MCE::Core(3) User Contributed Perl Documentation MCE::Core(3)
NAME
MCE::Core - Documentation describing the core MCE API
VERSION
This document describes MCE::Core version 1.868
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
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 => 0, ## Default 1
flush_stderr => 0, ## Default 1
flush_stdout => 0, ## Default 1
# 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 = $chunk_ref->[1];
task_end => \&task_end, ## Default undef
# This is called by the manager process after the task
# has completed processing. MCE 1.5 allows this option
# to be specified at the top level.
task_name => 'string', ## Default 'MCE'
# Added in MCE 1.5 and mainly beneficial for user_tasks.
# One may specify a unique name per each sub-task.
# The string is passed as the 3rd arg to task_end.
user_tasks => [ ## Default undef
{ ... }, ## Options for task 0
{ ... }, ## Options for task 1
{ ... }, ## Options for task 2
],
# Takes a list of hash references, each allowing up to 17
# options. All other MCE options are ignored. The init_relay,
# input_data, RS, and use_slurpio options are applicable to
# the first task only.
# max_workers, chunk_size, input_data, interval, sequence,
# bounds_only, user_args, user_begin, user_end, user_func,
# gather, task_end, task_name, use_slurpio, use_threads,
# init_relay, RS
# Options not specified here will default to same option
# specified at the top level.
);
EXPORT_CONST, CONST
There are 3 constants which are exportable. Using the constants in lieu
of 0,1,2 makes it more legible when accessing the user_func arguments
directly.
SELF CHUNK CID - MCE CONSTANTS
Exports SELF => 0, CHUNK => 1, and CID => 2.
use MCE export_const => 1;
use MCE const => 1; ## Shorter; MCE 1.415+
user_func => sub {
# my ($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";
}
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 "Sereal => 0" to use Storable instead.
use MCE Sereal => 0;
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
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
);
$mce->run;
-- Output (child processes)
2: PID_33223: 0: msg_foo: 1002
1: PID_33222: 0: msg_foo: 1001
3: PID_33224: 0: msg_foo: 1003
2: PID_33225: 0: msg_foo: 1002
-- Output (running with threads)
3: TID_3: 0: msg_foo: 1003
2: TID_2: 0: msg_foo: 1002
1: TID_1: 0: msg_foo: 1001
2: TID_4: 0: msg_foo: 1002
SYNTAX for ON_POST_RUN
The on_post_run option, if defined, is executed immediately by the
manager process after running MCE->process or MCE->run. This option
receives an array reference of hashes.
The difference between on_post_exit and on_post_run is that the former
is called immediately whereas the latter is called after all workers
have completed running.
sub on_post_run {
my ($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
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.
"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
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
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 "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 {
$_ += 1;
print MCE->chunk_id, "\n";
};
}
)->run;
print "final: ", MCE::relay_final(), "\n";
-- Output
1
2
Error: chunk 3 failed, retrying chunk attempt # 1
Error: chunk 5 failed, retrying chunk attempt # 1
Error: chunk 3 failed, retrying chunk attempt # 2
Error: chunk 5 failed, retrying chunk attempt # 2
3
4
Error: chunk 5 failed
6
7
final: 6
SYNTAX for USER_BEGIN and USER_END
The user_begin and user_end options, if specified, behave similarly to
awk 'BEGIN { begin } { func } { func } ... END { end }'. These are
called once per worker during each run.
MCE 1.510 passes 2 additional parameters ($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;
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
);
$mce->run;
SYNTAX for USER_FUNC with USE_SLURPIO => 1
Here, a reference to a scalar containing the raw chunk data is
processed.
sub user_func {
my ($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;
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;
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
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.
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
}
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";
}
}
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
});
METHODS for the MANAGER PROCESS and WORKERS
The methods listed below are callable by the main process and workers.
MCE->abort ( void )
$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;
MCE->chunk_id ( void )
$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;
MCE->chunk_size ( void )
$mce->chunk_size ( void )
Getter method for chunk_size used by MCE.
my $chunk_size = MCE->chunk_size;
my $chunk_size = $mce->chunk_size;
MCE->do ( 'callback_func' [, $arg1, ... ] )
$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, ... ]);
MCE->freeze ( $object_ref )
$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 ]);
MCE->max_retries ( void )
$mce->max_retries ( void )
Getter method for max_retries used by MCE.
my $max_retries = MCE->max_retries;
my $max_retries = $mce->max_retries;
MCE->max_workers ( void )
$mce->max_workers ( void )
Getter method for max_workers used by MCE.
my $max_workers = MCE->max_workers;
my $max_workers = $mce->max_workers;
MCE->pid ( void )
$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;
MCE->printf ( $format, $list [, ... ] )
MCE->print ( $list [, ... ] )
MCE->say ( $list [, ... ] )
$mce->printf ( $format, $list [, ... ] )
$mce->print ( $list [, ... ] )
$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 "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");
MCE->sess_dir ( void )
$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;
MCE->task_id ( void )
$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;
MCE->task_name ( void )
$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;
MCE->task_wid ( void )
$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;
MCE->thaw ( $frozen )
$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);
MCE->tmp_dir ( void )
$mce->tmp_dir ( void )
Returns the temporary directory used by MCE.
my $tmp_dir = MCE->tmp_dir;
my $tmp_dir = $mce->tmp_dir;
MCE->user_args ( void )
$mce->user_args ( void )
Returns the arguments specified via the user_args option.
my ($arg1, $arg2, $arg3) = MCE->user_args;
my ($arg1, $arg2, $arg3) = $mce->user_args;
MCE->wid ( void )
$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;
METHODS for the MANAGER PROCESS only
Methods listed below are callable by the main process only.
MCE->forchunk ( $input_data [, { options } ], sub { ... } )
MCE->foreach ( $input_data [, { options } ], sub { ... } )
MCE->forseq ( $sequence_spec [, { options } ], sub { ... } )
$mce->forchunk ( $input_data [, { options } ], sub { ... } )
$mce->foreach ( $input_data [, { options } ], sub { ... } )
$mce->forseq ( $sequence_spec [, { options } ], sub { ... } )
Forchunk, foreach, and forseq are sugar methods and described in
MCE::Candy. Stubs exist in MCE which load MCE::Candy automatically.
MCE->process ( $input_data [, { options } ] )
$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;
MCE->relay_final ( void )
$mce->relay_final ( void )
The relay methods are described in MCE::Relay. Relay capabilities are
enabled by specifying the "init_relay" MCE option.
MCE->restart_worker ( void )
$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
...
MCE->run ( [ $auto_shutdown [, { options } ] ] )
$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);
MCE->send ( $data_ref )
$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
MCE->shutdown ( void )
$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;
MCE->spawn ( void )
$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;
MCE->status ( void )
$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;
METHODS for WORKERS only
Methods listed below are callable by workers only.
MCE->exit ( [ $status [, $message [, $id ] ] ] )
$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');
MCE->gather ( $arg1, [, $arg2, ... ] )
$mce->gather ( $arg1, [, $arg2, ... ] )
A worker can submit data to the location specified via the gather
option by calling this method. See MCE::Flow and 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
MCE->last ( void )
$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
MCE->next ( void )
$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
MCE::relay { code }
MCE->relay ( sub { code } )
MCE->relay_recv ( void )
$mce->relay ( sub { code } )
$mce->relay_recv ( void )
The relay methods are described in MCE::Relay. Relay capabilities are
enabled by specifying the "init_relay" MCE option.
MCE->sendto ( $to, $arg1, ... )
$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");
MCE->sync ( void )
$mce->sync ( void )
A barrier sync operation means any worker must stop at this point until
all workers reach this barrier. Barrier syncing is useful for many
computer algorithms.
Barrier synchronization is supported for task 0 only or omitting
user_tasks. All workers assigned task_id 0 must call sync whenever
barrier syncing.
use MCE;
sub user_func {
my ($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.
}
}
MCE->yield ( void )
$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.
MCE PROGRESS DEMONSTRATIONS
The "progress" option takes a code block for receiving info on the
progress made while processing input data; e.g. "input_data" or
"sequence". To make this work, one provides the "progress" option a
closure block like so, passing along the size of the input_data; e.g
"scalar @array" or "-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 "progress" option.
When running many tasks simultaneously, via "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 / $total_size * 100;
};
}
my @input = (1..150);
MCE->new(
chunk_size => 10,
max_workers => 4,
input_data => \@input,
progress => make_progress( scalar @input ),
user_func => sub { sleep 1.5 }
)->run();
-- Output
6.7%
13.3%
20.0%
26.7%
33.3%
40.0%
46.7%
53.3%
60.0%
66.7%
73.3%
80.0%
86.7%
93.3%
100.0%
Next is the code using MCE::Flow and ProgressBar::Stack to do the same
thing, practically.
use Time::HiRes 'sleep';
use ProgressBar::Stack;
use MCE::Flow;
sub make_progress {
my ($total_size) = @_;
init_progress();
return sub {
my ($completed_size) = @_;
update_progress sprintf("%0.1f", $completed_size / $total_size * 100);
};
}
my @input = (1..150);
MCE::Flow->init(
chunk_size => 10,
max_workers => 4,
progress => make_progress( scalar @input )
);
MCE::Flow->run( sub { sleep 1.5 }, \@input );
MCE::Flow->finish();
print "\n";
-- Output
[################ ] 80.0% ETA: 0:01
For sequence of numbers, using the "sequence" option, one must account
for "step_size", typically set to 1 automatically.
use Time::HiRes 'sleep';
use MCE;
sub make_progress {
my ($total_size) = @_;
return sub {
my ($completed_size) = @_;
printf "%0.1f%%\n", $completed_size / $total_size * 100;
};
}
MCE->new(
chunk_size => 10,
max_workers => 4,
sequence => [ 1, 100, 2 ],
progress => make_progress( int( 100 / 2 + 0.5 ) ),
user_func => sub { sleep 1.5 }
)->run();
-- Output
20.0%
40.0%
60.0%
80.0%
100.0%
Changing "chunk_size" to 1 means workers notify the manager process
more often, thus increasing granularity. Take a look at the output.
2.0%
4.0%
6.0%
8.0%
10.0%
...
92.0%
94.0%
96.0%
98.0%
100.0%
Here is the same thing using MCE::Flow together with
ProgressBar::Stack.
use Time::HiRes 'sleep';
use ProgressBar::Stack;
use MCE::Flow;
sub make_progress {
my ($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 $input_file ),
user_func => sub { sleep 0.03 }
)->run();
For consistency, here is the example using MCE::Flow, again with
ProgressBar::Stack.
use Time::HiRes 'sleep';
use ProgressBar::Stack;
use MCE::Flow;
sub make_progress {
my ($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";
$mce->shutdown;
-- Output
[####################] 100.0% ETA: 0:00 array 1
[####################] 100.0% ETA: 0:00 array 2
[####################] 100.0% ETA: 0:00 array 3
When size is not know, such as reading from "STDIN", the only thing one
can do is report the size completed thus far.
# 1 kibibyte equals 1024 bytes
progress => sub {
my ($completed_size) = @_;
printf "%0.1f kibibytes\n", $completed_size / 1024;
}
SEE ALSO
o MCE::Examples(3)
INDEX
MCE(3)
AUTHOR
Mario E. Roy, <marioeroyA ATA gmailA DOTA com>
perl v5.28.2 2020-05-11 MCE::Core(3)
mce 1.868.0 - Generated Tue May 12 07:58:47 CDT 2020