new

 $pool = Thread::Pool->new(
  {
   optimize => 'cpu',                            # default: memory
   do => sub { print "doing with @_\n" },        # must have
   pre => sub { print "starting with @_\n",      # default: none
   post => sub { print "stopping with @_\n",     # default: none
   stream => sub { print "streamline with @_\n", # default: none
   monitor => sub { print "monitor with @_\n",   # default: none
   pre_post_monitor_only => 0, # default: 0 = also for "do"
   checkpoint => \&checkpoint,
   frequency => 1000,
   autoshutdown => 1, # default: 1 = yes
   workers => 10,     # default: 1
   maxjobs => 50,     # default: 5 * workers
   minjobs => 5,      # default: maxjobs / 2
  },
  qw(a b c)           # parameters to "pre" and "post" routines
 );

The "new" method returns the Thread::Pool object.

The first input parameter is a reference to a hash that should at least contain the "do" key with a subroutine reference.

The other input parameters are optional. If specified, they are passed to the the "pre" subroutine whenever a new worker is added.

Each time a worker thread is added, the "pre" subroutine (if available) will be called inside the thread. Each time a worker thread is removed, the "post" routine is called. Its return value(s) are saved only if a job ID was requested when removing the thread. Then the result method can be called to obtain the results of the "post" subroutine.

The following field must be specified in the hash reference:

optimize

 optimize => 'cpu', # default: 'memory'
    

The "optimize" field specifies which implementation of the belt will be selected. Currently there are two choices: 'cpu' and 'memory'. By default, the "memory" optimization will be selected if no specific optimization is specified.

You can call the class method optimize to change the default optimization.

do

 do => 'do_the_job',            # assume caller's namespace
    

or:

 do => 'Package::do_the_job',
    

or:

 do => \&SomeOther::do_the_job,
    

or:

 do => sub {print "anonymous sub doing the job\n"},
    

The "do" field specifies the subroutine to be executed for each job. It must be specified as either the name of a subroutine or as a reference to a (anonymous) subroutine.

The specified subroutine should expect the following parameters to be passed:

 1..N  any parameters that were passed with the call to L<job>.
    

Any values that are returned by this subroutine after finishing each job, are accessible with result if a job ID was requested when assigning the job.

The following fields are optional in the hash reference:

pre

 pre => 'prepare_jobs',         # assume caller's namespace
    

or:

 pre => 'Package::prepare_jobs',
    

or:

 pre => \&SomeOther::prepare_jobs,
    

or:

 pre => sub {print "anonymous sub preparing the jobs\n"},
    

The "pre" field specifies the subroutine to be executed each time a new worker thread is started (either when starting the pool, or when new worker threads are added with a call to either add or workers) and once when a "monitor" routine is specified. It must be specified as either the name of a subroutine or as a reference to a (anonymous) subroutine.

The specified subroutine should expect the following parameters to be passed:

 1..N  any additional parameters that were passed with the call to L<new>.
    

You can determine whether the "pre" routine is called for a new worker thread or for a monitoring thread by checking the self or monitor class method inside the "pre" routine.

post

 post => 'cleanup_after_worker',        # assume caller's namespace
    

or:

 post => 'Package::cleanup_after_worker',
    

or:

 post => \&SomeOther::cleanup_after_worker,
    

or:

 post => sub {print "anonymous sub cleaning up after the worker removed\n"},
    

The "post" field specifies the subroutine to be executed each time a worker thread is removed (either when being specifically removed, or when the pool is shutdown specifically or implicitly when the Thread::Pool object is destroyed. It must be specified as either the name of a subroutine or as a reference to a (anonymous) subroutine.

The specified subroutine should expect the following parameters to be passed:

 1..N  any additional parameters that were passed with the call to L<new>.
    

Any values that are returned by this subroutine after closing down the thread, are accessible with the result method, but only if the thread was removed and a job ID was requested.

You can determine whether the "post" routine is called for a new worker thread or for a monitoring thread by checking the self or monitor class method inside the "post" routine.

stream

 stream => 'in_order_of_submit',        # assume caller's namespace
    

or:

 stream => 'Package::in_order_of_submit',
    

or:

 stream => \&SomeOther::in_order_of_submit,
    

or:

 stream => sub {print "anonymous sub called in order of submit\n"},
    

The "stream" field specifies the subroutine to be executed for streaming the results of the "do" routine. If specified, the "stream" routine is called once for the result of each "do" subroutine, but in the order in which the jobs were submitted rather than in the order in which the result were obtained (which is by the very nature of threads, indeterminate).

The specified subroutine should expect the following parameters to be passed:

 1     the Thread::Pool object to which the worker thread belongs.
 2..N  the values that were returned by the "do" subroutine
    

The "stream" routine is executed in any of the threads that are created for the Thread::Pool object. The system attempts to call the "stream" routine in the same thread from which the values are obtained, but when things get out of sync, other threads may stream the result of a job. If you want only one thread to stream all results, use the "monitor" routine.

monitor

 monitor => 'in_order_of_submit',       # assume caller's namespace
    

or:

 monitor => 'Package::in_order_of_submit',
    

or:

 monitor => \&SomeOther::in_order_of_submit,
    

or:

 monitor => sub {print "anonymous sub called in order of submit\n"},
    

The "monitor" field specifies the subroutine to be executed for monitoring the results of the "do" routine. If specified, the "monitor" routine is called once for the result of each "do" subroutine, but in the order in which the jobs were submitted rather than in the order in which the result were obtained (which is by the very nature of threads, indeterminate).

The specified subroutine should expect the following parameters to be passed:

 1..N  the values that were returned by the "do" subroutine
    

The "monitor" routine is executed in its own thread. This means that all results have to be passed between threads, and therefore be frozen and thawed with Storable. If you can handle the streaming from different threads, it is probably wiser to use the "stream" routine feature.

pre_post_monitor_only

 pre_post_monitor_only => 1, # default 0
    

The "pre_post_monitor_only" field only makes sense if a "monitor" routine is specified. If specified with a true value, indicates that the "pre" and "post" routines (if specified) should only be called for the "monitor" routine only and not for the "do" routine. Otherwise, the same "pre" and "post" routine will be called for both the "do" as well as the "monitor" routine.

When the "pre" and "post" routine are called for the "do" subroutine, the self class method returns the Thread::Pool object (which it doesn't do when called in the "monitor" routine).

checkpoint

 checkpoint => 'checkpointing',                 # assume caller's namespace
    

or:

 checkpoint => 'Package::checkpointing',
    

or:

 checkpoint => \&SomeOther::checkpointing,
    

or:

 checkpoint => sub {print "anonymous sub to do checkpointing\n"},
    

The "checkpoint" field specifies the subroutine to be executed every time a checkpoint should be made by a monitoring routine (e.g. for saving or updating status). It must be specified as either the name of a subroutine or as a reference to a (anonymous) subroutine.

It only makes sense to specify a checkpoint routine if there is also a monitoring routine specified. No checkpointing will occur by default if a monitoring routine is specified. The frequency of checkpointing can be specified with the "frequency" field.

The specified subroutine should not expect any parameters to be passed. Any values returned by the checkpointing routine, will be lost.

frequency

 frequency => 100,                             # default = 1000
    

The "frequency" field specifies the number of jobs that should have been monitored before the "checkpoint" routine is called. If a checkpoint routine is specified but no frequency field is specified, then a frequency of 1000 will be assumed.

This field has no meaning if no checkpoint routine is specified with the "checkpoint" field. The default frequency can be changed with the frequency method.

autoshutdown

 autoshutdown => 0, # default: 1
    

The "autoshutdown" field specified whether the shutdown method should be called when the object is destroyed. By default, this flag is set to 1 indicating that the shutdown method should be called when the object is being destroyed. Setting the flag to a false value, will cause the shutdown method not to be called, causing potential loss of data and error messages when threads are not finished when the program exits.

The setting of the flag can be later changed by calling the autoshutdown method.

workers

 workers => 5, # default: 1
    

The "workers" field specifies the number of worker threads that should be created when the pool is created. If no "workers" field is specified, then only one worker thread will be created. The workers method can be used to change the number of workers later.

maxjobs

 maxjobs => 25, # default: 5 * workers
    

The "maxjobs" field specifies the maximum number of jobs that can be sitting on the belt to be handled (job throttling). If a new job submission would exceed this amount, job submission will be halted until the number of jobs waiting to be handled has become at least as low as the amount specified with the "minjobs" field.

If the "maxjobs" field is not specified, an amount of 5 * the number of worker threads will be assumed. If you do not want to have any job throttling, you can specify the value "undef" for the field. But beware! If you do not have job throttling active, you may wind up using excessive amounts of memory used for storing all of the job submission information.

The maxjobs method can be called to change the job throttling settings during the lifetime of the object.

minjobs

 minjobs => 10, # default: maxjobs / 2
    

The "minjobs" field specified the minimum number of jobs that can be waiting on the belt to be handled before job submission is allowed again (job throttling).

If job throttling is active and the "minjobs" field is not specified, then half of the "maxjobs" value will be assumed.

The minjobs method can be called to change the job throttling settings during the lifetime of the object.

frequency

 Thread::Pool->frequency( 100 );
 $frequency = Thread::Pool->frequency;

The "frequency" class method allows you to specify the default frequency that will be used when a checkpoint routine is specified with the "checkpoint" field. The default frequency is set to 1000 if no other value has been previously specified.

optimize

 Thread::Pool->optimize( 'cpu' );
 $optimize = Thread::Pool->optimize;

The "optimize" class method allows you to specify the default optimization type that will be used if no "optimize" field has been explicitly specified with a call to new. It returns the current default type of optimization.

Currently two types of optimization can be selected:

memory

Attempt to use as little memory as possible. Currently, this is achieved by starting a separate thread which hosts an unshared array. This uses the "Thread::Conveyor::Thread" sub-class.

cpu

Attempt to use as little CPU as possible. Currently, this is achieved by using a shared array (using the "Thread::Conveyor::Array" sub-class), encapsulated in a hash reference if throttling is activated (then also using the "Thread::Conveyor::Throttled" sub-class).

job

 $jobid = $pool->job( @parameter );     # saves result
 $pool->job( @parameter );              # does not save result

The "job" method specifies a job to be executed by any of the available workers. Which worker will execute the job, is indeterminate. When it will happen, depends on the number of jobs that still have to be done when this job was submitted.

The input parameters are passed to the "do" subroutine as is.

If a return value is requested, then the return value(s) of the "do" subroutine will be saved. The returned value is a job ID that should be used as the input parameter to result or result_dontwait.

waitfor

 @result = $pool->waitfor( @parameter ); # submit job and wait for result

The "waitfor" method specifies a job to be executed, wait for the result to become ready and return the result. It is in fact a shortcut for using job and result.

The input parameters are passed to the "do" subroutine as is.

The return value(s) are what was returned by the "do" routine. The meaning of the return value(s) is entirely up to you as the developer.

result

 @result = $pool->result( $jobid );

The "result" method waits for the specified job to be finished and returns the result of that job.

The input parameter is the job id as returned from the job assignment.

The return value(s) are what was returned by the "do" routine. The meaning of the return value(s) is entirely up to you as the developer.

If you want to wait for any job to be finished, use the result_any method.

If you don't want to wait for the job to be finished, but just want to see if there is a result already, use the result_dontwait method.

result_any

 @result = $pool->result_any;
 @result = $pool->result_any( \$jobid );

The "result_any" method waits for any job to be finished and returns the result of that job.

The optional input parameter is the reference to a scalar variable in which the job id will be stored.

The return value(s) are what was returned by the "do" routine. The meaning of the return value(s) is entirely up to you as the developer.

If you don't want to wait for a job to be finished, but just want to see if there is a result already, use the result_dontwait method.

result_dontwait

 @result = $pool->result_dontwait( $jobid );

The "result_dontwait" method returns the result of the job if it is available. If the job is not finished yet, it will return undef in scalar context or the empty list in list context.

The input parameter is the job id as returned from the job assignment.

If the result of the job is available, then the return value(s) are what was returned by the "do" routine. The meaning of the return value(s) is entirely up to you as the developer.

If you want to wait for the job to be finished, use the result method.

todo

 $todo = $pool->todo;

The "todo" method returns the number of jobs that are still left to be done.

results

 $results = $pool->results;
 @result = $pool->results;

The "results" method returns the jobids of which there are results available and which have not yet been fetched with result. Returns the number of results available in scalar context.

add

 $tid = $pool->add;             # add 1 worker thread
 @tid = $pool->add( 5 );

The "add" method adds the specified number of worker threads to the pool and returns the thread ID's (tid) of the threads that were created.

The input parameter specifies the number of workers to be added. If no number of workers is specified, then 1 worker thread will be added.

In scalar context, returns the thread ID (tid) of the first worker thread that was added. This usually only makes sense if you're adding only one worker thread.

In list context, returns the thread ID's (tid) of the worker threads that were created.

Each time a worker thread is added, the "pre" routine (if available) will be called inside the thread.

remove

 $pool->remove;                 # remove 1 worker thread
 $pool->remove( 5 );            # remove 5 worker threads
 $jobid = $pool->remove;        # remove 1 worker thread, save result
 @jobid = $pool->remove( 5 );   # remove 5 worker threads, save results

The "remove" method adds the specified number of special "remove" job to the lists of jobs to be done. It will return the job ID's if called in a non-void context.

The input parameter specifies the number of workers to be removed. If no number of workers is specified, then 1 worker thread will be removed.

In void context, the results of the execution of the "post" subroutine(s) is discarded.

In scalar context, returns the job ID of the result of the first worker thread that was removed. This usually only makes sense if you're removing only one worker thread.

In list context, returns the job ID's of the result of all the worker threads that were removed.

Each time a worker thread is removed, the "post" routine is called. Its return value(s) are saved only if a job ID was requested when removing the thread. Then the result method can be called to obtain the results of the "post" subroutine.

workers

 $workers = $pool->workers;     # find out number of worker threads
 $pool->workers( 10 );          # set number of worker threads

The "workers" method can be used to find out how many worker threads there are currently available, or it can be used to set the number of worker threads.

The input value, if specified, specifies the number of worker threads that should be available. If there are more worker threads available than the number specified, then superfluous worker threads will be removed. If there are not enough worker threads available, new worker threads will be added.

The return value is the current number of worker threads.

frequency

 $frequency = $pool->frequency;

The "frequency" instance method returns the frequency with which the checkpoint routine is being called. Returns undef if no checkpointing is being done.

maxjobs

 $pool->maxjobs( 100 );
 $maxjobs = $pool->maxjobs;

The "maxjobs" method returns the maximum number of jobs that can be on the belt before job throttling sets in. The input value, if specified, specifies the new maximum number of jobs that may be on the belt. Job throttling will be switched off if the value 0 is specified.

Specifying the "maxjobs" field when creating the pool object with new is equivalent to calling this method.

The minjobs method can be called to specify the minimum number of jobs that must be on the belt before job submission is allowed again after reaching the maximum number of jobs. By default, half of the "maxjobs" value is assumed.

minjobs

 $pool->minjobs( 50 );
 $minjobs = $pool->minjobs;

The "minjobs" method returns the minimum number of jobs that must be on the belt before job submission is allowed again after reaching the maximum number of jobs. The input value, if specified, specifies the new minimum number of jobs that must be on the belt.

Specifying the "minjobs" field when creating the pool object with new is equivalent to calling this method.

The maxjobs method can be called to set the maximum number of jobs that may be on the belt before job submission will be halted.

join

 $pool->join;

The "join" method waits until all of the worker threads that have been removed have finished their jobs. It basically cleans up the threads that are not needed anymore.

The "shutdown" method call the "join" method after removing all the active worker threads. You therefore seldom need to call the "join" method separately.

removed

 $removed = $pool->removed;

The "removed" method returns the number of worker threads that were removed over the lifetime of the object.

autoshutdown

 $pool->autoshutdown( 1 );
 $autoshutdown = $pool->autoshutdown;

The "autoshutdown" method sets and/or returns the flag indicating whether an automatic shutdown should be performed when the object is destroyed.

shutdown

 $pool->shutdown;

The "shutdown" method waits for all jobs to be executed, removes all worker threads, handles any results that still need to be streamed, before it returns. Call the abort method if you do not want to wait until all jobs have been executed.

It is called automatically when the object is destroyed, unless specifically disabled by providing a false value with the "autoshutdown" field when creating the pool with new, or by calling the autoshutdown method.

abort

The "abort" method waits for all worker threads to finish their current job, removes all worker threads, before it returns. Call the shutdown method if you want to wait until all jobs have been done.

You can restart the job handling process after calling "abort" by adding workers again.

done

 $done = $pool->done;

The "done" method returns the number of jobs that has been performed by the removed worker threads of the pool.

The "done" method is typically called after the shutdown method has been called.

notused

 $notused = $pool->notused;

The "notused" method returns the number of removed threads that have not performed any jobs. It provides a heuristic to determine how many workers you actually need for a specific application: a value > 0 indicates that you have specified too many worker threads for this application.

The "notused" method is typically called after the shutdown method has been called.

self

 $self = Thread::Pool->self;

The class method "self" returns the object to which this thread belongs. It is available within the "pre", "do", "post", "stream" and "monitor" subroutines only.

monitor

 $monitor = Thread::Pool->monitor;

The class method "monitor" returns the Thread::Conveyor::Monitored object that is associated with the pool. It is available only if the "monitor" field was specified in new. And then only within the "pre", "do", "post", "stream" and "monitor" subroutines only.

remove_me

 Thread::Pool->remove_me;

The "remove_me" class method only makes sense within the "do" subroutine. It indicates to the job dispatcher that this worker thread should be removed from the pool. After the "do" subroutine returns, the worker thread will be removed.

jobid

 $jobid = Thread::Pool->jobid;

The "jobid" class method only makes sense within the "do" subroutine in streaming mode. It returns the job ID value of the current job. This can be used connection with the dont_set_result and the set_result methods to have another thread set the result of the current job.

dont_set_result

 Thread::Pool->dont_set_result;

The "dont_set_result" class method only makes sense within the "do" subroutine. It indicates to the job dispatcher that the result of this job should not be saved. This is for cases where the result of this job will be placed in the result hash at some time in the future by another thread using the set_result method.

set_result

 Thread::Pool->self->set_result( $jobid,@param );

The "set_result" object method only makes sense within the "do" subroutine. It allows you to set the result of other jobs than the one currently being performed.

This method is only needed in very special situations. Normally, just returning values from the "do" subroutine is enough to have the result saved. This method is exposed to the outside world in those cases where a specific thread becomes responsible for setting the result of other threads (which used the dont_set_result method to defer saving their result.

The first input parameter specifies the job ID of the job for which to set the result. The rest of the input parameters is considered to be the result to be saved. Whatever is specified in the rest of the input parameters, will be returned with the result or result_dontwait methods.

simple asynchronous log file resolving filter

This is an example of a very simple asynchronous log file resolver filter.

Because the IP number to domain name translation is dependent on external DNS servers, it can take quite some (wallclock) time before a response is returned by the "gethostbyaddr" function. In a single threaded environment, a single bad DNS server can severely slow down the resolving process. In a threaded environment, you can have one thread waiting for a slow DNS server while other threads are able to obtain answers in the mean time.

This example uses a shared hash to keep results from DNS server responses, so that if an IP number was attempted to be resolved once (either successfully or unsuccessfully), it will not be attempted again: instead the value from the hash will be assumed.

 # You should always use strict!
 # Using Thread::Pool by itself is enough, no "use threads;" needed
 # Initialize the shared hash with IP numbers and their results
 use strict;
 use Thread::Pool;
 my %resolved : shared;
 # Create the pool of threads
 my $pool = Thread::Pool->new(
  {
   workers => 10,
   do => \&do,
   monitor => \&monitor,
  }
 );
 # Submit each line as a job to the pool
 $pool->job( $_ ) while <>;
 #--------------------------------------------------------------------
 # Handle a single job
 #  IN: 1 log line to resolve
 # OUT: 1 resolved log line
 sub do {
 # Substitute the IP number at the start with the name or with the original
 # Return the adapted value
   $_[0] =~ s#^(\d+\.\d+\.\d+\.\d+)#
    $resolved{$1} ||= gethostbyaddr( pack( 'C4',split(/\./,$1)),2 ) || $1#e;
   $_[0];
 } #do
 #--------------------------------------------------------------------
 # Output the results in the order they were submitted
 #  IN: 1 resolved log line
 sub monitor { print $_[0] } #monitor

This is a very simple filter. The main drawback is that many threads can be looking up the same IP number at the same time.

another asynchronous log file resolving filter

This is an example of a not so very simple asynchronous log file resolver filter. This is in fact the base code for the Thread::Pool::Resolve module.

In this example, the dont_set_result and set_result methods are used to put up all lines with the same unresolved IP number in the same thread until the DNS server returns, either with or without a result. Then all the lines with that IP number are handled by that thread: the other threads have long before that already continued attempting to handle other lines.

Because only the "do" subroutine is different from the previous example, we're only showing that.

 #--------------------------------------------------------------------
 # Handle a single job
 #  IN: 1 log line to resolve
 # OUT: 1 resolved log line (if already resolved, else ignored)
 sub do {
 # Obtain the line to work with
 # Return it now if it is already resolved (at least not an IP number there)
 # Save the IP number for later usage, line is now without IP number
   my $line = shift;
   return $line unless $line =~ s#^(\d+\.\d+\.\d+\.\d+)##;
   my $ip = $1;
 # Make sure we're the only one to access the resolved hash now
 # If there is already information for this IP number
 #  Return what is there with the line if it was resolved already
   {lock( %resolved );
    if (exists( $resolved{$ip} )) {
      return ($resolved{$ip} || $ip).$line unless ref( $resolved{$ip} );
 #  Set the rest of the line in the todo hash, keyed to jobid
 #  Set the flag that this result should not be set in the result hash
 #  And return without anything (thread will continue with next job)
      $resolved{$ip}->{Thread::Pool->jobid} = $line;
      Thread::Pool->dont_set_result;
      return;
 # Else (first time this IP number is encountered)
 #  Create a empty shared hash
 #  Save a reference to the hash in the todo hash as info for this IP number
    } else {
      my %hash : shared;
      $resolved{$ip} = \%hash;
    }
   } #%resolved
 # Do the actual name resolving (may take quite some time) or use IP number
 # Obtain local copy of the Thread::Pool object
 # Obtain local copy of the todo hash
   my $domain = gethostbyaddr( pack( 'C4',split(/\./,$ip)),2 ) || $ip;
   my $pool = Thread::Pool->self;
   my $todo = $resolved{$ip};
 # Make sure we're the only one accessing the resolved hash (rest of this sub)
 # For all the lines with this IP number
 #  Set the results
 # Remove the todo hash and replace by domain or blank string if unresolvable
 # Return the result for this job
   lock( %resolved );
   while (my $key = each %{$todo}) {
       $pool->set_result( $key,$domain.$todo->{$key} )
   }
   $resolved{$ip} = $domain eq $ip ? undef : $domain;
   $domain.$line;
 } #do