# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#  http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.

# Pig configuration file. All values can be overwritten by command line
# arguments; for a description of the properties, run
#
#     pig -h properties
#

############################################################################
#
# == Logging properties
#

# Location of pig log file. If blank, a file with a timestamped slug
# ('pig_1399336559369.log') will be generated in the current working directory.
#
# pig.logfile=
# pig.logfile=/tmp/pig-err.log

# Log4j configuration file. Set at runtime with the -4 parameter. The source
# distribution has a ./conf/log4j.properties.template file you can rename and
# customize.
#
# log4jconf=./conf/log4j.properties

# Verbose Output.
# * false (default): print only INFO and above to screen
# * true: Print all log messages to screen
#
# verbose=false

# Omit timestamps on log messages. (default: false)
#
# brief=false

# Logging level. debug=OFF|ERROR|WARN|INFO|DEBUG (default: INFO)
#
# debug=INFO

# Roll up warnings across tasks, so that when millions of mappers suddenly cry
# out in error they are partially silenced. (default, recommended: true)
#
# aggregate.warning=true

# Should DESCRIBE pretty-print its schema?
# * false (default): print on a single-line, suitable for pasting back in to your script
# * true (recommended): prints on multiple lines with indentation, much more readable
#
# pig.pretty.print.schema=false

# === Profiling UDFs  ===

# Turn on UDF timers? This will cause two counters to be
# tracked for every UDF and LoadFunc in your script: approx_microsecs measures
# approximate time spent inside a UDF approx_invocations reports the approximate
# number of times the UDF was invoked.
#
# * false (default): do not record timing information of UDFs.
# * true: report UDF performance. Uses more counters, but gives more insight
#   into script operation
#
# pig.udf.profile=false

# Specify frequency of profiling (default: every 100th).
# pig.udf.profile.frequency=100

############################################################################
#
# == Site-specific Properties
#

# Execution Mode. Local mode is much faster, but only suitable for small amounts
# of data. Local mode interprets paths on the local file system; Mapreduce mode
# on the HDFS. Read more under 'Execution Modes' within the Getting Started
# documentation.
#
# * mapreduce (default): use the Hadoop cluster defined in your Hadoop config files
# * local: use local mode
#
# exectype=mapreduce

# Bootstrap file with default statements to execute in every Pig job, similar to
# .bashrc.  If blank, uses the file '.pigbootup' from your home directory; If a
# value is supplied, that file is NOT loaded.  This does not do tilde expansion
# -- you must supply the full path to the file.
#
# pig.load.default.statements=
# pig.load.default.statements=/home/bob/.pigrc

# Kill all waiting/running MR jobs upon a MR job failure? (default: false) If
# false, jobs that can proceed independently will do so unless a parent stage
# fails. If true, the failure of any stage in the script kills all jobs.
#
# stop.on.failure=false

# File containing the pig script to run. Rarely set in the properties file.
# Commandline: -f
#
# file=

# Jarfile to load, colon separated. Rarely used.
#
# jar=

# Register additional .jar files to use with your Pig script.
# Most typically used as a command line option (see http://pig.apache.org/docs/r0.12.0/basic.html#register):
#
#     pig -Dpig.additional.jars=hdfs://nn.mydomain.com:9020/myjars/my.jar
#
# pig.additional.jars=<colon separated list of jars with optional wildcards>
# pig.additional.jars=/usr/local/share/pig/pig/contrib/piggybank/java/piggybank.jar:/usr/local/share/pig/datafu/datafu-pig/build/libs/datafu-pig-1.2.1.jar

# Specify potential packages to which a UDF or a group of UDFs belong,
# eliminating the need to qualify the UDF on every call. See
# http://pig.apache.org/docs/r0.12.0/udf.html#use-short-names
#
# Commandline use:
#
#     pig \
#       -Dpig.additional.jars=$PIG_HOME/contrib/piggybank/java/piggybank.jar:$PIG_HOME/../datafu/datafu-pig/build/libs/datafu-pig-1.2.1.jar \
#       -Dudf.import.list=org.apache.pig.piggybank.evaluation:datafu.pig.util \
#       happy_job.pig
#
# udf.import.list=<colon separated list of imports>
# udf.import.list=org.apache.pig.piggybank.evaluation:datafu.pig.bags:datafu.pig.hash:datafu.pig.stats:datafu.pig.util

#
# Reuse jars across jobs run by the same user? (default: false) If enabled, jars
# are placed in ${pig.user.cache.location}/${user.name}/.pigcache. Since most
# jars change infrequently, this gives a minor speedup.
#
# pig.user.cache.enabled=false

# Base path for storing jars cached by the pig.user.cache.enabled feature. (default: /tmp)
#
# pig.user.cache.location=/tmp

# Default UTC offset. (default: the host's current UTC offset) Supply a UTC
# offset in Java's timezone format: e.g., +08:00.
#
# pig.datetime.default.tz=

############################################################################
#
# Memory impacting properties
#

# Amount of memory (as fraction of heap) allocated to bags before a spill is
# forced. Default is 0.2, meaning 20% of available memory. Note that this memory
# is shared across all large bags used by the application. See
# http://pig.apache.org/docs/r0.12.0/perf.html#memory-management
#
# pig.cachedbag.memusage=0.2

# Don't spill bags smaller than this size (bytes). Default: 5000000, or about
# 5MB. Usually, the more spilling the longer runtime, so you might want to tune
# it according to heap size of each task and so forth.
#
# pig.spill.size.threshold=5000000

# EXPERIMENTAL: If a file bigger than this size (bytes) is spilled -- thus
# freeing a bunch of ram -- tell the JVM to perform garbage collection.  This
# should help reduce the number of files being spilled, but causes more-frequent
# garbage collection. Default: 40000000 (about 40 MB)
#
# pig.spill.gc.activation.size=40000000

# Maximum amount of data to replicate using the distributed cache when doing
# fragment-replicated join. (default: 1000000000, about 1GB) Consider increasing
# this in a production environment, but carefully.
#
# pig.join.replicated.max.bytes=1000000000

# Fraction of heap available for the reducer to perform a skewed join. A low
# fraction forces Pig to use more reducers, but increases the copying cost. See
# http://pig.apache.org/docs/r0.12.0/perf.html#skewed-joins
#
# pig.skewedjoin.reduce.memusage=0.3

#
# === SchemaTuple ===
#
# The SchemaTuple feature (PIG-2632) uses a tuple's schema (when known) to
# generate a custom Java class to hold records. Otherwise, tuples are loaded as
# a plain list that is unaware of its contents' schema -- and so each element
# has to be wrapped as a Java object on its own. This can provide more efficient
# CPU utilization, serialization, and most of all memory usage.
#
# This feature is considered experimental and is off by default. You can
# selectively enable it for specific operations using pig.schematuple.udf,
# pig.schematuple.load, pig.schematuple.fr_join and pig.schematuple.merge_join
#

# Enable the SchemaTuple optimization in all available cases? (default: false; recommended: true)
#
# pig.schematuple=false

# EXPERIMENTAL: Use SchemaTuples with UDFs (default: value of pig.schematuple).
# pig.schematuple.udf=false

# EXPERIMENTAL, CURRENTLY NOT IMPLEMENTED, but in the future, LoadFunc's with
# known schemas should output SchemaTuples. (default: value of pig.schematuple)
# pig.schematuple.load=false

# EXPERIMENTAL: Use SchemaTuples in replicated joins. The potential memory
# saving here is significant. (default: value of pig.schematuple)
# pig.schematuple.fr_join=false

# EXPERIMENTAL: Use SchemaTuples in merge joins. (default: value of pig.schematuple).
# pig.schematuple.merge_join=false

############################################################################
#
# Serialization options
#

# Omit empty part files from the output? (default: false)
#
# * false (default): reducers generates an output file, even if output is empty
# * true (recommended): do not generate zero-byte part files
#
# The default behavior of MapReduce is to generate an empty file for no data, so
# Pig follows that. But many small files can cause annoying extra map tasks and
# put load on the HDFS, so consider setting this to 'true'
#
# pig.output.lazy=false

#
# === Tempfile Handling
#

# EXPERIMENTAL: Storage format for temporary files generated by intermediate
# stages of Pig jobs. This can provide significant speed increases for certain
# codecs, as reducing the amount of data transferred to and from disk can more
# than make up for the cost of compression/compression. Recommend that you set
# up LZO compression in Hadoop and specify tfile storage.
#
# Compress temporary files?
# * false (default): do not compress
# * true (recommended): compress temporary files.
#
# pig.tmpfilecompression=false
# pig.tmpfilecompression=true

# Tempfile storage container type.
#
# * tfile (default, recommended): more efficient, but only supports supports gz(gzip) and lzo compression.
#   https://issues.apache.org/jira/secure/attachment/12396286/TFile%20Specification%2020081217.pdf
# * seqfile: only supports gz(gzip), lzo, snappy, and bzip2 compression
#
# pig.tmpfilecompression.storage=tfile

# Codec types for intermediate job files. tfile supports gz(gzip) and lzo;
# seqfile support gz(gzip), lzo, snappy, bzip2
#
# * lzo (recommended with caveats): moderate compression, low cpu burden;
#   typically leads to a noticeable speedup. Best default choice, but you must
#   set up LZO independently due to license incompatibility
# * snappy: moderate compression, low cpu burden; typically leads to a noticeable speedup..
# * gz (default): higher compression, high CPU burden. Typically leads to a noticeable slowdown.
# * bzip2: most compression, major CPU burden. Typically leads to a noticeable slowdown.
#
# pig.tmpfilecompression.codec=gzip

#
# === Split Combining
#

#
# Should pig try to combine small files for fewer map tasks? This improves the
# efficiency of jobs with many small input files, reduces the overhead on the
# jobtracker, and reduces the number of output files a map-only job
# produces. However, it only works with certain loaders and increases non-local
# map tasks. See http://pig.apache.org/docs/r0.12.0/perf.html#combine-files
#
# * false (default, recommended): _do_ combine files
# * true: do not combine files
#
# pig.noSplitCombination=false

#
# Size, in bytes, of data to be processed by a single map. Smaller files are
# combined untill this size is reached. If unset, defaults to the file system's
# default block size.
#
# pig.maxCombinedSplitSize=

# ###########################################################################
#
# Execution options
#

# Should pig omit combiners? (default, recommended: false -- meaning pig _will_
# use combiners)
#
# When combiners work well, they eliminate a significant amount of
# data. However, if they do not eliminate much data -- say, a DISTINCT operation
# that only eliminates 5% of the records -- they add a noticeable overhead to
# the job. So the recommended default is false (use combiners), selectively
# disabling them per-job:
#
#     pig -Dpig.exec.nocombiner=true distinct_but_not_too_much.pig
#
# pig.exec.nocombiner=false

# EXPERIMENTAL: Aggregate records in map task before sending to the combiner?
# (default: false, 10; recommended: true, 10). In cases where there is a massive
# reduction of data in the aggregation step, pig can do a first pass of
# aggregation before the data even leaves the mapper, saving much serialization
# overhead. It's off by default but can give a major improvement to
# group-and-aggregate operations. Pig skips partial aggregation unless reduction
# is better than a factor of minReduction (default: 10). See
# http://pig.apache.org/docs/r0.12.0/perf.html#hash-based-aggregation
#
# pig.exec.mapPartAgg=false
# pig.exec.mapPartAgg.minReduction=10

#
# === Control how many reducers are used.
#

# Estimate number of reducers naively using a fixed amount of data per
# reducer. Optimally, you have both fewer reducers than available reduce slots,
# and reducers that are neither getting too little data (less than a half-GB or
# so) nor too much data (more than 2-3 times the reducer child process max heap
# size). The default of 1000000000 (about 1GB) is probably low for a production
# cluster -- however it's much worse to set this too high (reducers spill many
# times over in group-sort) than too low (delay waiting for reduce slots).
#
# pig.exec.reducers.bytes.per.reducer=1000000000

#
# Don't ever use more than this many reducers. (default: 999)
#
# pig.exec.reducers.max=999

#
# === Local mode for small jobs
#

# EXPERIMENTAL: Use local mode for small jobs? If true, jobs with input data
# size smaller than pig.auto.local.input.maxbytes bytes and one or no reducers
# are run in local mode, which is much faster. Note that file paths are still
# interpreted as pig.exectype implies.
#
# * true (recommended): allow local mode for small jobs, which is much faster.
# * false (default): always use pig.exectype.
#
# pig.auto.local.enabled=false

#
# Definition of a small job for the pig.auto.local.enabled feature. Only jobs
# with less than this may bytes are candidates to run locally (default:
# 100000000 bytes, about 1GB)
#
# pig.auto.local.input.maxbytes=100000000

############################################################################
#
# Security Features
#

# Comma-delimited list of commands/operators that are disallowed. This security
# feature can be used by administrators to block use of certain commands by
# users.
#
# * <blank> (default): all commands and operators are allowed.
# * fs,set (for example): block all filesystem commands and config changes from pig scripts.
#
# pig.blacklist=
# pig.blacklist=fs,set

# Comma-delimited list of the only commands/operators that are allowed. This
# security feature can be used by administrators to block use of certain
# commands by users.
#
# * <blank> (default): all commands and operators not on the pig.blacklist are allowed.
# * load,store,filter,group: only LOAD, STORE, FILTER, GROUP
#   from pig scripts. All other commands and operators will fail.
#
# pig.whitelist=
# pig.whitelist=load,store,filter,group

#####################################################################
#
# Advanced Site-specific Customizations
#

# Remove intermediate output files?
#
# * true (default, recommended): remove the files
# * false: do NOT remove the files. You must clean them up yourself.
#
# Keeping them is useful for advanced debugging, but can be dangerous -- you
# must clean them up yourself.  Inspect the intermediate outputs with
#
#     LOAD '/path/to/tmp/file' USING org.apache.pig.impl.io.TFileStorage();
#
# (Or ...SequenceFileInterStorage if pig.tmpfilecompression.storage is seqfile)
#
# pig.delete.temp.files=true

# EXPERIMENTAL: A Pig Progress Notification Listener (PPNL) lets you wire pig's
# progress into your visibility stack. To use a PPNL, supply the fully qualified
# class name of a PPNL implementation. Note that only one PPNL can be set up, so
# if you need several, write a PPNL that will chain them.
#
# See https://github.com/twitter/ambrose for a pretty awesome one of these
#
# pig.notification.listener=<fully qualified class name of a PPNL implementation>

# String argument to pass to your PPNL constructor (optional). Only a single
# string value is allowed. (default none)
#
# pig.notification.listener.arg=<somevalue>

# EXPERIMENTAL: Class invoked to estimate the number of reducers to use.
# (default: org.apache.pig.backend.hadoop.executionengine.mapReduceLayer.InputSizeReducerEstimator)
#
# If you don't know how or why to write a PigReducerEstimator, you're unlikely
# to use this. By default, the naive mapReduceLayer.InputSizeReducerEstimator is
# used, but you can specify anything implementing the interface
# org.apache.pig.backend.hadoop.executionengine.mapReduceLayer.PigReducerEstimator
#
# pig.exec.reducer.estimator=<fully qualified class name of a PigReducerEstimator implementation>

# Optional String argument to pass to your PigReducerEstimator. (default: none;
# a single String argument is allowed).
#
# pig.exec.reducer.estimator.arg=<somevalue>

# Class invoked to report the size of reducers output. By default, the reducers'
# output is computed as the total size of output files. But not every storage is
# file-based, and so this logic can be replaced by implementing the interface
# org.apache.pig.backend.hadoop.executionengine.mapReduceLayer.PigStatsOutputSizeReader
# If you need to register more than one reader, you can register them as a comma
# separated list. Every reader implements a boolean supports(POStore sto) method.
# When there are more than one reader, they are consulted in order, and the
# first one whose supports() method returns true will be used.
#
# pig.stats.output.size.reader=<fully qualified class name of a PigStatsOutputSizeReader implementation>
# pig.stats.output.size.reader.unsupported=<comma separated list of StoreFuncs that are not supported by this reader>

# By default, Pig retrieves TaskReports for every launched task to compute
# various job statistics. But this can cause OOM if the number of tasks is
# large. In such case, you can disable it by setting this property to true.
# pig.stats.notaskreport=false

#
# Override hadoop configs programatically
#
# By default, Pig expects hadoop configs (hadoop-site.xml and core-site.xml)
# to be present on the classpath. There are cases when these configs are
# needed to be passed programatically, such as while using the PigServer API.
# In such cases, you can override hadoop configs by setting the property
# "pig.use.overriden.hadoop.configs".
#
# When this property is set to true, Pig ignores looking for hadoop configs
# in the classpath and instead picks it up from Properties/Configuration
# object passed to it.
#
# pig.use.overriden.hadoop.configs=false

# Implied LoadFunc for the LOAD operation when no USING clause is
# present. Supply the fully qualified class name of a LoadFunc
# implementation. Note: setting this means you will have to modify most code
# brought in from elsewhere on the web, as people generally omit the USING
# clause for TSV files.
#
# * org.apache.pig.builtin.PigStorage (default): the traditional tab-separated-values LoadFunc
# * my.custom.udfcollection.MyCustomLoadFunc (for example): use MyCustomLoadFunc instead
#
# pig.default.load.func=<fully qualified class name of a LoadFunc implementation>

# The implied StoreFunc for STORE operations with no USING clause. Supply the
# fully qualified class name of a StoreFunc implementation.
#
# * org.apache.pig.builtin.PigStorage (default): the traditional tab-separated-values StoreFunc.
# * my.custom.udfcollection.MyCustomStoreFunc (for example): use MyCustomStoreFunc instead
#
# pig.default.store.func=<fully qualified class name of a StoreFunc implementation>

# Recover jobs when the application master is restarted? (default: false). This
# is a Hadoop 2 specific property; enable it to take advantage of AM recovery.
#
# pig.output.committer.recovery.support=true

# Should scripts check to prevent multiple stores writing to the same location?
# (default: false) When set to true, stops the execution of script right away.
#
pig.location.check.strict=false

# In addition to the fs-style commands (rm, ls, etc) Pig can now execute
# SQL-style DDL commands, eg "sql create table pig_test(name string, age int)".
# The only implemented backend is hcat, and luckily that's also the default.
#
# pig.sql.type=hcat

# Path to the hcat executable, for use with pig.sql.type=hcat (default: null)
#
hcat.bin=/usr/local/hcat/bin/hcat

###########################################################################
#
# Overrides for extreme environments
#
# (Most people won't have to adjust these parameters)
#


# Limit the pig script length placed in the jobconf xml. (default:10240)
# Extremely long queries can waste space in the JobConf; since its contents are
# only advisory, the default is fine unless you are retaining it for forensics.
#
# pig.script.max.size=10240

# Disable use of counters by Pig. Note that the word 'counter' is singular here.
#
# * false (default, recommended): do NOT disable counters.
# * true: disable counters. Set this to true only when your Pig job will
#   otherwise die because of using more counters than hadoop configured limit
#
# pig.disable.counter=true

# Sample size (per-mapper, in number of rows) the ORDER..BY operation's
# RandomSampleLoader uses to estimate how your data should be
# partitioned. (default, recommended: 100 rows per task) Increase this if you
# have exceptionally large input splits and are unhappy with the reducer skew.
#
# pig.random.sampler.sample.size=100

# Process an entire script at once, reducing the amount of work and number of
# tasks? (default, recommended: true) See http://pig.apache.org/docs/r0.12.0/perf.html#multi-query-execution
#
# MultiQuery optimization is very useful, and so the recommended default is
# true. You may find a that a script fails to compile under MultiQuery. If so,
# disable it at runtime:
#
#     pig -no_multiquery script_that_makes_pig_sad.pig
#
# opt.multiquery=true

# For small queries, fetch data directly from the HDFS. (default, recommended:
# true). If you want to force Pig to launch a MR job, for example when you're
# testing a live cluster, disable with the -N option. See PIG-3642.
#
# opt.fetch=true

# Enable auto parallelism in tez. This should be used by default unless
# you encounter some bug in automatic parallelism. If set to false, use 1 as
# default parallelism
pig.tez.auto.parallelism=true

###########################################################################
#
# Streaming properties
#

# Define what properties will be set in the streaming environment. Just set this
# property to a comma-delimited list of properties to set, and those properties
# will be set in the environment.
#
# pig.streaming.environment=<comma-delimited list of propertes>

# Specify a comma-delimited list of local files to ship to distributed cache for
# streaming job.
#
# pig.streaming.ship.files=<comma-delimited list of local files>

# Specify a comma-delimited list of remote files to cache on distributed cache
# for streaming job.
#
# pig.streaming.cache.files=<comma-delimited list of remote files>

# Specify the python command to be used for python streaming udf. By default,
# python is used, but you can overwrite it with a non-default version such as
# python2.7.
#
# pig.streaming.udf.python.command=python