NAME Coro::ProcessPool - an asynchronous process pool SYNOPSIS use Coro::ProcessPool; my $pool = Coro::ProcessPool->new( max_procs => 4, max_reqs => 100, ); my $double = sub { $_[0] * 2 }; # Process in sequence my %result; foreach my $i (1 .. 1000) { $result{$i} = $pool->process($double, [$i]); } # Process as a batch my @results = $pool->map($double, 1 .. 1000); # Defer waiting for result my %deferred = map { $_ => $pool->defer($double, [$_]) } 1 .. 1000); foreach my $i (keys %deferred) { print "$i = " . $deferred{$i}->() . "\n"; } # Use a "task class", implementing 'new' and 'run' my $result = $pool->process('Task::Doubler', 21); $pool->shutdown; DESCRIPTION Processes tasks using a pool of external Perl processes. METHODS new Creates a new process pool. Processes will be spawned as needed. max_procs This is the maximum number of child processes to maintain. If all processes are busy handling tasks, further calls to "process" in . will yield until a process becomes available. If not specified, defaults to the number of CPUs on the system. max_reqs If this is a positive number (defaults to 0), child processes will be terminated and replaced after handling "max_reqs" tasks. Choosing the correct value for "max_reqs" is a tradeoff between the need to clear memory leaks in the child process and the time it takes to spawn a new process and import any packages used by client code. process($f, $args, $timeout) Processes code ref $f in a child process from the pool. If $args is provided, it is an array ref of arguments that will be passed to $f. Returns the result of calling $f->(@$args). Alternately, $f may be the name of a class implementing the methods "new" and "run", in which case the result is equivalent to calling $f->new(@$args)->run(). Note that the include path for worker processes is identical to that of the calling process. This call will yield until the results become available. If all processes are busy, this method will block until one becomes available. Processes are spawned as needed, up to "max_procs", from this method. Also note that the use of "max_reqs" can cause this method to yield while a new process is spawned. A timeout may be optionally specified in fractional seconds. If specified, $timeout will cause "process" to croak if $timeout seconds pass an no process becomes available to handle the task. Note that the timeout only applies to the time it takes to acquire an available process. It does not watch the time it takes to perform the task. map($f, @args) Applies $f to each value in @args in turn and returns a list of the results. Although the order in which each argument is processed is not guaranteed, the results are guaranteed to be in the same order as @args, even if the result of calling $f returns a list itself (in which case, the results of that calcuation is flattened into the list returned by "map". defer($f, $args) Similar to "process" in ., but returns immediately. The return value is a code reference that, when called, returns the results of calling "$f-"(@$args)>. my $deferred = $pool->defer($coderef, [ $x, $y, $z ]); my $result = $deferred->(); queue($f, $args, $callback) Queues the execution of "$f-"(@$args)> and returns immediately. If $callback is specified and is a code ref, it will be called with the result of the executed code once the task is processed. Note that the callback is not provided with any identifying information about the task being executed. That is the responsibility of the caller. It is also the caller's responsibility to coordinate any code that depends on the result, for example using a condition variable. my $cv = AnyEvent->condvar; sub make_callback { my $n = shift; return sub { my $result = shift; print "2 * $n = $result\n"; $cv->send; } } $pool->queue($doubler_function, [21], make_callback(21)); $cv->recv; shutdown Shuts down all processes and resets state on the process pool. After calling this method, the pool is effectively in a new state and may be used normally. A NOTE ABOUT IMPORTS AND CLOSURES Code refs are serialized using Storable to pass them to the worker processes. Once deserialized in the pool process, these functions can no longer see the stack as it is in the parent process. Therefore, imports and variables external to the function are unavailable. Something like this will not work: use Foo; my $foo = Foo->new(); my $result = $pool->process(sub { return $foo->bar; # $foo not found }); Nor will this: use Foo; my $result = $pool->process(sub { my $foo = Foo->new; # Foo not found return $foo->bar; }); The correct way to do this is to import from within the function: my $result = $pool->process(sub { require Foo; my $foo = Foo->new(); return $foo->bar; }); ...or to pass in external variables that are needed by the function: use Foo; my $foo = Foo->new(); my $result = $pool->process(sub { $_[0]->bar }, [ $foo ]); Use versus require The "use" pragma is run at compile time, whereas "require" is evaluated at runtime. Because of this, the use of "use" in code passed directly to the "process" method can fail because the "use" statement has already been evaluated when the calling code was compiled. This will not work: $pool->process(sub { use Foo; my $foo = Foo->new(); }); This will work: $pool->process(sub { require Foo; my $foo = Foo->new(); }); If "use" is necessary (for example, to import a method or transform the calling code via import), it is recommended to move the code into its own module, which can then be called in the anonymous routine: package Bar; use Foo; sub dostuff { ... } Then, in your caller: $pool->process(sub { require Bar; Bar::dostuff(); }); If it's a problem... Use the task class method if the loading requirements are causing headaches: my $result = $pool->process('Task::Class', [@args]); COMPATIBILITY "Coro::ProcessPool" will likely break on Win32 due to missing support for non-blocking file descriptors (Win32 can only call "select" and "poll" on actual network sockets). Without rewriting this as a network server, which would impact performance and be really annoying, it is likely this module will not support Win32 in the near future. The following modules will get you started if you wish to explore a synchronous process pool on Windows: Win32::Process Win32::IPC Win32::Pipe AUTHOR Jeff Ober LICENSE BSD License