=head1 NAME
MCE::Core - Documentation describing the core MCE API
=head1 VERSION
This document describes MCE::Core version 1.862
=head1 SYNOPSIS
This is a simplistic use case of MCE running with 5 workers.
## Construction using the Core API
use MCE;
my $mce = MCE->new(
max_workers => 5,
user_func => sub {
my ($mce) = @_;
$mce->say("Hello from " . $mce->wid);
}
);
$mce->run;
## Construction using a MCE model
use MCE::Flow max_workers => 5;
mce_flow sub {
my ($mce) = @_;
MCE->say("Hello from " . MCE->wid);
};
-- Output
Hello from 2
Hello from 4
Hello from 5
Hello from 1
Hello from 3
=head2 MCE->new ( [ options ] )
Below, a new instance is configured with all available options.
use MCE;
my $mce = MCE->new(
max_workers => 8, ## Default 1
# Number of workers to spawn. This can be set automatically
# with MCE 1.412 and later releases.
# MCE 1.521 sets an upper-limit of 8 for 'auto'.
# See MCE::Util::get_ncpu for more info.
# max_workers => 'auto', ## # of lcores, 8 maximum
# max_workers => 'auto-1', ## 7 on HW with 16-lcores
# max_workers => 'auto-1', ## 3 on HW with 4-lcores
# max_workers => MCE::Util::get_ncpu, # run on all lcores
chunk_size => 2000, ## Default 1
# Can also take a suffix; k (kibiBytes) or m (mebiBytes).
# The default is 1 when using the Core API and 'auto' for
# MCE Models. For arrays or queues, chunk_size means the
# number of records per chunk. For iterators, MCE will not
# use chunk_size, though the iterator may use it to determine
# how much to return per iteration. For files, smaller than or
# equal to 8192 is the number of records. Greater than 8192
# is the number of bytes. MCE reads until the end of record
# before calling user_func.
# chunk_size => 1, ## Consists of 1 record
# chunk_size => 1000, ## Consists of 1000 records
# chunk_size => '16k', ## Approximate 16 kibiBytes (KiB)
# chunk_size => '20m', ## Approximate 20 mebiBytes (MiB)
tmp_dir => $tmp_dir, ## Default $MCE::Signal::tmp_dir
# Default is $MCE::Signal::tmp_dir which points to
# $ENV{TEMP} if defined. Otherwise, tmp_dir points
# to a location under /tmp.
freeze => \&encode_sereal, ## Default \&Storable::freeze
thaw => \&decode_sereal, ## Default \&Storable::thaw
# Release 1.412 allows freeze and thaw to be overridden.
# Simply include a serialization module prior to loading
# MCE. Configure freeze/thaw options.
# use Sereal qw( encode_sereal decode_sereal );
# use CBOR::XS qw( encode_cbor decode_cbor );
# use JSON::XS qw( encode_json decode_json );
#
# use MCE;
gather => \@a, ## Default undef
# Release 1.5 allows for gathering of data to an array or
# hash reference, a MCE::Queue/Thread::Queue object, or code
# reference. One invokes gathering by calling the gather
# method as often as needed.
# gather => \@array,
# gather => \%hash,
# gather => $queue,
# gather => \&order,
init_relay => 0, ## Default undef
# For specifying the initial relay value. Allowed values
# are array_ref, hash_ref, or scalar. The MCE::Relay module
# is loaded automatically when specified.
# init_relay => \@array,
# init_relay => \%hash,
# init_relay => scalar,
input_data => $input_file, ## Default undef
RS => "\n>", ## Default undef
# input_data => '/path/to/file' ## Process file
# input_data => \@array ## Process array
# input_data => \*FILE_HNDL ## Process file handle
# input_data => $io ## Process IO::All { File, Pipe, STDIO }
# input_data => \$scalar ## Treated like a file
# input_data => \&iterator ## User specified iterator
# The RS option (for input record separator) applies to files
# and file handles.
# MCE applies additional logic when RS begins with a newline
# character; e.g. RS => "\n>". It trims away characters after
# the newline and prepends them to the next record.
#
# Typically, the left side is what happens for $/ = "\n>".
# The right side is what user_func receives.
#
# All records begin with > and end with \n
# Record 1: >seq1 ... \n> (to) >seq1 ... \n
# Record 2: seq2 ... \n> >seq2 ... \n
# Record 3: seq3 ... \n> >seq3 ... \n
# Last Rec: seqN ... \n >seqN ... \n
loop_timeout => 20, ## Default 0
# Added in 1.7, enables the manager process to timeout of a read
# operation on channel 0 (UNIX platforms only). The manager process
# decrements the total workers running for any worker which have
# died in an uncontrollable manner. Specify this option if on
# occassion a worker dies unexpectedly (i.e. from an XS module).
# Option works with init_relay on UNIX platforms since MCE 1.844.
# A number smaller than 5 is silently increased to 5.
max_retries => 2, ## Default 0
# This option, added in 1.7, causes MCE to retry a failed
# chunk from a worker dying while processing input data or
# sequence of numbers.
parallel_io => 1, ## Default 0
posix_exit => 1, ## Default 0
use_slurpio => 1, ## Default 0
# The parallel_io option enables parallel reads during large
# slurpio, useful when reading from fast storage. Do not enable
# parallel_io when running MCE on many nodes with input coming
# from shared storage.
# Set posix_exit to avoid all END and destructor processing.
# Constructing MCE inside a thread implies 1 or if present CGI,
# FCGI, Coro, Curses, Gearman::Util, Gearman::XS, LWP::UserAgent,
# Mojo::IOLoop, STFL, Tk, Wx, or Win32::GUI.
# Enable slurpio to pass the raw chunk (scalar ref) to the user
# function when reading input files.
use_threads => 1, ## Auto 0 or 1
# MCE spawns child processes by default, not threads.
#
# However, MCE supports threads via 2 threading
# libraries if threads is desired.
# The use of threads in MCE requires that you include
# threads support prior to loading MCE. The use_threads
# option defaults to 1 when a thread library is loaded.
# Threads is loaded automatically for $^O eq 'MSWin32'.
#
# use threads; use forks;
# use threads::shared; (or) use forks::shared;
# use MCE use MCE;
spawn_delay => 0.045, ## Default undef
submit_delay => 0.015, ## Default undef
job_delay => 0.060, ## Default undef
# Time to wait in fractional seconds after spawning a worker,
# after submitting parameters to worker (MCE->run, MCE->process),
# and worker running (one time staggered delay).
# Specify job_delay to stagger workers connecting to a database.
on_post_exit => \&on_post_exit, ## Default undef
on_post_run => \&on_post_run, ## Default undef
# Execute the code block after a worker exits or dies.
# (i.e. MCE->exit, exit, die)
# Execute the code block after running.
# (i.e. MCE->process, MCE->run)
progress => sub { ... }, ## Default undef
# A code block for receiving info on the progress made.
# See section labeled "MCE PROGRESS DEMONSTRATIONS" at the
# end of this document.
user_args => { env => 'test' }, ## Default undef
# MCE release 1.4 added a new parameter to allow one to
# specify arbitrary arguments such as a string, an ARRAY
# or HASH reference. Workers can access this directly.
# (i.e. my $args = $mce->{user_args} or MCE->user_args)
user_begin => \&user_begin, ## Default undef
user_func => \&user_func, ## Default undef
user_end => \&user_end, ## Default undef
# Think of user_begin, user_func, and user_end as in
# the awk scripting language:
# awk 'BEGIN { begin } { func } { func } ... END { end }'
# MCE workers call user_begin once at the start of a job,
# then user_func repeatedly until no chunks remain.
# Afterwards, user_end is called.
user_error => \&user_error, ## Default undef
user_output => \&user_output, ## Default undef
# MCE will forward data to user_error/user_output,
# when defined, for the following methods.
# MCE->sendto(\*STDERR, "sent to user_error\n");
# MCE->printf(\*STDERR, "%s\n", "sent to user_error");
# MCE->print(\*STDERR, "sent to user_error\n");
# MCE->say(\*STDERR, "sent to user_error");
# MCE->sendto(\*STDOUT, "sent to user_output\n");
# MCE->printf("%s\n", "sent to user_output");
# MCE->print("sent to user_output\n");
# MCE->say("sent to user_output");
stderr_file => 'err_file', ## Default STDERR
stdout_file => 'out_file', ## Default STDOUT
# Or to file; user_error and user_output take precedence.
flush_file => 1, ## Default 0
flush_stderr => 1, ## Default 0
flush_stdout => 1, ## Default 0
# Flush sendto file, standard error, or standard output.
interval => {
delay => 0.007 [, max_nodes => 4, node_id => 1 ]
},
# For use with the yield method introduced in MCE 1.5.
# Both max_nodes & node_id are optional and default to 1.
# Delay is the amount of time between intervals.
# interval => 0.007 ## Shorter; MCE 1.506+
sequence => { ## Default undef
begin => -1, end => 1 [, step => 0.1 [, format => "%4.1f" ] ]
},
bounds_only => 1, ## Default undef
# For looping through a sequence of numbers in parallel.
# STEP, if omitted, defaults to 1 if BEGIN is smaller than
# END or -1 if BEGIN is greater than END. The FORMAT string
# is passed to sprintf behind the scene (% may be omitted).
# e.g. $seq_n_formatted = sprintf("%4.1f", $seq_n);
# Do not specify both options; input_data and sequence.
# Release 1.4 allows one to specify an array reference.
# e.g. sequence => [ -1, 1, 0.1, "%4.1f" ]
# The bounds_only => 1 option will compute the 'begin' and
# 'end' items only for the chunk and not the items in between
# (hence boundaries only). This option has no effect when
# sequence is not specified or chunk_size equals 1.
# my $begin = $chunk_ref->[0]; my $end = $chunk_ref->[1];
task_end => \&task_end, ## Default undef
# This is called by the manager process after the task
# has completed processing. MCE 1.5 allows this option
# to be specified at the top level.
task_name => 'string', ## Default 'MCE'
# Added in MCE 1.5 and mainly beneficial for user_tasks.
# One may specify a unique name per each sub-task.
# The string is passed as the 3rd arg to task_end.
user_tasks => [ ## Default undef
{ ... }, ## Options for task 0
{ ... }, ## Options for task 1
{ ... }, ## Options for task 2
],
# Takes a list of hash references, each allowing up to 17
# options. All other MCE options are ignored. The init_relay,
# input_data, RS, and use_slurpio options are applicable to
# the first task only.
# max_workers, chunk_size, input_data, interval, sequence,
# bounds_only, user_args, user_begin, user_end, user_func,
# gather, task_end, task_name, use_slurpio, use_threads,
# init_relay, RS
# Options not specified here will default to same option
# specified at the top level.
);
=head2 EXPORT_CONST, CONST
There are 3 constants which are exportable. Using the constants in lieu of
0,1,2 makes it more legible when accessing the user_func arguments directly.
=head3 SELF CHUNK CID - MCE CONSTANTS
Exports SELF => 0, CHUNK => 1, and CID => 2.
use MCE export_const => 1;
use MCE const => 1; ## Shorter; MCE 1.415+
user_func => sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
print "Hello from ", $_[SELF]->wid, "\n";
}
MCE 1.5 allows all public method to be called directly.
use MCE;
user_func => sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
print "Hello from ", MCE->wid, "\n";
}
=head2 OVERRIDING DEFAULTS
The following list options which may be overridden when loading the module.
use Sereal qw( encode_sereal decode_sereal );
use CBOR::XS qw( encode_cbor decode_cbor );
use JSON::XS qw( encode_json decode_json );
use MCE
max_workers => 4, ## Default 1
chunk_size => 100, ## Default 1
tmp_dir => "/path/to/app/tmp", ## $MCE::Signal::tmp_dir
freeze => \&encode_sereal, ## \&Storable::freeze
thaw => \&decode_sereal ## \&Storable::thaw
;
my $mce = MCE->new( ... );
From MCE 1.8 onwards, Sereal 3.015+ is loaded automatically if available.
Specify C<< Sereal => 0 >> to use Storable instead.
use MCE Sereal => 0;
=head2 RUNNING
Run calls spawn, submits the job; workers call user_begin, user_func, and
user_end. Run shuts down workers afterwards. Call spawn whenever the need
arises for large data structures prior to running.
$mce->spawn; ## Call early if desired
$mce->run; ## Call run or process below
## Acquire data arrays and/or input_files. Workers persist after
## processing.
$mce->process(\@input_data_1); ## Process array
$mce->process(\@input_data_2);
$mce->process(\@input_data_n);
$mce->process(\%input_hash_1); ## Process hash, current API
$mce->process(\%input_hash_2); ## available since 1.828
$mce->process(\%input_hash_n);
$mce->process('input_file_1'); ## Process file
$mce->process('input_file_2');
$mce->process('input_file_n');
$mce->shutdown; ## Shutdown workers
=head2 SYNTAX for ON_POST_EXIT
Often times, one may want to capture the exit status. The on_post_exit option,
if defined, is executed immediately by the manager process after a worker exits
via exit (children only), MCE->exit (children and threads), or die.
The format of $e->{pid} is PID_123 for children and THR_123 for threads.
my $restart_flag = 1;
sub on_post_exit {
my ($mce, $e) = @_;
## Display all possible hash elements.
print "$e->{wid}: $e->{pid}: $e->{status}: $e->{msg}: $e->{id}\n";
## Restart this worker if desired.
if ($restart_flag && $e->{wid} == 2) {
$mce->restart_worker;
$restart_flag = 0;
}
}
sub user_func {
my ($mce) = @_;
MCE->exit(0, 'msg_foo', 1000 + MCE->wid); ## Args, not necessary
}
my $mce = MCE->new(
on_post_exit => \&on_post_exit,
user_func => \&user_func,
max_workers => 3
);
$mce->run;
-- Output (child processes)
2: PID_33223: 0: msg_foo: 1002
1: PID_33222: 0: msg_foo: 1001
3: PID_33224: 0: msg_foo: 1003
2: PID_33225: 0: msg_foo: 1002
-- Output (running with threads)
3: TID_3: 0: msg_foo: 1003
2: TID_2: 0: msg_foo: 1002
1: TID_1: 0: msg_foo: 1001
2: TID_4: 0: msg_foo: 1002
=head2 SYNTAX for ON_POST_RUN
The on_post_run option, if defined, is executed immediately by the manager
process after running MCE->process or MCE->run. This option receives an
array reference of hashes.
The difference between on_post_exit and on_post_run is that the former is
called immediately whereas the latter is called after all workers have
completed running.
sub on_post_run {
my ($mce, $status_ref) = @_;
foreach my $e ( @{ $status_ref } ) {
## Display all possible hash elements.
print "$e->{wid}: $e->{pid}: $e->{status}: $e->{msg}: $e->{id}\n";
}
}
sub user_func {
my ($mce) = @_;
MCE->exit(0, 'msg_foo', 1000 + MCE->wid); ## Args, not necessary
}
my $mce = MCE->new(
on_post_run => \&on_post_run,
user_func => \&user_func,
max_workers => 3
);
$mce->run;
-- Output (child processes)
3: PID_33174: 0: msg_foo: 1003
1: PID_33172: 0: msg_foo: 1001
2: PID_33173: 0: msg_foo: 1002
-- Output (running with threads)
2: TID_2: 0: msg_foo: 1002
3: TID_3: 0: msg_foo: 1003
1: TID_1: 0: msg_foo: 1001
=head2 SYNTAX for INPUT_DATA
MCE supports many ways to specify input_data. Support for iterators was added
in MCE 1.505. The RS option allows one to specify the record separator when
processing files.
MCE is a chunking engine. Therefore, chunk_size is applicable to input_data.
Specifying 1 for use_slurpio causes user_func to receive a scalar reference
containing the raw data (applicable to files only) instead of an array
reference.
C<IO::All> { File, Pipe, STDIO } is supported since MCE 1.845.
input_data => '/path/to/file', ## process file
input_data => \@array, ## process array
input_data => \%hash, ## process hash, API since 1.828
input_data => \*FILE_HNDL, ## process file handle
input_data => $fh, ## open $fh, "<", "file"
input_data => $fh, ## IO::File "file", "r"
input_data => $fh, ## IO::Uncompress::Gunzip "file.gz"
input_data => $io, ## IO::All { File, Pipe, STDIO }
input_data => \$scalar, ## treated like a file
input_data => \&iterator, ## user specified iterator
chunk_size => 1, ## >1 means looping inside user_func
use_slurpio => 1, ## $chunk_ref is a scalar ref
RS => "\n>", ## input record separator
The chunk_size value determines the chunking mode to use when processing files.
Otherwise, chunk_size is the number of elements for arrays. For files, a chunk
size value of <= 8192 is how many records to read. Greater than 8192 is how
many bytes to read. MCE appends (the rest) up to the next record separator.
chunk_size => 8192, ## Consists of 8192 records
chunk_size => 8193, ## Approximate 8193 bytes for files
chunk_size => 1, ## Consists of 1 record or element
chunk_size => 1000, ## Consists of 1000 records
chunk_size => '16k', ## Approximate 16 kibiBytes (KiB)
chunk_size => '20m', ## Approximate 20 mebiBytes (MiB)
The construction for user_func when chunk_size > 1 and assuming use_slurpio
equals 0.
user_func => sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
## $_ is $chunk_ref->[0] when chunk_size equals 1
## $_ is $chunk_ref otherwise; $_ can be used below
for my $record ( @{ $chunk_ref } ) {
print "$chunk_id: $record\n";
}
}
# input_data => \%hash
# current API available since 1.828
user_func => sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
## $_ points to $chunk_ref regardless of chunk_size
for my $key ( keys %{ $chunk_ref } ) {
print "$key: ", $chunk_ref->{$key}, "\n";
}
}
Specifying a value for input_data is straight forward for arrays and files.
The next several examples specify an iterator reference for input_data.
use MCE;
## A factory function which creates a closure (the iterator itself)
## for generating a sequence of numbers. The external variables
## ($n, $max, $step) are used for keeping state across successive
## calls to the closure. The iterator simply returns when $n > max.
sub input_iterator {
my ($n, $max, $step) = @_;
return sub {
return if $n > $max;
my $current = $n;
$n += $step;
return $current;
};
}
## Run user_func in parallel. Input data can be specified during
## the construction or as an argument to the process method.
my $mce = MCE->new(
# input_data => input_iterator(10, 30, 2),
chunk_size => 1, max_workers => 4,
user_func => sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
MCE->print("$_: ", $_ * 2, "\n");
}
)->spawn;
$mce->process( input_iterator(10, 30, 2) );
-- Output Note that output order is not guaranteed
Take a look at iterator.pl for ordered output
10: 20
12: 24
16: 32
20: 40
14: 28
22: 44
18: 36
24: 48
26: 52
28: 56
30: 60
The following example queries the DB for the next 1000 rows. Notice the use of
fetchall_arrayref. The iterator function itself receives one argument which is
chunk_size (added in MCE 1.510) to determine how much to return per iteration.
The default is 1 for the Core API and MCE Models.
use DBI;
use MCE;
sub db_iter {
my $dsn = "DBI:Oracle:host=db_server;port=db_port;sid=db_name";
my $dbh = DBI->connect($dsn, 'db_user', 'db_passwd') ||
die "Could not connect to database: $DBI::errstr";
my $sth = $dbh->prepare('select color, desc from table');
$sth->execute;
return sub {
my ($chunk_size) = @_;
if (my $aref = $sth->fetchall_arrayref(undef, $chunk_size)) {
return @{ $aref };
}
return;
};
}
## Let's enumerate column indexes for easy column retrieval.
my ($i_color, $i_desc) = (0 .. 1);
my $mce = MCE->new(
max_workers => 3, chunk_size => 1000,
input_data => db_iter(),
user_func => sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
my $ret = '';
foreach my $row (@{ $chunk_ref }) {
$ret .= $row->[$i_color] .": ". $row->[$i_desc] ."\n";
}
MCE->print($ret);
}
);
$mce->run;
There are many modules on CPAN which return an iterator reference. Showing
one such example below. The demonstration ensures MCE workers are spawned
before obtaining the iterator. Note the worker_id value (left column) in
the output.
use Path::Iterator::Rule;
use MCE;
my $start_dir = shift
or die "Please specify a starting directory";
-d $start_dir
or die "Cannot open ($start_dir): No such file or directory";
my $mce = MCE->new(
max_workers => 'auto',
user_func => sub { MCE->say( MCE->wid . ": $_" ) }
)->spawn;
my $rule = Path::Iterator::Rule->new->file->name( qr/[.](pm)$/ );
my $iterator = $rule->iter(
$start_dir, { follow_symlinks => 0, depthfirst => 1 }
);
$mce->process( $iterator );
-- Output
8: lib/MCE/Core/Input/Generator.pm
5: lib/MCE/Core/Input/Handle.pm
6: lib/MCE/Core/Input/Iterator.pm
2: lib/MCE/Core/Input/Request.pm
3: lib/MCE/Core/Manager.pm
4: lib/MCE/Core/Input/Sequence.pm
7: lib/MCE/Core/Validation.pm
1: lib/MCE/Core/Worker.pm
8: lib/MCE/Flow.pm
5: lib/MCE/Grep.pm
6: lib/MCE/Loop.pm
2: lib/MCE/Map.pm
3: lib/MCE/Queue.pm
4: lib/MCE/Signal.pm
7: lib/MCE/Stream.pm
1: lib/MCE/Subs.pm
8: lib/MCE/Util.pm
5: lib/MCE.pm
Although MCE supports arrays, extra measures are needed to use a "lazy" array
as input data. The reason for this is that MCE needs the size of the array
before processing which may be unknown for lazy arrays. Therefore, closures
provides an excellent mechanism for this.
The code block belonging to the lazy array must return undef after exhausting
its input data. Otherwise, the process will never end.
use Tie::Array::Lazy;
use MCE;
tie my @a, 'Tie::Array::Lazy', [], sub {
my $i = $_[0]->index;
return ($i < 10) ? $i : undef;
};
sub make_iterator {
my $i = 0; my $a_ref = shift;
return sub {
return $a_ref->[$i++];
};
}
my $mce = MCE->new(
max_workers => 4, input_data => make_iterator(\@a),
user_func => sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
MCE->say($_);
}
)->run;
-- Output
0
1
2
3
4
6
7
8
5
9
The following demonstrates how to retrieve a chunk from the lazy array per
each successive call. Here, undef is sent by the iterator block when $i is
greater than $max. Iterators may optionally use chunk_size to determine how
much to return per iteration.
use Tie::Array::Lazy;
use MCE;
tie my @a, 'Tie::Array::Lazy', [], sub {
$_[0]->index;
};
sub make_iterator {
my $j = 0; my ($a_ref, $max) = @_;
return sub {
my ($chunk_size) = @_;
my $i = $j; $j += $chunk_size;
return if $i > $max;
return $j <= $max ? @$a_ref[$i .. $j - 1] : @$a_ref[$i .. $max];
};
}
my $mce = MCE->new(
chunk_size => 15, max_workers => 4,
input_data => make_iterator(\@a, 100),
user_func => sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
MCE->say("$chunk_id: " . join(' ', @{ $chunk_ref }));
}
)->run;
-- Output
1: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
2: 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
3: 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
4: 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59
5: 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
6: 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89
7: 90 91 92 93 94 95 96 97 98 99 100
=head2 SYNTAX for SEQUENCE
The 1.3 release and above allows workers to loop through a sequence of numbers
computed mathematically without the overhead of an array. The sequence can be
specified separately per each user_task entry unlike input_data which is
applicable to the first task only.
See the seq_demo.pl example, included with this distribution, on applying
sequences with the user_tasks option.
Sequence can be defined using an array or a hash reference.
use MCE;
my $mce = MCE->new(
max_workers => 3,
# sequence => [ 10, 19, 0.7, "%4.1f" ], ## up to 4 options
sequence => {
begin => 10, end => 19, step => 0.7, format => "%4.1f"
},
user_func => sub {
my ($mce, $n, $chunk_id) = @_;
print $n, " from ", MCE->wid, " id ", $chunk_id, "\n";
}
);
$mce->run;
-- Output (sorted afterwards, notice wid and chunk_id in output)
10.0 from 1 id 1
10.7 from 2 id 2
11.4 from 3 id 3
12.1 from 1 id 4
12.8 from 2 id 5
13.5 from 3 id 6
14.2 from 1 id 7
14.9 from 2 id 8
15.6 from 3 id 9
16.3 from 1 id 10
17.0 from 2 id 11
17.7 from 3 id 12
18.4 from 1 id 13
The 1.5 release includes a new option (bounds_only). This option tells the
sequence engine to compute 'begin' and 'end' items only, for the chunk,
and not the items in between (hence boundaries only). This option applies
to sequence only and has no effect when chunk_size equals 1.
The time to run is 0.006s below. This becomes 0.827s without the bounds_only
option due to computing all items in between, thus creating a very large
array. Basically, specify bounds_only => 1 when boundaries is all you need
for looping inside the block; e.g. Monte Carlo simulations.
Time was measured using 1 worker to emphasize the difference.
use MCE;
my $mce = MCE->new(
max_workers => 1, chunk_size => 1_250_000,
sequence => { begin => 1, end => 10_000_000 },
bounds_only => 1,
## For sequence, the input scalar $_ points to $chunk_ref
## when chunk_size > 1, otherwise $chunk_ref->[0].
##
## user_func => sub {
## my $begin = $_->[0]; my $end = $_->[-1];
##
## for ($begin .. $end) {
## ...
## }
## },
user_func => sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
## $chunk_ref contains 2 items, not 1_250_000
my $begin = $chunk_ref->[ 0];
my $end = $chunk_ref->[-1]; ## or $chunk_ref->[1]
MCE->printf("%7d .. %8d\n", $begin, $end);
}
);
$mce->run;
-- Output
1 .. 1250000
1250001 .. 2500000
2500001 .. 3750000
3750001 .. 5000000
5000001 .. 6250000
6250001 .. 7500000
7500001 .. 8750000
8750001 .. 10000000
=head2 SYNTAX for MAX_RETRIES
The max_retries option, added in 1.7, allows MCE to retry a failed chunk from
a worker dying while processing input data or a sequence of numbers.
When max_retries is set, MCE configures the on_post_exit option automatically
using the following code before running. Specify on_post_exit explicitly for
any further tailoring. The restart_worker line is necessary, obviously.
on_post_exit => sub {
my ( $mce, $e, $retry_cnt ) = @_;
if ( $e->{id} ) {
my $cnt = $retry_cnt + 1;
my $msg = "Error: chunk $e->{id} failed";
if ( defined $mce->{init_relay} ) {
print {*STDERR} "$msg, retrying chunk attempt # $cnt\n"
if ( $retry_cnt < $mce->{max_retries} );
}
else {
( $retry_cnt < $mce->{max_retries} )
? print {*STDERR} "$msg, retrying chunk attempt # $cnt\n"
: print {*STDERR} "$msg\n";
}
$mce->restart_worker;
}
}
We let MCE handle on_post_exit automatically below, which is essentially the
same code shown above. For max_retries to work, the worker must die, abnormally
included, or call MCE->exit. Notice that we pass the chunk_id value for the 3rd
argument to MCE->exit (defaults to chunk_id if omitted since MCE 1.844).
## max_retries demonstration
use strict;
use warnings;
use MCE;
sub user_func {
my ( $mce, $chunk_ref, $chunk_id ) = @_;
# die "Died : chunk_id = 3\n" if $chunk_id == 3;
MCE->exit(1, undef, $chunk_id) if $chunk_id == 3;
print "$chunk_id\n";
}
my $mce = MCE->new(
max_workers => 1,
max_retries => 2,
user_func => \&user_func,
)->spawn;
my $input_data = [ 0..7 ];
$mce->process( { chunk_size => 1 }, $input_data );
$mce->shutdown;
-- Output
1
2
Error: chunk 3 failed, retrying chunk attempt # 1
Error: chunk 3 failed, retrying chunk attempt # 2
Error: chunk 3 failed
4
5
6
7
8
Orderly output with max_retries is possible since MCE 1.844. Below, chunk 3
succeeds whereas chunk 5 fails due to exceeding the number of retries. Be sure
to call MCE::relay inside C<user_func> and near the end of the block.
## max_retries demonstration with init_relay
use strict;
use warnings;
use MCE;
use MCE::Shared;
tie my $retries1, 'MCE::Shared', 0;
tie my $retries2, 'MCE::Shared', 0;
MCE->new(
max_workers => 4,
input_data => [ 1..7 ],
chunk_size => 1,
max_retries => 2,
init_relay => 0,
user_func => sub {
if ( MCE->chunk_id == 3 ) {
MCE->exit if ++$retries1 <= 2;
}
if ( MCE->chunk_id == 5 ) {
MCE->exit if ++$retries2 <= 3;
}
MCE::relay {
$_ += 1;
print MCE->chunk_id, "\n";
};
}
)->run;
print "final: ", MCE::relay_final(), "\n";
-- Output
1
2
Error: chunk 3 failed, retrying chunk attempt # 1
Error: chunk 5 failed, retrying chunk attempt # 1
Error: chunk 3 failed, retrying chunk attempt # 2
Error: chunk 5 failed, retrying chunk attempt # 2
3
4
Error: chunk 5 failed
6
7
final: 6
=head2 SYNTAX for USER_BEGIN and USER_END
The user_begin and user_end options, if specified, behave similarly to
awk 'BEGIN { begin } { func } { func } ... END { end }'. These are called
once per worker during each run.
MCE 1.510 passes 2 additional parameters ($task_id and $task_name).
sub user_begin { ## Called once at the beginning
my ($mce, $task_id, $task_name) = @_;
$mce->{wk_total_rows} = 0;
}
sub user_func { ## Called while processing
my $mce = shift;
$mce->{wk_total_rows} += 1;
}
sub user_end { ## Called once at the end
my ($mce, $task_id, $task_name) = @_;
printf "## %d: Processed %d rows\n",
MCE->wid, $mce->{wk_total_rows};
}
my $mce = MCE->new(
user_begin => \&user_begin,
user_func => \&user_func,
user_end => \&user_end
);
$mce->run;
=head2 SYNTAX for USER_FUNC with USE_SLURPIO => 0
When processing input data, MCE can pass an array of rows or a slurped chunk.
Below, a reference to an array containing the chunk data is processed.
e.g. $chunk_ref = [ record1, record2, record3, ... ]
sub user_func {
my ($mce, $chunk_ref, $chunk_id) = @_;
foreach my $row ( @{ $chunk_ref } ) {
$mce->{wk_total_rows} += 1;
print $row;
}
}
my $mce = MCE->new(
chunk_size => 100,
input_data => "/path/to/file",
user_func => \&user_func,
use_slurpio => 0
);
$mce->run;
=head2 SYNTAX for USER_FUNC with USE_SLURPIO => 1
Here, a reference to a scalar containing the raw chunk data is processed.
sub user_func {
my ($mce, $chunk_ref, $chunk_id) = @_;
my $count = () = $$chunk_ref =~ /abc/;
}
my $mce = MCE->new(
chunk_size => 16000,
input_data => "/path/to/file",
user_func => \&user_func,
use_slurpio => 1
);
$mce->run;
=head2 SYNTAX for USER_ERROR and USER_OUTPUT
Output from MCE->sendto('STDERR/STDOUT', ...), MCE->printf, MCE->print, and
MCE->say can be intercepted by specifying the user_error and user_output
options. MCE on receiving output will forward to user_error or user_output
in a serialized fashion.
Handy when wanting to filter, modify, and/or direct the output elsewhere.
sub user_error { ## Redirect STDERR to STDOUT
my $error = shift;
print {*STDOUT} $error;
}
sub user_output { ## Redirect STDOUT to STDERR
my $output = shift;
print {*STDERR} $output;
}
sub user_func {
my ($mce, $chunk_ref, $chunk_id) = @_;
my $count = 0;
foreach my $row ( @{ $chunk_ref } ) {
MCE->print($row);
$count += 1;
}
MCE->print(\*STDERR, "$chunk_id: processed $count rows\n");
}
my $mce = MCE->new(
chunk_size => 1000,
input_data => "/path/to/file",
user_error => \&user_error,
user_output => \&user_output,
user_func => \&user_func
);
$mce->run;
=head2 SYNTAX for USER_TASKS and TASK_END
This option takes an array of tasks. Each task allows up to 17 options.
The init_relay, input_data, RS, and use_slurpio options may be defined
inside the first task or at the top level, otherwise ignored under
other sub-tasks.
max_workers, chunk_size, input_data, interval, sequence,
bounds_only, user_args, user_begin, user_end, user_func,
gather, task_end, task_name, use_slurpio, use_threads,
init_relay, RS
Sequence and chunk_size were added in 1.3. User_args was introduced in 1.4.
Name and input_data are new options allowed in 1.5. In addition, one can
specify task_end at the top level. Task_end also receives 2 additional
arguments $task_id and $task_name (shown below).
Options not specified here will default to the same option specified at the
top level. The task_end option is called by the manager process when all
workers for that sub-task have completed processing.
Forking and threading can be intermixed among tasks unless running Cygwin.
The run method will continue running until all workers have completed
processing.
use threads;
use threads::shared;
use MCE;
sub parallel_task1 { sleep 2; }
sub parallel_task2 { sleep 1; }
my $mce = MCE->new(
task_end => sub {
my ($mce, $task_id, $task_name) = @_;
print "Task [$task_id -- $task_name] completed processing\n";
},
user_tasks => [{
task_name => 'foo',
max_workers => 2,
user_func => \¶llel_task1,
use_threads => 0 ## Not using threads
},{
task_name => 'bar',
max_workers => 4,
user_func => \¶llel_task2,
use_threads => 1 ## Yes, threads
}]
);
$mce->run;
-- Output
Task [1 -- bar] completed processing
Task [0 -- foo] completed processing
=head1 DEFAULT INPUT SCALAR
Beginning with MCE 1.5, the input scalar $_ is localized prior to calling
user_func for input_data and sequence of numbers. The following applies.
=over 3
=item use_slurpio => 1
$_ is a reference to the buffer e.g. $_ = \$_buffer;
$_ is a reference regardless of whether chunk_size is 1 or greater
user_func => sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
print ${ $_ }; ## $_ is same as $chunk_ref
}
=item chunk_size is greater than 1, use_slurpio => 0
$_ is a reference to an array. $_ = \@_records; $_ = \@_seq_n;
$_ is same as $chunk_ref or $_[CHUNK]
user_func => sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
for my $row ( @{ $_ } ) {
print $row, "\n";
}
}
use MCE const => 1;
user_func => sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
for my $row ( @{ $_[CHUNK] } ) {
print $row, "\n";
}
}
=item chunk_size equals 1, use_slurpio => 0
$_ contains the actual value. $_ = $_buffer; $_ = $seq_n;
## Note that $_ and $chunk_ref are not the same below.
## $chunk_ref is a reference to an array.
user_func => sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
print $_, "\n; ## Same as $chunk_ref->[0];
}
$mce->foreach("/path/to/file", sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
print $_; ## Same as $chunk_ref->[0];
});
## However, that is not the case for the forseq method.
## Both $_ and $n_seq are the same when chunk_size => 1.
$mce->forseq([ 1, 9 ], sub {
# my ($mce, $n_seq, $chunk_id) = @_;
print $_, "\n"; ## Same as $n_seq
});
Sequence can also be specified using an array reference. The below is the
same as the example afterwards.
$mce->forseq( { begin => 10, end => 40, step => 2 }, ... );
The code block receives an array containing the next 5 sequences. Chunk 1
(chunk_id 1) contains 10,12,14,16,18. $n_seq is a reference to an array,
same as $_, due to chunk_size being greater than 1.
$mce->forseq( [ 10, 40000, 2 ], { chunk_size => 5 }, sub {
# my ($mce, $n_seq, $chunk_id) = @_;
my @result;
for my $n ( @{ $_ } ) {
... do work, append to result for 5
}
... do something with result afterwards
});
=back
=head1 METHODS for the MANAGER PROCESS and WORKERS
The methods listed below are callable by the main process and workers.
=head2 MCE->abort ( void )
=head2 $mce->abort ( void )
The 'abort' method is applicable when processing input_data only. This
causes all workers to abort after processing the current chunk.
Workers write the next offset position to the queue socket for the next
available worker. In essence, the 'abort' method writes the last offset
position. Workers, on requesting the next offset position, will think
the end of input_data has been reached and leave the chunking loop.
MCE->abort;
$mce->abort;
=head2 MCE->chunk_id ( void )
=head2 $mce->chunk_id ( void )
Returns the chunk_id for the current chunk. The value starts at 1. Chunking
applies to input_data or sequence. The value is 0 for the manager process.
my $chunk_id = MCE->chunk_id;
my $chunk_id = $mce->chunk_id;
=head2 MCE->chunk_size ( void )
=head2 $mce->chunk_size ( void )
Getter method for chunk_size used by MCE.
my $chunk_size = MCE->chunk_size;
my $chunk_size = $mce->chunk_size;
=head2 MCE->do ( 'callback_func' [, $arg1, ... ] )
=head2 $mce->do ( 'callback_func' [, $arg1, ... ] )
MCE serializes data transfers from a worker process via helper functions
do & sendto to the manager process. The callback function can optionally
return a reply. Support for calling by the manager process was enabled
in MCE 1.839.
[ $reply = ] MCE->do('callback' [, $arg1, ... ]);
Passing args to a callback function using references & scalar.
sub callback {
my ($array_ref, $hash_ref, $scalar_ref, $scalar) = @_;
...
}
MCE->do('main::callback', \@a, \%h, \$s, 'foo');
MCE->do('callback', \@a, \%h, \$s, 'foo');
MCE knows if wanting a void, list, hash, or a scalar return value.
MCE->do('callback' [, $arg1, ... ]);
my @array = MCE->do('callback' [, $arg1, ... ]);
my %hash = MCE->do('callback' [, $arg1, ... ]);
my $scalar = MCE->do('callback' [, $arg1, ... ]);
=head2 MCE->freeze ( $object_ref )
=head2 $mce->freeze ( $object_ref )
Calls the internal freeze method to serialize an object. The default
serialization routines are handled by Sereal if available or Storable.
my $frozen = MCE->freeze([ 0, 2, 4 ]);
my $frozen = $mce->freeze([ 0, 2, 4 ]);
=head2 MCE->max_retries ( void )
=head2 $mce->max_retries ( void )
Getter method for max_retries used by MCE.
my $max_retries = MCE->max_retries;
my $max_retries = $mce->max_retries;
=head2 MCE->max_workers ( void )
=head2 $mce->max_workers ( void )
Getter method for max_workers used by MCE.
my $max_workers = MCE->max_workers;
my $max_workers = $mce->max_workers;
=head2 MCE->pid ( void )
=head2 $mce->pid ( void )
Returns the Process ID. Threads have thread ID attached to the value.
my $pid = MCE->pid; ## 16180 (pid) ; 16180.2 (pid.tid)
my $pid = $mce->pid;
=head2 MCE->printf ( $format, $list [, ... ] )
=head2 MCE->print ( $list [, ... ] )
=head2 MCE->say ( $list [, ... ] )
=head2 $mce->printf ( $format, $list [, ... ] )
=head2 $mce->print ( $list [, ... ] )
=head2 $mce->say ( $list [, ... ] )
Use the printf, print, and say methods when wanting to serialize output among
workers and the manager process. These are sugar syntax for the sendto method.
These behave similar to the native subroutines in Perl with the exception
that barewords must be passed as a reference and require the comma after
it including file handles.
Say is like print, but implicitly appends a newline.
MCE->printf(\*STDOUT, "%s: %d\n", $name, $age);
MCE->printf($fh, "%s: %d\n", $name, $age);
MCE->printf("%s: %d\n", $name, $age);
MCE->print(\*STDERR, "$error_msg\n");
MCE->print($fh, $log_msg."\n");
MCE->print("$output_msg\n");
MCE->say(\*STDERR, $error_msg);
MCE->say($fh, $log_msg);
MCE->say($output_msg);
Caveat: Use the following syntax when passing a reference not a glob or file
handle. Otherwise, MCE will error indicating the first argument is not a glob
reference.
MCE->print(\*STDOUT, \@array, "\n");
MCE->print("", \@array, "\n"); ## ok
Sending to C<IO::All> { File, Pipe, STDIO } is supported since MCE 1.845.
use IO::All;
my $out = io->stdout;
my $err = io->stderr;
MCE->printf($out, "%s\n", "sent to stdout");
MCE->printf($err, "%s\n", "sent to stderr");
MCE->print($out, "sent to stdout\n");
MCE->print($err, "sent to stderr\n");
MCE->say($out, "sent to stdout");
MCE->say($err, "sent to stderr");
=head2 MCE->sess_dir ( void )
=head2 $mce->sess_dir ( void )
Returns the session directory used by the MCE instance. This is defined
during spawning and removed during shutdown.
my $sess_dir = MCE->sess_dir;
my $sess_dir = $mce->sess_dir;
=head2 MCE->task_id ( void )
=head2 $mce->task_id ( void )
Returns the task ID. This applies to the user_tasks option (starts at 0).
my $task_id = MCE->task_id;
my $task_id = $mce->task_id;
=head2 MCE->task_name ( void )
=head2 $mce->task_name ( void )
Returns the task_name value specified via the task_name option when
configuring MCE.
my $task_name = MCE->task_name;
my $task_name = $mce->task_name;
=head2 MCE->task_wid ( void )
=head2 $mce->task_wid ( void )
Returns the task worker ID (applies to user_tasks). The value starts at 1 per
each task configured within user_tasks. The value is 0 for the manager process.
my $task_wid = MCE->task_wid;
my $task_wid = $mce->task_wid;
=head2 MCE->thaw ( $frozen )
=head2 $mce->thaw ( $frozen )
Calls the internal thaw method to un-serialize the frozen object.
my $object_ref = MCE->thaw($frozen);
my $object_ref = $mce->thaw($frozen);
=head2 MCE->tmp_dir ( void )
=head2 $mce->tmp_dir ( void )
Returns the temporary directory used by MCE.
my $tmp_dir = MCE->tmp_dir;
my $tmp_dir = $mce->tmp_dir;
=head2 MCE->user_args ( void )
=head2 $mce->user_args ( void )
Returns the arguments specified via the user_args option.
my ($arg1, $arg2, $arg3) = MCE->user_args;
my ($arg1, $arg2, $arg3) = $mce->user_args;
=head2 MCE->wid ( void )
=head2 $mce->wid ( void )
Returns the MCE worker ID. Starts at 1 per each MCE instance. The value is
0 for the manager process.
my $wid = MCE->wid;
my $wid = $mce->wid;
=head1 METHODS for the MANAGER PROCESS only
Methods listed below are callable by the main process only.
=head2 MCE->forchunk ( $input_data [, { options } ], sub { ... } )
=head2 MCE->foreach ( $input_data [, { options } ], sub { ... } )
=head2 MCE->forseq ( $sequence_spec [, { options } ], sub { ... } )
=head2 $mce->forchunk ( $input_data [, { options } ], sub { ... } )
=head2 $mce->foreach ( $input_data [, { options } ], sub { ... } )
=head2 $mce->forseq ( $sequence_spec [, { options } ], sub { ... } )
Forchunk, foreach, and forseq are sugar methods and described in
L<MCE::Candy>. Stubs exist in MCE which load MCE::Candy automatically.
=head2 MCE->process ( $input_data [, { options } ] )
=head2 $mce->process ( $input_data [, { options } ] )
The process method will spawn workers automatically if not already spawned.
It will set input_data => $input_data. It calls run(0) to not auto-shutdown
workers. Specifying options is optional.
Allowable options { key => value, ... } are:
chunk_size input_data job_delay spawn_delay submit_delay
flush_file flush_stderr flush_stdout stderr_file stdout_file
on_post_exit on_post_run sequence user_args user_begin user_end
user_func user_error user_output use_slurpio RS
Options remain persistent going forward unless changed. Setting user_begin,
user_end, or user_func will cause already spawned workers to shut down and
re-spawn automatically. Therefore, define these during instantiation.
The below will cause workers to re-spawn after running.
my $mce = MCE->new( max_workers => 'auto' );
$mce->process( {
user_begin => sub { ## connect to DB },
user_func => sub { ## process each row },
user_end => sub { ## close handle to DB },
}, \@input_data );
$mce->process( {
user_begin => sub { ## connect to DB },
user_func => sub { ## process each file },
user_end => sub { ## close handle to DB },
}, "/list/of/files" );
Do the following if wanting workers to persist between jobs.
use MCE max_workers => 'auto';
my $mce = MCE->new(
user_begin => sub { ## connect to DB },
user_func => sub { ## process each chunk or row or host },
user_end => sub { ## close handle to DB },
);
$mce->spawn; ## Spawn early if desired
$mce->process("/one/very_big_file/_mce_/will_chunk_in_parallel");
$mce->process(\@array_of_files_to_grep);
$mce->process("/path/to/host/list");
$mce->process($array_ref);
$mce->process($array_ref, { stdout_file => $output_file });
## This was not allowed before. Fixed in 1.415.
$mce->process({ sequence => { begin => 10, end => 90, step 2 } });
$mce->process({ sequence => [ 10, 90, 2 ] });
$mce->shutdown;
=head2 MCE->relay_final ( void )
=head2 $mce->relay_final ( void )
The relay methods are described in L<MCE::Relay>. Relay capabilities are
enabled by specifying the C<init_relay> MCE option.
=head2 MCE->restart_worker ( void )
=head2 $mce->restart_worker ( void )
One can restart a worker who has died or exited. The job never ends below
due to restarting each time. Recommended is to call MCE->exit or $mce->exit
instead of the native exit function for better handling, especially under
the Windows environment.
The $e->{wid} argument is no longer necessary starting with the 1.5 release.
Press [ctrl-c] to terminate the script.
my $mce = MCE->new(
on_post_exit => sub {
my ($mce, $e) = @_;
print "$e->{wid}: $e->{pid}: status $e->{status}: $e->{msg}";
# $mce->restart_worker($e->{wid}); ## MCE-1.415 and below
$mce->restart_worker; ## MCE-1.500 and above
},
user_begin => sub {
my ($mce, $task_id, $task_name) = @_;
## Not interested in die messages going to STDERR,
## because the die handler calls MCE->exit(255, $_[0]).
close STDERR;
},
user_tasks => [{
max_workers => 5,
user_func => sub {
my ($mce) = @_; sleep MCE->wid;
MCE->exit(3, "exited from " . MCE->wid . "\n");
}
},{
max_workers => 4,
user_func => sub {
my ($mce) = @_; sleep MCE->wid;
die("died from " . MCE->wid . "\n");
}
}]
);
$mce->run;
-- Output
1: PID_85388: status 3: exited from 1
2: PID_85389: status 3: exited from 2
1: PID_85397: status 3: exited from 1
3: PID_85390: status 3: exited from 3
1: PID_85399: status 3: exited from 1
4: PID_85391: status 3: exited from 4
2: PID_85398: status 3: exited from 2
1: PID_85401: status 3: exited from 1
5: PID_85392: status 3: exited from 5
1: PID_85404: status 3: exited from 1
6: PID_85393: status 255: died from 6
3: PID_85400: status 3: exited from 3
2: PID_85403: status 3: exited from 2
1: PID_85406: status 3: exited from 1
7: PID_85394: status 255: died from 7
1: PID_85410: status 3: exited from 1
8: PID_85395: status 255: died from 8
4: PID_85402: status 3: exited from 4
2: PID_85409: status 3: exited from 2
1: PID_85412: status 3: exited from 1
9: PID_85396: status 255: died from 9
3: PID_85408: status 3: exited from 3
1: PID_85416: status 3: exited from 1
...
=head2 MCE->run ( [ $auto_shutdown [, { options } ] ] )
=head2 $mce->run ( [ $auto_shutdown [, { options } ] ] )
The run method, by default, spawns workers, processes once, and shuts down
afterwards. Specify 0 for $auto_shutdown when wanting workers to persist
after running (default 1).
Specifying options is optional. Valid options are the same as for the
process method.
my $mce = MCE->new( ... );
## Disables auto-shutdown
$mce->run(0);
=head2 MCE->send ( $data_ref )
=head2 $mce->send ( $data_ref )
The 'send' method is useful when wanting to spawn workers early to minimize
memory consumption and afterwards send data individually to each worker. One
cannot send more than the total workers spawned. Workers store the received
data as $mce->{user_data}.
The data which can be sent is restricted to an ARRAY, HASH, or PDL reference.
Workers begin processing immediately after receiving data. Workers set
$mce->{user_data} to undef after processing. One cannot specify input_data,
sequence, or user_tasks when using the "send" method.
Passing any options e.g. run(0, { options }) is ignored due to workers running
immediately after receiving user data. There is no guarantee to which worker
will receive data first. It depends on which worker is available awaiting data.
use MCE;
my $mce = MCE->new(
max_workers => 5,
user_func => sub {
my ($mce) = @_;
my $data = $mce->{user_data};
my $first_name = $data->{first_name};
print MCE->wid, ": Hello from $first_name\n";
}
);
$mce->spawn; ## Optional, send will spawn if necessary.
$mce->send( { first_name => "Theresa" } );
$mce->send( { first_name => "Francis" } );
$mce->send( { first_name => "Padre" } );
$mce->send( { first_name => "Anthony" } );
$mce->run; ## Wait for workers to complete processing.
-- Output
2: Hello from Theresa
5: Hello from Anthony
3: Hello from Francis
4: Hello from Padre
=head2 MCE->shutdown ( void )
=head2 $mce->shutdown ( void )
The run method will automatically spawn workers, run once, and shutdown workers
automatically. Workers persist after running below. Shutdown may be called as
needed or prior to exiting.
my $mce = MCE->new( ... );
$mce->spawn;
$mce->process(\@input_data_1); ## Processing multiple arrays
$mce->process(\@input_data_2);
$mce->process(\@input_data_n);
$mce->shutdown;
$mce->process('input_file_1'); ## Processing multiple files
$mce->process('input_file_2');
$mce->process('input_file_n');
$mce->shutdown;
=head2 MCE->spawn ( void )
=head2 $mce->spawn ( void )
Workers are normally spawned automatically. The spawn method allows one to
spawn workers early if so desired.
my $mce = MCE->new( ... );
$mce->spawn;
=head2 MCE->status ( void )
=head2 $mce->status ( void )
The greatest exit status is saved among workers while running. Look at
the on_post_exit or on_post_run options for callback support.
my $mce = MCE->new( ... );
$mce->run;
my $exit_status = $mce->status;
=head1 METHODS for WORKERS only
Methods listed below are callable by workers only.
=head2 MCE->exit ( [ $status [, $message [, $id ] ] ] )
=head2 $mce->exit ( [ $status [, $message [, $id ] ] ] )
A worker exits from MCE entirely. $id (optional) can be used for passing the
primary key or a string along with the message. Look at the on_post_exit
or on_post_run options for callback support.
MCE->exit; ## default 0
MCE->exit(1);
MCE->exit(2, 'chunk failed', $chunk_id);
MCE->exit(0, 'msg_foo', 'id_1000');
=head2 MCE->gather ( $arg1, [, $arg2, ... ] )
=head2 $mce->gather ( $arg1, [, $arg2, ... ] )
A worker can submit data to the location specified via the gather option by
calling this method. See L<MCE::Flow> and L<MCE::Loop> for additional use-case.
use MCE;
my @hosts = qw(
hosta hostb hostc hostd hoste
);
my $mce = MCE->new(
chunk_size => 1, max_workers => 3,
user_func => sub {
# my ($mce, $chunk_ref, $chunk_id) = @_;
my ($output, $error, $status); my $host = $_;
## Do something with $host;
$output = "Worker ". MCE->wid .": Hello from $host";
if (MCE->chunk_id % 3 == 0) {
## Simulating an error condition
local $? = 1; $status = $?;
$error = "Error from $host"
}
else {
$status = 0;
}
## Ensure unique keys (key, value) when gathering to a
## hash.
MCE->gather("$host.out", $output, "$host.sta", $status);
MCE->gather("$host.err", $error) if (defined $error);
}
);
my %h; $mce->process(\@hosts, { gather => \%h });
foreach my $host (@hosts) {
print $h{"$host.out"}, "\n";
print $h{"$host.err"}, "\n" if (exists $h{"$host.err"});
print "Exit status: ", $h{"$host.sta"}, "\n\n";
}
-- Output
Worker 2: Hello from hosta
Exit status: 0
Worker 1: Hello from hostb
Exit status: 0
Worker 3: Hello from hostc
Error from hostc
Exit status: 1
Worker 2: Hello from hostd
Exit status: 0
Worker 1: Hello from hoste
Exit status: 0
=head2 MCE->last ( void )
=head2 $mce->last ( void )
Worker leaves the chunking loop or user_func block immediately. Callable from
inside foreach, forchunk, forseq, and user_func.
use MCE;
my $mce = MCE->new(
max_workers => 5
);
my @list = (1 .. 80);
$mce->forchunk(\@list, { chunk_size => 2 }, sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
MCE->last if ($chunk_id > 4);
my @output = ();
foreach my $rec ( @{ $chunk_ref } ) {
push @output, $rec, "\n";
}
MCE->print(@output);
});
-- Output (each chunk above consists of 2 elements)
3
4
1
2
7
8
5
6
=head2 MCE->next ( void )
=head2 $mce->next ( void )
Worker starts the next iteration of the chunking loop. Callable from inside
foreach, forchunk, forseq, and user_func.
use MCE;
my $mce = MCE->new(
max_workers => 5
);
my @list = (1 .. 80);
$mce->forchunk(\@list, { chunk_size => 4 }, sub {
my ($mce, $chunk_ref, $chunk_id) = @_;
MCE->next if ($chunk_id < 20);
my @output = ();
foreach my $rec ( @{ $chunk_ref } ) {
push @output, $rec, "\n";
}
MCE->print(@output);
});
-- Output (each chunk above consists of 4 elements)
77
78
79
80
=head2 MCE::relay { code }
=head2 MCE->relay ( sub { code } )
=head2 MCE->relay_recv ( void )
=head2 $mce->relay ( sub { code } )
=head2 $mce->relay_recv ( void )
The relay methods are described in L<MCE::Relay>. Relay capabilities are
enabled by specifying the C<init_relay> MCE option.
=head2 MCE->sendto ( $to, $arg1, ... )
=head2 $mce->sendto ( $to, $arg1, ... )
The sendto method is called by workers for serializing data to standard output,
standard error, or end of file. The action is done by the manager process.
Release 1.00x supported 1 data argument, not more.
MCE->sendto('file', \@array, '/path/to/file');
MCE->sendto('file', \$scalar, '/path/to/file');
MCE->sendto('file', $scalar, '/path/to/file');
MCE->sendto('STDERR', \@array);
MCE->sendto('STDERR', \$scalar);
MCE->sendto('STDERR', $scalar);
MCE->sendto('STDOUT', \@array);
MCE->sendto('STDOUT', \$scalar);
MCE->sendto('STDOUT', $scalar);
Release 1.100 added the ability to pass multiple arguments. Notice the syntax
change for sending to a file. Passing a reference to an array is no longer
necessary.
MCE->sendto('file:/path/to/file', $arg1 [, $arg2, ... ]);
MCE->sendto('STDERR', $arg1 [, $arg2, ... ]);
MCE->sendto('STDOUT', $arg1 [, $arg2, ... ]);
MCE->sendto('STDOUT', @a, "\n", %h, "\n", $s, "\n");
To retain 1.00x compatibility, sendto outputs the content when a single data
reference is specified. Otherwise, the reference for \@array or \$scalar is
shown in 1.500, not the content.
MCE->sendto('STDERR', \@array); ## 1.00x behavior, content
MCE->sendto('STDOUT', \$scalar);
MCE->sendto('file:/path/to/file', \@array);
## Output matches the print statement
MCE->sendto(\*STDERR, \@array); ## 1.500 behavior, reference
MCE->sendto(\*STDOUT, \$scalar);
MCE->sendto($fh, \@array);
MCE->sendto('STDOUT', \@array, "\n", \$scalar, "\n");
print {*STDOUT} \@array, "\n", \$scalar, "\n";
MCE 1.500 added support for sending to a glob reference, file descriptor, and
file handle.
MCE->sendto(\*STDERR, "foo\n", \@array, \$scalar, "\n");
MCE->sendto('fd:2', "foo\n", \@array, \$scalar, "\n");
MCE->sendto($fh, "foo\n", \@array, \$scalar, "\n");
=head2 MCE->sync ( void )
=head2 $mce->sync ( void )
A barrier sync operation means any worker must stop at this point until all
workers reach this barrier. Barrier syncing is useful for many computer
algorithms.
Barrier synchronization is supported for task 0 only or omitting user_tasks.
All workers assigned task_id 0 must call sync whenever barrier syncing.
use MCE;
sub user_func {
my ($mce) = @_;
my $wid = MCE->wid;
MCE->sendto("STDOUT", "a: $wid\n"); ## MCE 1.0+
MCE->sync;
MCE->sendto(\*STDOUT, "b: $wid\n"); ## MCE 1.5+
MCE->sync;
MCE->print("c: $wid\n"); ## MCE 1.5+
MCE->sync;
return;
}
my $mce = MCE->new(
max_workers => 4, user_func => \&user_func
)->run;
-- Output (without barrier synchronization)
a: 1
a: 2
b: 1
b: 2
c: 1
c: 2
a: 3
b: 3
c: 3
a: 4
b: 4
c: 4
-- Output (with barrier synchronization)
a: 1
a: 2
a: 4
a: 3
b: 2
b: 1
b: 3
b: 4
c: 1
c: 4
c: 2
c: 3
Consider the following example. The MCE->sync operation is done inside a loop
along with MCE->do. A stall may occur for workers calling sync the 2nd or 3rd
time while other workers are sending results via MCE->do or MCE->sendto.
It requires another semaphore lock in MCE to solve this which was not done in
order to keep resources low. Therefore, please keep this in mind when mixing
MCE->sync with MCE->do or output serialization methods inside a loop.
sub user_func {
my ($mce) = @_;
my @result;
for (1 .. 3) {
... compute algorithm ...
MCE->sync;
... compute algorithm ...
MCE->sync;
MCE->do('aggregate_result', \@result); ## or MCE->sendto
MCE->sync; ## The sync operation is also needed here to
## prevent MCE from stalling.
}
}
=head2 MCE->yield ( void )
=head2 $mce->yield ( void )
There may be on occasion when the MCE driven app is too fast. The interval
option combined with the yield method, both introduced with MCE 1.5, allows
one to throttle the app. It adds a "grace" factor to the design.
A use case is an app configured with 100 workers running on a 24 logical way
box. Data is polled from a database containing over 2.5 million rows. Workers
chunk away at 300 rows per chunk performing SNMP gets (300 sockets per worker)
polling 25 metrics from each device. With this scenario, the load on the box
may rise beyond 90+. In addition, IP_Tables may reach its contention point
causing the entire application to fail.
The scenario above is solved by simply having workers yield among themselves
in a synchronized fashion. A delay of 0.007 seconds between intervals is all
that's needed. The load on the box will hover between 23 ~ 27 for the duration
of the run. Polling completes in under 17 minutes time. This is quite fast
considering the app polls 62.5 million metrics combined. The math equates
to 3,676,470 per minute or rather 61,275 per second from a single box.
## Both max_nodes and node_id are optional (default 1).
interval => {
delay => 0.007, max_nodes => $max_nodes, node_id => $node_id
}
A 4 node setup can poll 10 million devices without the additional overhead of a
distribution agent. The difference between the 4 nodes are simply node_id and
the where clause used to query the database. The mac addresses are random such
that the data divides equally to any power of 2. The distribution key lies in
the mac address itself. In fact, the 2nd character from the right is sufficient
for maximizing on the power of randomness for equal distribution.
Query NodeID 1: ... AND substr(MAC, -2, 1) IN ('0', '1', '2', '3')
Query NodeID 2: ... AND substr(MAC, -2, 1) IN ('4', '5', '6', '7')
Query NodeID 3: ... AND substr(MAC, -2, 1) IN ('8', '9', 'A', 'B')
Query NodeID 4: ... AND substr(MAC, -2, 1) IN ('C', 'D', 'E', 'F')
Below, the user_tasks is configured to simulate 4 nodes. This demonstration
uses 2 workers to minimize the output size. Input is from the sequence option.
use Time::HiRes qw(time);
use MCE;
my $d = shift || 0.1;
local $| = 1;
sub create_task {
my ($node_id) = @_;
my $seq_size = 6;
my $seq_start = ($node_id - 1) * $seq_size + 1;
my $seq_end = $seq_start + $seq_size - 1;
return {
max_workers => 2, sequence => [ $seq_start, $seq_end ],
interval => { delay => $d, max_nodes => 4, node_id => $node_id }
};
}
sub user_begin {
my ($mce, $task_id, $task_name) = @_;
## The yield method causes this worker to wait for its next time
## interval slot before running. Yield has no effect without the
## 'interval' option.
## Yielding is beneficial inside a user_begin block. A use case
## is staggering database connections among workers in order
## to not impact the DB server.
MCE->yield;
MCE->printf(
"Node %2d: %0.5f -- Worker %2d: %12s -- Started\n",
MCE->task_id + 1, time, MCE->task_wid, ''
);
return;
}
{
my $prev_time = time;
sub user_func {
my ($mce, $seq_n, $chunk_id) = @_;
## Yield simply waits for the next time interval.
MCE->yield;
## Calculate how long this worker has waited.
my $curr_time = time;
my $time_waited = $curr_time - $prev_time;
$prev_time = $curr_time;
MCE->printf(
"Node %2d: %0.5f -- Worker %2d: %12.5f -- Seq_N %3d\n",
MCE->task_id + 1, time, MCE->task_wid, $time_waited, $seq_n
);
return;
}
}
## Simulate a 4 node environment passing node_id to create_task.
print "Node_ID Current_Time Worker_ID Time_Waited Comment\n";
MCE->new(
user_begin => \&user_begin,
user_func => \&user_func,
user_tasks => [
create_task(1),
create_task(2),
create_task(3),
create_task(4)
]
)->run;
-- Output (notice Current_Time below, stays 0.10 apart)
Node_ID Current_Time Worker_ID Time_Waited Comment
Node 1: 1374807976.74634 -- Worker 1: -- Started
Node 2: 1374807976.84634 -- Worker 1: -- Started
Node 3: 1374807976.94638 -- Worker 1: -- Started
Node 4: 1374807977.04639 -- Worker 1: -- Started
Node 1: 1374807977.14634 -- Worker 2: -- Started
Node 2: 1374807977.24640 -- Worker 2: -- Started
Node 3: 1374807977.34649 -- Worker 2: -- Started
Node 4: 1374807977.44657 -- Worker 2: -- Started
Node 1: 1374807977.54636 -- Worker 1: 0.90037 -- Seq_N 1
Node 2: 1374807977.64638 -- Worker 1: 1.00040 -- Seq_N 7
Node 3: 1374807977.74642 -- Worker 1: 1.10043 -- Seq_N 13
Node 4: 1374807977.84643 -- Worker 1: 1.20045 -- Seq_N 19
Node 1: 1374807977.94636 -- Worker 2: 1.30037 -- Seq_N 2
Node 2: 1374807978.04638 -- Worker 2: 1.40040 -- Seq_N 8
Node 3: 1374807978.14641 -- Worker 2: 1.50042 -- Seq_N 14
Node 4: 1374807978.24644 -- Worker 2: 1.60045 -- Seq_N 20
Node 1: 1374807978.34628 -- Worker 1: 0.79996 -- Seq_N 3
Node 2: 1374807978.44631 -- Worker 1: 0.79996 -- Seq_N 9
Node 3: 1374807978.54634 -- Worker 1: 0.79996 -- Seq_N 15
Node 4: 1374807978.64636 -- Worker 1: 0.79997 -- Seq_N 21
Node 1: 1374807978.74628 -- Worker 2: 0.79996 -- Seq_N 4
Node 2: 1374807978.84632 -- Worker 2: 0.79997 -- Seq_N 10
Node 3: 1374807978.94634 -- Worker 2: 0.79996 -- Seq_N 16
Node 4: 1374807979.04636 -- Worker 2: 0.79996 -- Seq_N 22
Node 1: 1374807979.14628 -- Worker 1: 0.80001 -- Seq_N 5
Node 2: 1374807979.24631 -- Worker 1: 0.80000 -- Seq_N 11
Node 3: 1374807979.34634 -- Worker 1: 0.80001 -- Seq_N 17
Node 4: 1374807979.44636 -- Worker 1: 0.80000 -- Seq_N 23
Node 1: 1374807979.54628 -- Worker 2: 0.80000 -- Seq_N 6
Node 2: 1374807979.64631 -- Worker 2: 0.80000 -- Seq_N 12
Node 3: 1374807979.74633 -- Worker 2: 0.80000 -- Seq_N 18
Node 4: 1374807979.84636 -- Worker 2: 0.80000 -- Seq_N 24
The interval.pl example above is included with MCE.
=head1 MCE PROGRESS DEMONSTRATIONS
The C<progress> option takes a code block for receiving info on the progress
made while processing input data; e.g. C<input_data> or C<sequence>. To make
this work, one provides the C<progress> option a closure block like so, passing
along the size of the input_data; e.g C<scalar @array> or C<-s /path/to/file>.
Current API available since 1.813.
A worker, upon completing processing its chunk, notifies the manager-process
with the size of the chunk. That could be the number of rows or literally the
size of the chunk when processing an input file. The manager-process accumulates
the size before calling the code block associated with the C<progress> option.
When running many tasks simultaneously, via C<user_tasks>, the call is initiated
by workers at level 0 only or rather the first task, not shown here.
use Time::HiRes 'sleep';
use MCE;
sub make_progress {
my ($total_size) = @_;
return sub {
my ($completed_size) = @_;
printf "%0.1f%%\n", $completed_size / $total_size * 100;
};
}
my @input = (1..150);
MCE->new(
chunk_size => 10,
max_workers => 4,
input_data => \@input,
progress => make_progress( scalar @input ),
user_func => sub { sleep 1.5 }
)->run();
-- Output
6.7%
13.3%
20.0%
26.7%
33.3%
40.0%
46.7%
53.3%
60.0%
66.7%
73.3%
80.0%
86.7%
93.3%
100.0%
Next is the code using L<MCE::Flow> and L<ProgressBar::Stack> to do the
same thing, practically.
use Time::HiRes 'sleep';
use ProgressBar::Stack;
use MCE::Flow;
sub make_progress {
my ($total_size) = @_;
init_progress();
return sub {
my ($completed_size) = @_;
update_progress sprintf("%0.1f", $completed_size / $total_size * 100);
};
}
my @input = (1..150);
MCE::Flow->init(
chunk_size => 10,
max_workers => 4,
progress => make_progress( scalar @input )
);
MCE::Flow->run( sub { sleep 1.5 }, \@input );
MCE::Flow->finish();
print "\n";
-- Output
[################ ] 80.0% ETA: 0:01
For sequence of numbers, using the C<sequence> option, one must account for
C<step_size>, typically set to C<1> automatically.
use Time::HiRes 'sleep';
use MCE;
sub make_progress {
my ($total_size) = @_;
return sub {
my ($completed_size) = @_;
printf "%0.1f%%\n", $completed_size / $total_size * 100;
};
}
MCE->new(
chunk_size => 10,
max_workers => 4,
sequence => [ 1, 100, 2 ],
progress => make_progress( int( 100 / 2 + 0.5 ) ),
user_func => sub { sleep 1.5 }
)->run();
-- Output
20.0%
40.0%
60.0%
80.0%
100.0%
Changing C<chunk_size> to C<1> means workers notify the manager process more
often, thus increasing granularity. Take a look at the output.
2.0%
4.0%
6.0%
8.0%
10.0%
...
92.0%
94.0%
96.0%
98.0%
100.0%
Here is the same thing using L<MCE::Flow> together with L<ProgressBar::Stack>.
use Time::HiRes 'sleep';
use ProgressBar::Stack;
use MCE::Flow;
sub make_progress {
my ($total_size) = @_;
init_progress();
return sub {
my ($completed_size) = @_;
update_progress sprintf("%0.1f", $completed_size / $total_size * 100);
};
}
MCE::Flow->init(
chunk_size => 1,
max_workers => 4,
progress => make_progress( int( 100 / 2 + 0.5 ) )
);
MCE::Flow->run_seq( sub { sleep 0.5 }, 1, 100, 2 );
MCE::Flow->finish();
print "\n";
-- Output
[######### ] 48.0% ETA: 0:03
For files and file handles, workers send the actual size of the data read
versus counting rows.
use Time::HiRes 'sleep';
use MCE;
sub make_progress {
my ($total_size) = @_;
return sub {
my ($completed_size) = @_;
printf "%0.1f%%\n", $completed_size / $total_size * 100;
};
}
my $input_file = "/path/to/input_file.txt";
MCE->new(
chunk_size => 5,
max_workers => 4,
input_data => $input_file,
progress => make_progress( -s $input_file ),
user_func => sub { sleep 0.03 }
)->run();
For consistency, here is the example using L<MCE::Flow>, again with
L<ProgressBar::Stack>.
use Time::HiRes 'sleep';
use ProgressBar::Stack;
use MCE::Flow;
sub make_progress {
my ($total_size) = @_;
init_progress();
return sub {
my ($completed_size) = @_;
update_progress sprintf("%0.1f", $completed_size / $total_size * 100);
};
}
my $input_file = "/path/to/input_file.txt";
MCE::Flow->init(
chunk_size => 5,
max_workers => 4,
progress => make_progress( -s $input_file )
);
MCE::Flow->run_file( sub { sleep 0.03 }, $input_file );
MCE::Flow->finish();
The next demonstration processes three arrays consecutively. For this one, MCE
workers persist after running. This needs MCE 1.814 or later to run. Otherwise,
the progress output is not shown in MCE 1.813.
use Time::HiRes 'sleep';
use ProgressBar::Stack;
use MCE;
sub make_progress {
my ($total_size, $message) = @_;
init_progress();
return sub {
my ($completed_size) = @_;
update_progress(
sprintf("%0.1f", $completed_size / $total_size * 100),
$message
);
};
}
my $mce = MCE->new(
chunk_size => 10,
max_workers => 4,
user_func => sub { sleep 0.5 }
)->spawn();
my @a1 = ( 1 .. 200 );
my @a2 = ( 1 .. 500 );
my @a3 = ( 1 .. 300 );
$mce->process({ progress => make_progress(scalar(@a1), "array 1") }, \@a1);
print "\n";
$mce->process({ progress => make_progress(scalar(@a2), "array 2") }, \@a2);
print "\n";
$mce->process({ progress => make_progress(scalar(@a3), "array 3") }, \@a3);
print "\n";
$mce->shutdown;
-- Output
[####################] 100.0% ETA: 0:00 array 1
[####################] 100.0% ETA: 0:00 array 2
[####################] 100.0% ETA: 0:00 array 3
When size is not know, such as reading from C<STDIN>, the only thing one
can do is report the size completed thus far.
# 1 kibibyte equals 1024 bytes
progress => sub {
my ($completed_size) = @_;
printf "%0.1f kibibytes\n", $completed_size / 1024;
}
=head1 SEE ALSO
=over 3
=item * L<MCE::Examples>
=back
=head1 INDEX
L<MCE|MCE>
=head1 AUTHOR
Mario E. Roy, S<E<lt>marioeroy AT gmail DOT comE<gt>>
=cut