EV - perl interface to libev, a high performance full-featured event loop
use EV;
# TIMERS
my $w = EV::timer 2, 0, sub {
warn "is called after 2s";
};
my $w = EV::timer 2, 2, sub {
warn "is called roughly every 2s (repeat = 2)";
};
undef $w; # destroy event watcher again
my $w = EV::periodic 0, 60, 0, sub {
warn "is called every minute, on the minute, exactly";
};
# IO
my $w = EV::io *STDIN, EV::READ, sub {
my ($w, $revents) = @_; # all callbacks receive the watcher and event mask
warn "stdin is readable, you entered: ", <STDIN>;
};
# SIGNALS
my $w = EV::signal 'QUIT', sub {
warn "sigquit received\n";
};
# CHILD/PID STATUS CHANGES
my $w = EV::child 666, 0, sub {
my ($w, $revents) = @_;
my $status = $w->rstatus;
};
# STAT CHANGES
my $w = EV::stat "/etc/passwd", 10, sub {
my ($w, $revents) = @_;
warn $w->path, " has changed somehow.\n";
};
# MAINLOOP
EV::loop; # loop until EV::unloop is called or all watchers stop
EV::loop EV::LOOP_ONESHOT; # block until at least one event could be handled
EV::loop EV::LOOP_NONBLOCK; # try to handle same events, but do not block
This module provides an interface to libev
(http://software.schmorp.de/pkg/libev.html). While the documentation
below is comprehensive, one might also consult the documentation of
libev itself (http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod or
perldoc EV::libev) for more subtle details on watcher semantics or some
discussion on the available backends, or how to force a specific backend
with LIBEV_FLAGS, or just about in any case because it has much more
detailed information.
This module is very fast and scalable. It is actually so fast that you can use it through the AnyEvent module, stay portable to other event loops (if you don't rely on any watcher types not available through it) and still be faster than with any other event loop currently supported in Perl.
This module does not export any symbols.
EV supports multiple event loops: There is a single "default event loop" that can handle everything including signals and child watchers, and any number of "dynamic event loops" that can use different backends (with various limitations), but no child and signal watchers.
You do not have to do anything to create the default event loop: When the module is loaded a suitable backend is selected on the premise of selecting a working backend (which for example rules out kqueue on most BSDs). Modules should, unless they have "special needs" always use the default loop as this is fastest (perl-wise), best supported by other modules (e.g. AnyEvent or Coro) and most portable event loop.
For specific programs you can create additional event loops dynamically.
If you want to take advantage of kqueue (which often works properly for sockets only) even though the default loop doesn't enable it, you can embed a kqueue loop into the default loop: running the default loop will then also service the kqueue loop to some extent. See the example in the section about embed watchers for an example on how to achieve that.
Create a new event loop as per the specified flags. Please refer to
the ev_loop_new () function description in the libev documentation
(http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod#GLOBAL_FUNCTIONS,
or locally-installed as EV::libev manpage) for more info.
The loop will automatically be destroyed when it is no longer referenced by any watcher and the loop object goes out of scope.
If you are not embedding the loop, then Using EV::FLAG_FORKCHECK
is recommended, as only the default event loop is protected by this
module. If you are embedding this loop in the default loop, this is not
necessary, as EV::embed automatically does the right thing on fork.
EV::FLAG_FORKCHECK which calls
this function automatically, at some performance loss (refer to the libev
documentation).
ev_verify to make internal consistency checks (for debugging
libev) and abort the program if any data structures were found to be
corrupted.
Must contain a reference to a function that is called when a callback throws an exception (with $@ containing the error). The default prints an informative message and continues.
If this callback throws an exception it will be silently ignored.
EV::BACKEND_* flags) of backends supported by this
instance of EV, the set of recommended backends (supposed to be good) for
this platform and the set of embeddable backends (see EMBED WATCHERS).
Establishes the current time by querying the kernel, updating the time
returned by EV::now in the progress. This is a costly operation and
is usually done automatically within EV::loop.
This function is rarely useful, but when some event callback runs for a very long time without entering the event loop, updating libev's idea of the current time is a good idea.
These two functions suspend and resume a loop, for use when the loop is not used for a while and timeouts should not be processed.
A typical use case would be an interactive program such as a game: When
the user presses ^Z to suspend the game and resumes it an hour later it
would be best to handle timeouts as if no time had actually passed while
the program was suspended. This can be achieved by calling suspend
in your SIGTSTP handler, sending yourself a SIGSTOP and calling
resume directly afterwards to resume timer processing.
Effectively, all timer watchers will be delayed by the time spend
between suspend and resume, and all periodic watchers
will be rescheduled (that is, they will lose any events that would have
occured while suspended).
After calling suspend you must not call any function on the given
loop other than resume, and you must not call resume
without a previous call to suspend.
Calling suspend/resume has the side effect of updating the event
loop time (see now_update).
Begin checking for events and calling callbacks. It returns when a callback calls EV::unloop.
The $flags argument can be one of the following:
0 as above EV::LOOP_ONESHOT block at most once (wait, but do not loop) EV::LOOP_NONBLOCK do not block at all (fetch/handle events but do not wait)
When called with no arguments or an argument of EV::UNLOOP_ONE, makes the innermost call to EV::loop return.
When called with an argument of EV::UNLOOP_ALL, all calls to EV::loop will return as fast as possible.
This function rolls together an I/O and a timer watcher for a single one-shot event without the need for managing a watcher object.
If $fh_or_undef is a filehandle or file descriptor, then $events
must be a bitset containing either EV::READ, EV::WRITE or EV::READ
| EV::WRITE, indicating the type of I/O event you want to wait for. If
you do not want to wait for some I/O event, specify undef for
$fh_or_undef and 0 for $events).
If timeout is undef or negative, then there will be no
timeout. Otherwise a EV::timer with this value will be started.
When an error occurs or either the timeout or I/O watcher triggers, then
the callback will be called with the received event set (in general
you can expect it to be a combination of EV::ERROR, EV::READ,
EV::WRITE and EV::TIMEOUT).
EV::once doesn't return anything: the watchers stay active till either of them triggers, then they will be stopped and freed, and the callback invoked.
$revents (a combination of
EV::READ and EV::WRITE) happened on the file descriptor $fd.
$signal had occured.
A watcher is an object that gets created to record your interest in some event. For instance, if you want to wait for STDIN to become readable, you would create an EV::io watcher for that:
my $watcher = EV::io *STDIN, EV::READ, sub {
my ($watcher, $revents) = @_;
warn "yeah, STDIN should now be readable without blocking!\n"
};
All watchers can be active (waiting for events) or inactive (paused). Only active watchers will have their callbacks invoked. All callbacks will be called with at least two arguments: the watcher and a bitmask of received events.
Each watcher type has its associated bit in revents, so you can use the same callback for multiple watchers. The event mask is named after the type, i.e. EV::child sets EV::CHILD, EV::prepare sets EV::PREPARE, EV::periodic sets EV::PERIODIC and so on, with the exception of I/O events (which can set both EV::READ and EV::WRITE bits), and EV::timer (which uses EV::TIMEOUT).
In the rare case where one wants to create a watcher but not start it at
the same time, each constructor has a variant with a trailing _ns in
its name, e.g. EV::io has a non-starting variant EV::io_ns and so on.
Please note that a watcher will automatically be stopped when the watcher object is destroyed, so you need to keep the watcher objects returned by the constructors.
Also, all methods changing some aspect of a watcher (->set, ->priority, ->fh and so on) automatically stop and start it again if it is active, which means pending events get lost.
This section lists methods common to all watchers.
_ns variants if you need stopped watchers).
Queries a freely usable data scalar on the watcher and optionally changes it. This is a way to associate custom data with a watcher:
my $w = EV::timer 60, 0, sub {
warn $_[0]->data;
};
$w->data ("print me!");
Queries the priority on the watcher and optionally changes it. Pending watchers with higher priority will be invoked first. The valid range of priorities lies between EV::MAXPRI (default 2) and EV::MINPRI (default -2). If the priority is outside this range it will automatically be normalised to the nearest valid priority.
The default priority of any newly-created watcher is 0.
Note that the priority semantics have not yet been fleshed out and are subject to almost certain change.
$revents mask.
$revents bitset (as if its callback was invoked). If the
watcher isn't pending it does nothing and returns 0.
Normally, EV::loop will return when there are no active watchers
(which is a "deadlock" because no progress can be made anymore). This is
convinient because it allows you to start your watchers (and your jobs),
call EV::loop once and when it returns you know that all your jobs are
finished (or they forgot to register some watchers for their task :).
Sometimes, however, this gets in your way, for example when the module
that calls EV::loop (usually the main program) is not the same module
as a long-living watcher (for example a DNS client module written by
somebody else even). Then you might want any outstanding requests to be
handled, but you would not want to keep EV::loop from returning just
because you happen to have this long-running UDP port watcher.
In this case you can clear the keepalive status, which means that even
though your watcher is active, it won't keep EV::loop from returning.
The initial value for keepalive is true (enabled), and you can change it any time.
Example: Register an I/O watcher for some UDP socket but do not keep the event loop from running just because of that watcher.
my $udp_socket = ...
my $udp_watcher = EV::io $udp_socket, EV::READ, sub { ... };
$udp_watcher->keepalive (0);
Each of the following subsections describes a single watcher type.
As long as the returned watcher object is alive, call the $callback
when at least one of events specified in $eventmask occurs.
The $eventmask can be one or more of these constants ORed together:
EV::READ wait until read() wouldn't block anymore EV::WRITE wait until write() wouldn't block anymore
The io_ns variant doesn't start (activate) the newly created watcher.
Calls the callback after $after seconds (which may be fractional). If
$repeat is non-zero, the timer will be restarted (with the $repeat
value as $after) after the callback returns.
This means that the callback would be called roughly after $after
seconds, and then every $repeat seconds. The timer does his best not
to drift, but it will not invoke the timer more often then once per event
loop iteration, and might drift in other cases. If that isn't acceptable,
look at EV::periodic, which can provide long-term stable timers.
The timer is based on a monotonic clock, that is, if somebody is sitting in front of the machine while the timer is running and changes the system clock, the timer will nevertheless run (roughly) the same time.
The timer_ns variant doesn't start (activate) the newly created watcher.
Similar to the start method, but has special semantics for repeating timers:
If the timer is active and non-repeating, it will be stopped.
If the timer is active and repeating, reset the timeout to occur
$repeat seconds after now.
If the timer is inactive and repeating, start it using the repeat value.
Otherwise do nothing.
This behaviour is useful when you have a timeout for some IO
operation. You create a timer object with the same value for $after and
$repeat, and then, in the read/write watcher, run the again method
on the timeout.
Similar to EV::timer, but is not based on relative timeouts but on absolute times. Apart from creating "simple" timers that trigger "at" the specified time, it can also be used for non-drifting absolute timers and more complex, cron-like, setups that are not adversely affected by time jumps (i.e. when the system clock is changed by explicit date -s or other means such as ntpd). It is also the most complex watcher type in EV.
It has three distinct "modes":
$at and doesn't repeat. It
will not adjust when a time jump occurs, that is, if it is to be run
at January 1st 2011 then it will run when the system time reaches or
surpasses this time.
In this mode the watcher will always be scheduled to time out at the
next $at + N * $interval time (for some integer N) and then repeat,
regardless of any time jumps.
This can be used to create timers that do not drift with respect to system time:
my $hourly = EV::periodic 0, 3600, 0, sub { print "once/hour\n" };
That doesn't mean there will always be 3600 seconds in between triggers, but only that the the clalback will be called when the system time shows a full hour (UTC).
Another way to think about it (for the mathematically inclined) is that
EV::periodic will try to run the callback in this mode at the next
possible time where $time = $at (mod $interval), regardless of any time
jumps.
In this mode $interval and $at are both being ignored. Instead, each time the periodic watcher gets scheduled, the reschedule callback ($reschedule_cb) will be called with the watcher as first, and the current time as second argument.
This callback MUST NOT stop or destroy this or any other periodic
watcher, ever, and MUST NOT call any event loop functions or methods. If
you need to stop it, return 1e30 and stop it afterwards. You may create
and start a EV::prepare watcher for this task.
It must return the next time to trigger, based on the passed time value (that is, the lowest time value larger than or equal to to the second argument). It will usually be called just before the callback will be triggered, but might be called at other times, too.
This can be used to create very complex timers, such as a timer that triggers on each midnight, local time (actually 24 hours after the last midnight, to keep the example simple. If you know a way to do it correctly in about the same space (without requiring elaborate modules), drop me a note :):
my $daily = EV::periodic 0, 0, sub {
my ($w, $now) = @_;
use Time::Local ();
my (undef, undef, undef, $d, $m, $y) = localtime $now;
86400 + Time::Local::timelocal 0, 0, 0, $d, $m, $y
}, sub {
print "it's midnight or likely shortly after, now\n";
};
The periodic_ns variant doesn't start (activate) the newly created watcher.
Call the callback when $signal is received (the signal can be specified by
number or by name, just as with kill or %SIG).
EV will grab the signal for the process (the kernel only allows one
component to receive a signal at a time) when you start a signal watcher,
and removes it again when you stop it. Perl does the same when you
add/remove callbacks to %SIG, so watch out.
You can have as many signal watchers per signal as you want.
The signal_ns variant doesn't start (activate) the newly created watcher.
Call the callback when a status change for pid $pid (or any pid
if $pid is 0) has been received (a status change happens when the
process terminates or is killed, or, when trace is true, additionally when
it is stopped or continued). More precisely: when the process receives
a SIGCHLD, EV will fetch the outstanding exit/wait status for all
changed/zombie children and call the callback.
It is valid (and fully supported) to install a child watcher after a child
has exited but before the event loop has started its next iteration (for
example, first you fork, then the new child process might exit, and
only then do you install a child watcher in the parent for the new pid).
You can access both exit (or tracing) status and pid by using the
rstatus and rpid methods on the watcher object.
You can have as many pid watchers per pid as you want, they will all be called.
The child_ns variant doesn't start (activate) the newly created watcher.
Call the callback when a file status change has been detected on
$path. The $path does not need to exist, changing from "path exists"
to "path does not exist" is a status change like any other.
The $interval is a recommended polling interval for systems where
OS-supported change notifications don't exist or are not supported. If
you use 0 then an unspecified default is used (which is highly
recommended!), which is to be expected to be around five seconds usually.
This watcher type is not meant for massive numbers of stat watchers, as even with OS-supported change notifications, this can be resource-intensive.
The stat_ns variant doesn't start (activate) the newly created watcher.
This call is very similar to the perl stat built-in: It stats (using
lstat) the path specified in the watcher and sets perls stat cache (as
well as EV's idea of the current stat values) to the values found.
In scalar context, a boolean is return indicating success or failure of the stat. In list context, the same 13-value list as with stat is returned (except that the blksize and blocks fields are not reliable).
In the case of an error, errno is set to ENOENT (regardless of the
actual error value) and the nlink value is forced to zero (if the stat
was successful then nlink is guaranteed to be non-zero).
See also the next two entries for more info.
$w->stat, but without the initial stat'ing: this returns
the values most recently detected by EV. See the next entry for more info.
Just like $w->stat, but without the initial stat'ing: this returns
the previous set of values, before the change.
That is, when the watcher callback is invoked, $w->prev will be set
to the values found before a change was detected, while $w->attr
returns the values found leading to the change detection. The difference (if any)
between prev and attr is what triggered the callback.
If you did something to the filesystem object and do not want to trigger
yet another change, you can call stat to update EV's idea of what the
current attributes are.
Call the callback when there are no other pending watchers of the same or higher priority (excluding check, prepare and other idle watchers of the same or lower priority, of course). They are called idle watchers because when the watcher is the highest priority pending event in the process, the process is considered to be idle at that priority.
If you want a watcher that is only ever called when no other events are
outstanding you have to set the priority to EV::MINPRI.
The process will not block as long as any idle watchers are active, and they will be called repeatedly until stopped.
For example, if you have idle watchers at priority 0 and 1, and
an I/O watcher at priority 0, then the idle watcher at priority 1
and the I/O watcher will always run when ready. Only when the idle watcher
at priority 1 is stopped and the I/O watcher at priority 0 is not
pending with the 0-priority idle watcher be invoked.
The idle_ns variant doesn't start (activate) the newly created watcher.
Call the callback just before the process would block. You can still create/modify any watchers at this point.
See the EV::check watcher, below, for explanations and an example.
The prepare_ns variant doesn't start (activate) the newly created watcher.
Call the callback just after the process wakes up again (after it has gathered events), but before any other callbacks have been invoked.
This is used to integrate other event-based software into the EV mainloop: You register a prepare callback and in there, you create io and timer watchers as required by the other software. Here is a real-world example of integrating Net::SNMP (with some details left out):
our @snmp_watcher;
our $snmp_prepare = EV::prepare sub {
# do nothing unless active
$dispatcher->{_event_queue_h}
or return;
# make the dispatcher handle any outstanding stuff
... not shown
# create an I/O watcher for each and every socket
@snmp_watcher = (
(map { EV::io $_, EV::READ, sub { } }
keys %{ $dispatcher->{_descriptors} }),
EV::timer +($event->[Net::SNMP::Dispatcher::_ACTIVE]
? $event->[Net::SNMP::Dispatcher::_TIME] - EV::now : 0),
0, sub { },
);
};
The callbacks are irrelevant (and are not even being called), the only purpose of those watchers is to wake up the process as soon as one of those events occurs (socket readable, or timer timed out). The corresponding EV::check watcher will then clean up:
our $snmp_check = EV::check sub {
# destroy all watchers
@snmp_watcher = ();
# make the dispatcher handle any new stuff
... not shown
};
The callbacks of the created watchers will not be called as the watchers are destroyed before this can happen (remember EV::check gets called first).
The check_ns variant doesn't start (activate) the newly created watcher.
Fork watchers are called when a fork () was detected. The invocation
is done before the event loop blocks next and before check watchers
are being called, and only in the child after the fork.
Call the callback before the event loop is resumed in the child process after a fork.
The fork_ns variant doesn't start (activate) the newly created watcher.
This is a rather advanced watcher type that lets you embed one event loop into another (currently only IO events are supported in the embedded loop, other types of watchers might be handled in a delayed or incorrect fashion and must not be used).
See the libev documentation at http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod#code_ev_embed_code_when_one_backend_ (locally installed as EV::libev) for more details.
In short, this watcher is most useful on BSD systems without working kqueue to still be able to handle a large number of sockets:
my $socket_loop;
# check wether we use SELECT or POLL _and_ KQUEUE is supported
if (
(EV::backend & (EV::BACKEND_POLL | EV::BACKEND_SELECT))
&& (EV::supported_backends & EV::embeddable_backends & EV::BACKEND_KQUEUE)
) {
# use kqueue for sockets
$socket_loop = new EV::Loop EV::BACKEND_KQUEUE | EV::FLAG_NOENV;
}
# use the default loop otherwise
$socket_loop ||= EV::default_loop;
Call the callback when the embedded event loop ($otherloop) has any
I/O activity. The $callback is optional: if it is missing, then the
embedded event loop will be managed automatically (which is recommended),
otherwise you have to invoke sweep yourself.
The embed_ns variant doesn't start (activate) the newly created watcher.
Async watchers are provided by EV, but have little use in perl directly, as perl neither supports threads nor direct access to signal handlers or other contexts where they could be of value.
It is, however, possible to use them from the XS level.
Please see the libev documentation for further details.
While Perl signal handling (%SIG) is not affected by EV, the behaviour
with EV is as the same as any other C library: Perl-signals will only be
handled when Perl runs, which means your signal handler might be invoked
only the next time an event callback is invoked.
The solution is to use EV signal watchers (see EV::signal), which will
ensure proper operations with regards to other event watchers.
If you cannot do this for whatever reason, you can also force a watcher
to be called on every event loop iteration by installing a EV::check
watcher:
my $async_check = EV::check sub { };
This ensures that perl gets into control for a short time to handle any pending signals, and also ensures (slightly) slower overall operation.
Threads are not supported by this module in any way. Perl pseudo-threads is evil stuff and must die. As soon as Perl gains real threads I will work on thread support for it.
Most of the "improved" event delivering mechanisms of modern operating systems have quite a few problems with fork(2) (to put it bluntly: it is not supported and usually destructive). Libev makes it possible to work around this by having a function that recreates the kernel state after fork in the child.
On non-win32 platforms, this module requires the pthread_atfork functionality to do this automatically for you. This function is quite buggy on most BSDs, though, so YMMV. The overhead for this is quite negligible, because everything the function currently does is set a flag that is checked only when the event loop gets used the next time, so when you do fork but not use EV, the overhead is minimal.
On win32, there is no notion of fork so all this doesn't apply, of course.
EV::ADNS (asynchronous DNS), Glib::EV (makes Glib/Gtk2 use EV as event loop), EV::Glib (embed Glib into EV), Coro::EV (efficient coroutines with EV), Net::SNMP::EV (asynchronous SNMP), AnyEvent for event-loop agnostic and portable event driven programming.
Marc Lehmann <schmorp@schmorp.de> http://home.schmorp.de/