class Thread::ConditionVariable
ConditionVariable
objects augment class Mutex
. Using condition variables, it is possible to suspend while in the middle of a critical section until a condition is met, such as a resource becomes available.
Due to non-deterministic scheduling and spurious wake-ups, users of condition variables should always use a separate boolean predicate (such as reading from a boolean variable) to check if the condition is actually met before starting to wait, and should wait in a loop, re-checking the condition every time the ConditionVariable
is waken up. The idiomatic way of using condition variables is calling the wait
method in an until
loop with the predicate as the loop condition.
condvar.wait(mutex) until condition_is_met
In the example below, we use the boolean variable resource_available
(which is protected by mutex
) to indicate the availability of the resource, and use condvar
to wait for that variable to become true. Note that:
-
Thread
b
may be scheduled before threada1
anda2
, and may run so fast that it have already made the resource available before eithera1
ora2
starts. Therefore,a1
anda2
should check ifresource_available
is already true before starting to wait. -
The
wait
method may spuriously wake up without signalling. Therefore, threada1
anda2
should recheckresource_available
after thewait
method returns, and go back to wait if the condition is not actually met. -
It is possible that thread
a2
starts right after threada1
is waken up byb
.Thread
a2
may have acquired themutex
and consumed the resource before threada1
acquires themutex
. This necessitates rechecking afterwait
, too.
Example:
mutex = Thread::Mutex.new resource_available = false condvar = Thread::ConditionVariable.new a1 = Thread.new { # Thread 'a1' waits for the resource to become available and consumes # the resource. mutex.synchronize { condvar.wait(mutex) until resource_available # After the loop, 'resource_available' is guaranteed to be true. resource_available = false puts "a1 consumed the resource" } } a2 = Thread.new { # Thread 'a2' behaves like 'a1'. mutex.synchronize { condvar.wait(mutex) until resource_available resource_available = false puts "a2 consumed the resource" } } b = Thread.new { # Thread 'b' periodically makes the resource available. loop { mutex.synchronize { resource_available = true # Notify one waiting thread if any. It is possible that neither # 'a1' nor 'a2 is waiting on 'condvar' at this moment. That's OK. condvar.signal } sleep 1 } } # Eventually both 'a1' and 'a2' will have their resources, albeit in an # unspecified order. [a1, a2].each {|th| th.join}
Public Class Methods
Source
static VALUE rb_condvar_initialize(VALUE self) { struct rb_condvar *cv = condvar_ptr(self); ccan_list_head_init(&cv->waitq); return self; }
Creates a new condition variable instance.
Public Instance Methods
Source
static VALUE rb_condvar_broadcast(VALUE self) { struct rb_condvar *cv = condvar_ptr(self); wakeup_all(&cv->waitq); return self; }
Wakes up all threads waiting for this lock.
Source
static VALUE rb_condvar_signal(VALUE self) { struct rb_condvar *cv = condvar_ptr(self); wakeup_one(&cv->waitq); return self; }
Wakes up the first thread in line waiting for this lock.
Source
static VALUE rb_condvar_wait(int argc, VALUE *argv, VALUE self) { rb_execution_context_t *ec = GET_EC(); struct rb_condvar *cv = condvar_ptr(self); struct sleep_call args; rb_scan_args(argc, argv, "11", &args.mutex, &args.timeout); struct sync_waiter sync_waiter = { .self = args.mutex, .th = ec->thread_ptr, .fiber = nonblocking_fiber(ec->fiber_ptr) }; ccan_list_add_tail(&cv->waitq, &sync_waiter.node); return rb_ensure(do_sleep, (VALUE)&args, delete_from_waitq, (VALUE)&sync_waiter); }
Releases the lock held in mutex
and waits; reacquires the lock on wakeup.
If timeout
is given, this method returns after timeout
seconds passed, even if no other thread doesn’t signal.
This method may wake up spuriously due to underlying implementation details.
Returns the slept result on mutex
.