NAME
Coro - the only real threads in perl
SYNOPSIS
use Coro;
async {
# some asynchronous thread of execution
print "2\n";
cede; # yield back to main
print "4\n";
};
print "1\n";
cede; # yield to coro
print "3\n";
cede; # and again
# use locking
use Coro::Semaphore;
my $lock = new Coro::Semaphore;
my $locked;
$lock->down;
$locked = 1;
$lock->up;
DESCRIPTION
For a tutorial-style introduction, please read the Coro::Intro manpage.
This manpage mainly contains reference information.
This module collection manages continuations in general, most often in
the form of cooperative threads (also called coros, or simply "coro" in
the documentation). They are similar to kernel threads but don't (in
general) run in parallel at the same time even on SMP machines. The
specific flavor of thread offered by this module also guarantees you
that it will not switch between threads unless necessary, at
easily-identified points in your program, so locking and parallel access
are rarely an issue, making thread programming much safer and easier
than using other thread models.
Unlike the so-called "Perl threads" (which are not actually real threads
but only the windows process emulation (see section of same name for
more details) ported to unix, and as such act as processes), Coro
provides a full shared address space, which makes communication between
threads very easy. And Coro's threads are fast, too: disabling the
Windows process emulation code in your perl and using Coro can easily
result in a two to four times speed increase for your programs. A
parallel matrix multiplication benchmark runs over 300 times faster on a
single core than perl's pseudo-threads on a quad core using all four
cores.
Coro achieves that by supporting multiple running interpreters that
share data, which is especially useful to code pseudo-parallel processes
and for event-based programming, such as multiple HTTP-GET requests
running concurrently. See Coro::AnyEvent to learn more on how to
integrate Coro into an event-based environment.
In this module, a thread is defined as "callchain + lexical variables +
some package variables + C stack), that is, a thread has its own
callchain, its own set of lexicals and its own set of perls most
important global variables (see Coro::State for more configuration and
background info).
See also the "SEE ALSO" section at the end of this document - the Coro
module family is quite large.
GLOBAL VARIABLES
$Coro::main
This variable stores the Coro object that represents the main
program. While you cna "ready" it and do most other things you can
do to coro, it is mainly useful to compare again $Coro::current, to
see whether you are running in the main program or not.
$Coro::current
The Coro object representing the current coro (the last coro that
the Coro scheduler switched to). The initial value is $Coro::main
(of course).
This variable is strictly *read-only*. You can take copies of the
value stored in it and use it as any other Coro object, but you must
not otherwise modify the variable itself.
$Coro::idle
This variable is mainly useful to integrate Coro into event loops.
It is usually better to rely on Coro::AnyEvent or Coro::EV, as this
is pretty low-level functionality.
This variable stores a Coro object that is put into the ready queue
when there are no other ready threads (without invoking any ready
hooks).
The default implementation dies with "FATAL: deadlock detected.",
followed by a thread listing, because the program has no other way
to continue.
This hook is overwritten by modules such as "Coro::EV" and
"Coro::AnyEvent" to wait on an external event that hopefully wake up
a coro so the scheduler can run it.
See Coro::EV or Coro::AnyEvent for examples of using this technique.
SIMPLE CORO CREATION
async { ... } [@args...]
Create a new coro and return its Coro object (usually unused). The
coro will be put into the ready queue, so it will start running
automatically on the next scheduler run.
The first argument is a codeblock/closure that should be executed in
the coro. When it returns argument returns the coro is automatically
terminated.
The remaining arguments are passed as arguments to the closure.
See the "Coro::State::new" constructor for info about the coro
environment in which coro are executed.
Calling "exit" in a coro will do the same as calling exit outside
the coro. Likewise, when the coro dies, the program will exit, just
as it would in the main program.
If you do not want that, you can provide a default "die" handler, or
simply avoid dieing (by use of "eval").
Example: Create a new coro that just prints its arguments.
async {
print "@_\n";
} 1,2,3,4;
async_pool { ... } [@args...]
Similar to "async", but uses a coro pool, so you should not call
terminate or join on it (although you are allowed to), and you get a
coro that might have executed other code already (which can be good
or bad :).
On the plus side, this function is about twice as fast as creating
(and destroying) a completely new coro, so if you need a lot of
generic coros in quick successsion, use "async_pool", not "async".
The code block is executed in an "eval" context and a warning will
be issued in case of an exception instead of terminating the
program, as "async" does. As the coro is being reused, stuff like
"on_destroy" will not work in the expected way, unless you call
terminate or cancel, which somehow defeats the purpose of pooling
(but is fine in the exceptional case).
The priority will be reset to 0 after each run, tracing will be
disabled, the description will be reset and the default output
filehandle gets restored, so you can change all these. Otherwise the
coro will be re-used "as-is": most notably if you change other
per-coro global stuff such as $/ you *must needs* revert that
change, which is most simply done by using local as in: "local $/".
The idle pool size is limited to 8 idle coros (this can be adjusted
by changing $Coro::POOL_SIZE), but there can be as many non-idle
coros as required.
If you are concerned about pooled coros growing a lot because a
single "async_pool" used a lot of stackspace you can e.g.
"async_pool { terminate }" once per second or so to slowly replenish
the pool. In addition to that, when the stacks used by a handler
grows larger than 32kb (adjustable via $Coro::POOL_RSS) it will also
be destroyed.
STATIC METHODS
Static methods are actually functions that implicitly operate on the
current coro.
schedule
Calls the scheduler. The scheduler will find the next coro that is
to be run from the ready queue and switches to it. The next coro to
be run is simply the one with the highest priority that is longest
in its ready queue. If there is no coro ready, it will call the
$Coro::idle hook.
Please note that the current coro will *not* be put into the ready
queue, so calling this function usually means you will never be
called again unless something else (e.g. an event handler) calls
"->ready", thus waking you up.
This makes "schedule" *the* generic method to use to block the
current coro and wait for events: first you remember the current
coro in a variable, then arrange for some callback of yours to call
"->ready" on that once some event happens, and last you call
"schedule" to put yourself to sleep. Note that a lot of things can
wake your coro up, so you need to check whether the event indeed
happened, e.g. by storing the status in a variable.
See HOW TO WAIT FOR A CALLBACK, below, for some ways to wait for
callbacks.
cede
"Cede" to other coros. This function puts the current coro into the
ready queue and calls "schedule", which has the effect of giving up
the current "timeslice" to other coros of the same or higher
priority. Once your coro gets its turn again it will automatically
be resumed.
This function is often called "yield" in other languages.
Coro::cede_notself
Works like cede, but is not exported by default and will cede to
*any* coro, regardless of priority. This is useful sometimes to
ensure progress is made.
terminate [arg...]
Terminates the current coro with the given status values (see
cancel).
Coro::on_enter BLOCK, Coro::on_leave BLOCK
These function install enter and leave winders in the current scope.
The enter block will be executed when on_enter is called and
whenever the current coro is re-entered by the scheduler, while the
leave block is executed whenever the current coro is blocked by the
scheduler, and also when the containing scope is exited (by whatever
means, be it exit, die, last etc.).
*Neither invoking the scheduler, nor exceptions, are allowed within
those BLOCKs*. That means: do not even think about calling "die"
without an eval, and do not even think of entering the scheduler in
any way.
Since both BLOCKs are tied to the current scope, they will
automatically be removed when the current scope exits.
These functions implement the same concept as "dynamic-wind" in
scheme does, and are useful when you want to localise some resource
to a specific coro.
They slow down thread switching considerably for coros that use them
(about 40% for a BLOCK with a single assignment, so thread switching
is still reasonably fast if the handlers are fast).
These functions are best understood by an example: The following
function will change the current timezone to
"Antarctica/South_Pole", which requires a call to "tzset", but by
using "on_enter" and "on_leave", which remember/change the current
timezone and restore the previous value, respectively, the timezone
is only changed for the coro that installed those handlers.
use POSIX qw(tzset);
async {
my $old_tz; # store outside TZ value here
Coro::on_enter {
$old_tz = $ENV{TZ}; # remember the old value
$ENV{TZ} = "Antarctica/South_Pole";
tzset; # enable new value
};
Coro::on_leave {
$ENV{TZ} = $old_tz;
tzset; # restore old value
};
# at this place, the timezone is Antarctica/South_Pole,
# without disturbing the TZ of any other coro.
};
This can be used to localise about any resource (locale, uid,
current working directory etc.) to a block, despite the existance of
other coros.
Another interesting example implements time-sliced multitasking
using interval timers (this could obviously be optimised, but does
the job):
# "timeslice" the given block
sub timeslice(&) {
use Time::HiRes ();
Coro::on_enter {
# on entering the thread, we set an VTALRM handler to cede
$SIG{VTALRM} = sub { cede };
# and then start the interval timer
Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0.01, 0.01;
};
Coro::on_leave {
# on leaving the thread, we stop the interval timer again
Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0, 0;
};
&{+shift};
}
# use like this:
timeslice {
# The following is an endless loop that would normally
# monopolise the process. Since it runs in a timesliced
# environment, it will regularly cede to other threads.
while () { }
};
killall
Kills/terminates/cancels all coros except the currently running one.
Note that while this will try to free some of the main interpreter
resources if the calling coro isn't the main coro, but one cannot
free all of them, so if a coro that is not the main coro calls this
function, there will be some one-time resource leak.
CORO OBJECT METHODS
These are the methods you can call on coro objects (or to create them).
new Coro \&sub [, @args...]
Create a new coro and return it. When the sub returns, the coro
automatically terminates as if "terminate" with the returned values
were called. To make the coro run you must first put it into the
ready queue by calling the ready method.
See "async" and "Coro::State::new" for additional info about the
coro environment.
$success = $coro->ready
Put the given coro into the end of its ready queue (there is one
queue for each priority) and return true. If the coro is already in
the ready queue, do nothing and return false.
This ensures that the scheduler will resume this coro automatically
once all the coro of higher priority and all coro of the same
priority that were put into the ready queue earlier have been
resumed.
$coro->suspend
Suspends the specified coro. A suspended coro works just like any
other coro, except that the scheduler will not select a suspended
coro for execution.
Suspending a coro can be useful when you want to keep the coro from
running, but you don't want to destroy it, or when you want to
temporarily freeze a coro (e.g. for debugging) to resume it later.
A scenario for the former would be to suspend all (other) coros
after a fork and keep them alive, so their destructors aren't
called, but new coros can be created.
$coro->resume
If the specified coro was suspended, it will be resumed. Note that
when the coro was in the ready queue when it was suspended, it might
have been unreadied by the scheduler, so an activation might have
been lost.
To avoid this, it is best to put a suspended coro into the ready
queue unconditionally, as every synchronisation mechanism must
protect itself against spurious wakeups, and the one in the Coro
family certainly do that.
$is_ready = $coro->is_ready
Returns true iff the Coro object is in the ready queue. Unless the
Coro object gets destroyed, it will eventually be scheduled by the
scheduler.
$is_running = $coro->is_running
Returns true iff the Coro object is currently running. Only one Coro
object can ever be in the running state (but it currently is
possible to have multiple running Coro::States).
$is_suspended = $coro->is_suspended
Returns true iff this Coro object has been suspended. Suspended
Coros will not ever be scheduled.
$coro->cancel (arg...)
Terminates the given Coro and makes it return the given arguments as
status (default: the empty list). Never returns if the Coro is the
current Coro.
$coro->schedule_to
Puts the current coro to sleep (like "Coro::schedule"), but instead
of continuing with the next coro from the ready queue, always switch
to the given coro object (regardless of priority etc.). The
readyness state of that coro isn't changed.
This is an advanced method for special cases - I'd love to hear
about any uses for this one.
$coro->cede_to
Like "schedule_to", but puts the current coro into the ready queue.
This has the effect of temporarily switching to the given coro, and
continuing some time later.
This is an advanced method for special cases - I'd love to hear
about any uses for this one.
$coro->throw ([$scalar])
If $throw is specified and defined, it will be thrown as an
exception inside the coro at the next convenient point in time.
Otherwise clears the exception object.
Coro will check for the exception each time a schedule-like-function
returns, i.e. after each "schedule", "cede",
"Coro::Semaphore->down", "Coro::Handle->readable" and so on. Most of
these functions detect this case and return early in case an
exception is pending.
The exception object will be thrown "as is" with the specified
scalar in $@, i.e. if it is a string, no line number or newline will
be appended (unlike with "die").
This can be used as a softer means than "cancel" to ask a coro to
end itself, although there is no guarantee that the exception will
lead to termination, and if the exception isn't caught it might well
end the whole program.
You might also think of "throw" as being the moral equivalent of
"kill"ing a coro with a signal (in this case, a scalar).
$coro->join
Wait until the coro terminates and return any values given to the
"terminate" or "cancel" functions. "join" can be called concurrently
from multiple coro, and all will be resumed and given the status
return once the $coro terminates.
$coro->on_destroy (\&cb)
Registers a callback that is called when this coro gets destroyed,
but before it is joined. The callback gets passed the terminate
arguments, if any, and *must not* die, under any circumstances.
$oldprio = $coro->prio ($newprio)
Sets (or gets, if the argument is missing) the priority of the coro.
Higher priority coro get run before lower priority coro. Priorities
are small signed integers (currently -4 .. +3), that you can refer
to using PRIO_xxx constants (use the import tag :prio to get then):
PRIO_MAX > PRIO_HIGH > PRIO_NORMAL > PRIO_LOW > PRIO_IDLE > PRIO_MIN
3 > 1 > 0 > -1 > -3 > -4
# set priority to HIGH
current->prio (PRIO_HIGH);
The idle coro ($Coro::idle) always has a lower priority than any
existing coro.
Changing the priority of the current coro will take effect
immediately, but changing the priority of coro in the ready queue
(but not running) will only take effect after the next schedule (of
that coro). This is a bug that will be fixed in some future version.
$newprio = $coro->nice ($change)
Similar to "prio", but subtract the given value from the priority
(i.e. higher values mean lower priority, just as in unix).
$olddesc = $coro->desc ($newdesc)
Sets (or gets in case the argument is missing) the description for
this coro. This is just a free-form string you can associate with a
coro.
This method simply sets the "$coro->{desc}" member to the given
string. You can modify this member directly if you wish.
GLOBAL FUNCTIONS
Coro::nready
Returns the number of coro that are currently in the ready state,
i.e. that can be switched to by calling "schedule" directory or
indirectly. The value 0 means that the only runnable coro is the
currently running one, so "cede" would have no effect, and
"schedule" would cause a deadlock unless there is an idle handler
that wakes up some coro.
my $guard = Coro::guard { ... }
This function still exists, but is deprecated. Please use the
"Guard::guard" function instead.
unblock_sub { ... }
This utility function takes a BLOCK or code reference and "unblocks"
it, returning a new coderef. Unblocking means that calling the new
coderef will return immediately without blocking, returning nothing,
while the original code ref will be called (with parameters) from
within another coro.
The reason this function exists is that many event libraries (such
as the venerable Event module) are not thread-safe (a weaker form of
reentrancy). This means you must not block within event callbacks,
otherwise you might suffer from crashes or worse. The only event
library currently known that is safe to use without "unblock_sub" is
EV.
Coro will try to catch you when you block in the event loop
("FATAL:$Coro::IDLE blocked itself"), but this is just best effort
and only works when you do not run your own event loop.
This function allows your callbacks to block by executing them in
another coro where it is safe to block. One example where blocking
is handy is when you use the Coro::AIO functions to save results to
disk, for example.
In short: simply use "unblock_sub { ... }" instead of "sub { ... }"
when creating event callbacks that want to block.
If your handler does not plan to block (e.g. simply sends a message
to another coro, or puts some other coro into the ready queue),
there is no reason to use "unblock_sub".
Note that you also need to use "unblock_sub" for any other callbacks
that are indirectly executed by any C-based event loop. For example,
when you use a module that uses AnyEvent (and you use
Coro::AnyEvent) and it provides callbacks that are the result of
some event callback, then you must not block either, or use
"unblock_sub".
$cb = rouse_cb
Create and return a "rouse callback". That's a code reference that,
when called, will remember a copy of its arguments and notify the
owner coro of the callback.
See the next function.
@args = rouse_wait [$cb]
Wait for the specified rouse callback (or the last one that was
created in this coro).
As soon as the callback is invoked (or when the callback was invoked
before "rouse_wait"), it will return the arguments originally passed
to the rouse callback. In scalar context, that means you get the
*last* argument, just as if "rouse_wait" had a "return ($a1, $a2,
$a3...)" statement at the end.
See the section HOW TO WAIT FOR A CALLBACK for an actual usage
example.
HOW TO WAIT FOR A CALLBACK
It is very common for a coro to wait for some callback to be called.
This occurs naturally when you use coro in an otherwise event-based
program, or when you use event-based libraries.
These typically register a callback for some event, and call that
callback when the event occured. In a coro, however, you typically want
to just wait for the event, simplyifying things.
For example "AnyEvent->child" registers a callback to be called when a
specific child has exited:
my $child_watcher = AnyEvent->child (pid => $pid, cb => sub { ... });
But from within a coro, you often just want to write this:
my $status = wait_for_child $pid;
Coro offers two functions specifically designed to make this easy,
"Coro::rouse_cb" and "Coro::rouse_wait".
The first function, "rouse_cb", generates and returns a callback that,
when invoked, will save its arguments and notify the coro that created
the callback.
The second function, "rouse_wait", waits for the callback to be called
(by calling "schedule" to go to sleep) and returns the arguments
originally passed to the callback.
Using these functions, it becomes easy to write the "wait_for_child"
function mentioned above:
sub wait_for_child($) {
my ($pid) = @_;
my $watcher = AnyEvent->child (pid => $pid, cb => Coro::rouse_cb);
my ($rpid, $rstatus) = Coro::rouse_wait;
$rstatus
}
In the case where "rouse_cb" and "rouse_wait" are not flexible enough,
you can roll your own, using "schedule":
sub wait_for_child($) {
my ($pid) = @_;
# store the current coro in $current,
# and provide result variables for the closure passed to ->child
my $current = $Coro::current;
my ($done, $rstatus);
# pass a closure to ->child
my $watcher = AnyEvent->child (pid => $pid, cb => sub {
$rstatus = $_[1]; # remember rstatus
$done = 1; # mark $rstatus as valud
});
# wait until the closure has been called
schedule while !$done;
$rstatus
}
BUGS/LIMITATIONS
fork with pthread backend
When Coro is compiled using the pthread backend (which isn't
recommended but required on many BSDs as their libcs are completely
broken), then coro will not survive a fork. There is no known
workaround except to fix your libc and use a saner backend.
perl process emulation ("threads")
This module is not perl-pseudo-thread-safe. You should only ever use
this module from the first thread (this requirement might be removed
in the future to allow per-thread schedulers, but Coro::State does
not yet allow this). I recommend disabling thread support and using
processes, as having the windows process emulation enabled under
unix roughly halves perl performance, even when not used.
coro switching is not signal safe
You must not switch to another coro from within a signal handler
(only relevant with %SIG - most event libraries provide safe
signals), *unless* you are sure you are not interrupting a Coro
function.
That means you *MUST NOT* call any function that might "block" the
current coro - "cede", "schedule" "Coro::Semaphore->down" or
anything that calls those. Everything else, including calling
"ready", works.
WINDOWS PROCESS EMULATION
A great many people seem to be confused about ithreads (for example,
Chip Salzenberg called me unintelligent, incapable, stupid and gullible,
while in the same mail making rather confused statements about perl
ithreads (for example, that memory or files would be shared), showing
his lack of understanding of this area - if it is hard to understand for
Chip, it is probably not obvious to everybody).
What follows is an ultra-condensed version of my talk about threads in
scripting languages given onthe perl workshop 2009:
The so-called "ithreads" were originally implemented for two reasons:
first, to (badly) emulate unix processes on native win32 perls, and
secondly, to replace the older, real thread model ("5.005-threads").
It does that by using threads instead of OS processes. The difference
between processes and threads is that threads share memory (and other
state, such as files) between threads within a single process, while
processes do not share anything (at least not semantically). That means
that modifications done by one thread are seen by others, while
modifications by one process are not seen by other processes.
The "ithreads" work exactly like that: when creating a new ithreads
process, all state is copied (memory is copied physically, files and
code is copied logically). Afterwards, it isolates all modifications. On
UNIX, the same behaviour can be achieved by using operating system
processes, except that UNIX typically uses hardware built into the
system to do this efficiently, while the windows process emulation
emulates this hardware in software (rather efficiently, but of course it
is still much slower than dedicated hardware).
As mentioned before, loading code, modifying code, modifying data
structures and so on is only visible in the ithreads process doing the
modification, not in other ithread processes within the same OS process.
This is why "ithreads" do not implement threads for perl at all, only
processes. What makes it so bad is that on non-windows platforms, you
can actually take advantage of custom hardware for this purpose (as
evidenced by the forks module, which gives you the (i-) threads API,
just much faster).
Sharing data is in the i-threads model is done by transfering data
structures between threads using copying semantics, which is very slow -
shared data simply does not exist. Benchmarks using i-threads which are
communication-intensive show extremely bad behaviour with i-threads (in
fact, so bad that Coro, which cannot take direct advantage of multiple
CPUs, is often orders of magnitude faster because it shares data using
real threads, refer to my talk for details).
As summary, i-threads *use* threads to implement processes, while the
compatible forks module *uses* processes to emulate, uhm, processes.
I-threads slow down every perl program when enabled, and outside of
windows, serve no (or little) practical purpose, but disadvantages every
single-threaded Perl program.
This is the reason that I try to avoid the name "ithreads", as it is
misleading as it implies that it implements some kind of thread model
for perl, and prefer the name "windows process emulation", which
describes the actual use and behaviour of it much better.
SEE ALSO
Event-Loop integration: Coro::AnyEvent, Coro::EV, Coro::Event.
Debugging: Coro::Debug.
Support/Utility: Coro::Specific, Coro::Util.
Locking and IPC: Coro::Signal, Coro::Channel, Coro::Semaphore,
Coro::SemaphoreSet, Coro::RWLock.
I/O and Timers: Coro::Timer, Coro::Handle, Coro::Socket, Coro::AIO.
Compatibility with other modules: Coro::LWP (but see also AnyEvent::HTTP
for a better-working alternative), Coro::BDB, Coro::Storable,
Coro::Select.
XS API: Coro::MakeMaker.
Low level Configuration, Thread Environment, Continuations: Coro::State.
AUTHOR
Marc Lehmann <schmorp@schmorp.de>
http://home.schmorp.de/