored. This will be called on the front end during planning if the store is associated with a schema. A Store function should implement this function to check that a given schema is acceptable to it. For example, it can check that the correct partition keys are included; a storage function to be written directly to an OutputFormat can make sure the schema will translate in a well defined way. Default implementation is a no-op.

Specified by:

checkSchema in interface StoreFuncInterface

Parameters:

s - to be checked

Throws:

IOException - if this schema is not acceptable. It should include a detailed error message indicating what is wrong with the schema.

prepareToWrite

public abstract void prepareToWrite(org.apache.hadoop.mapreduce.RecordWriter writer)
                             throws IOException

Initialize StoreFunc to write data. This will be called during execution on the backend before the call to putNext.

Specified by:

prepareToWrite in interface StoreFuncInterface

Parameters:

writer - RecordWriter to use.

Throws:

IOException - if an exception occurs during initialization

putNext

public abstract void putNext(Tuple t)
                      throws IOException

Write a tuple to the data store.

Specified by:

putNext in interface StoreFuncInterface

Parameters:

t - the tuple to store.

Throws:

IOException - if an exception occurs during the write

setStoreFuncUDFContextSignature

public void setStoreFuncUDFContextSignature(String signature)

This method will be called by Pig both in the front end and back end to pass a unique signature to the StoreFunc which it can use to store information in the UDFContext which it needs to store between various method invocations in the front end and back end. This method will be called before other methods in StoreFunc. This is necessary because in a Pig Latin script with multiple stores, the different instances of store functions need to be able to find their (and only their) data in the UDFContext object. The default implementation is a no-op.

Specified by:

setStoreFuncUDFContextSignature in interface StoreFuncInterface

Parameters:

signature - a unique signature to identify this StoreFunc

cleanupOnFailure

public void cleanupOnFailure(String location,
                    org.apache.hadoop.mapreduce.Job job)
                      throws IOException

This method will be called by Pig if the job which contains this store fails. Implementations can clean up output locations in this method to ensure that no incorrect/incomplete results are left in the output location. The default implementation deletes the output location if it is a FileSystem location.

Specified by:

cleanupOnFailure in interface StoreFuncInterface

Parameters:

location - Location returned by relToAbsPathForStoreLocation(String, Path)

job - The Job object - this should be used only to obtain cluster properties through JobContextImpl.getConfiguration() and not to set/query any runtime job information.

Throws:

IOException

cleanupOnSuccess

public void cleanupOnSuccess(String location,
                    org.apache.hadoop.mapreduce.Job job)
                      throws IOException

This method will be called by Pig if the job which contains this store is successful, and some cleanup of intermediate resources is required. Implementations can clean up output locations in this method to ensure that no incorrect/incomplete results are left in the output location.

Specified by:

cleanupOnSuccess in interface StoreFuncInterface

Parameters:

location - Location returned by relToAbsPathForStoreLocation(String, Path)

job - The Job object - this should be used only to obtain cluster properties through JobContextImpl.getConfiguration() and not to set/query any runtime job information.

Throws:

IOException

cleanupOnFailureImpl

public static void cleanupOnFailureImpl(String location,
                        org.apache.hadoop.mapreduce.Job job)
                                 throws IOException

Default implementation for cleanupOnFailure(String, Job) and cleanupOnSuccess(String, Job). This removes a file from HDFS.

Parameters:

location - file name (or URI) of file to remove

job - Hadoop job, used to access the appropriate file system.

Throws:

IOException

supportsParallelWriteToStoreLocation

public Boolean supportsParallelWriteToStoreLocation()

DAG execution engines like Tez support optimizing union by writing to output location in parallel from tasks of different vertices. Commit is called once all the vertices in the union are complete. This eliminates need to have a separate phase to read data output from previous phases, union them and write out again. Enabling the union optimization requires the OutputFormat to 1) Support creation of different part file names for tasks of different vertices. Conflicting filenames can create data corruption and loss. For eg: If task 0 of vertex1 and vertex2 both create filename as part-r-00000, then one of the files will be overwritten when promoting from temporary to final location leading to data loss. FileOutputFormat has mapreduce.output.basename config which enables naming files differently in different vertices. Classes extending FileOutputFormat and those prefixing file names with mapreduce.output.basename value will not encounter conflict. Cases like HBaseStorage which write to key value store and do not produce files also should not face any conflict. 2) Support calling of commit once at the end takes care of promoting temporary files of the different vertices into the final location. For eg: FileOutputFormat commit algorithm handles promoting of files produced by tasks of different vertices into final output location without issues if there is no file name conflict. In cases like HBaseStorage, the TableOutputCommitter does nothing on commit. If custom OutputFormat used by the StoreFunc does not support the above two criteria, then false should be returned. Union optimization will be disabled for the StoreFunc. Default implementation returns null and in that case planner falls back to PigConfiguration.PIG_TEZ_OPT_UNION_SUPPORTED_STOREFUNCS and PigConfiguration.PIG_TEZ_OPT_UNION_UNSUPPORTED_STOREFUNCS settings to determine if the StoreFunc supports it.

warn

public final void warn(String msg,
        Enum warningEnum)

Issue a warning. Warning messages are aggregated and reported to the user.

Parameters:

msg - String message of the warning

warningEnum - type of warning