def test_scaled_multiple():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
dur = 15.0
lat = 0.10
log = [] # [(timeout, 'name', priority, delay, latency), ...]
# Schedule a bunch of events, for immediate execution at the same
# instant, and 0.0 delay on reschedule. This will force the scheduler
# to select them based on priority scaling of their age
mix = [
('a -1/2', 0.50),
('b 1-' , 1),
('c 1-1/4', 1.25),
('d 2-', 2.00),
('e 1-', 1.00),
('f 1-1/2', 1.50),
('g 0-1/4', 0.25),
('h 3-1/4', 3.25),
('i -1/8', 0.125),
]
for name, prio in mix:
sch.enterabs(beg, prio,
reschedule, (log, beg + dur, sch, name, prio, 0.0, lat),
kwargs={'event_time':beg})
siz = len(sch.queue)
sch.run()
# Given a certain function latency 'lat', and a total duration of
# 'dur', we can compute roughly how many times reschedule should have
# triggered... Of course,
print( "\ntime name overdue" )
for expiry, name, prio, delay, lat, overdue in log:
print( "%7.3f %-10s %7.3fs" % (expiry - beg, name, overdue) )
runs = len(log); assert dur/lat * .9 < runs-siz < dur/lat * 1.1
# Count up the incidence of each type of event, and make sure their
# distribution reflects their relative priority.
prio_sum = 0.
for name, prio in mix:
prio_sum += prio
ovr = {}
rel = {}
for now, name, prio, delay, latency, overdue in log:
try: rel[name] += 1
except: rel[name] = 1
try: ovr[name] += overdue
except: ovr[name] = overdue
for name, prio in mix:
strength = prio / prio_sum
incidence = rel[name] / len(log)
error = (incidence - strength) / strength
overdue = ovr[name]/rel[name] # Average overdue in seconds
print( "%3d x %-10s prio == %7f; %4.2fs overdue; strengh == % 5.1f%% vs. % 5.1f%% incidence (% 5.1f%% err.)" % (
rel[name], name, prio, overdue, strength*100, incidence*100, error*100))
# Within 5 percentage points of where they should be... This is a very loose metric,
# because (especially for very low priority events), even one extra scheduling out of 100
# may result in a very high error "percentage".
assert abs(incidence - strength) < .05
def test_scaled_basic():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
when = []
sch.enter( 1.0, 0, lambda: when.append(timer()), () )
sch.run()
now = timer()
elapsed = now - beg; assert 0.90 < elapsed < 1.10 # Total run should take ~1 second
delayed =when[0] - beg; assert 0.90 < delayed < 1.10
def test_scaled_cancellation():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
when = []
e = sch.enterabs( beg + 2.0, 0, lambda: when.append(timer()), () )
threading.Timer( 0.5, lambda: sch.cancel(e)).start()
sch.run()
now = timer()
elapsed = now - beg; assert 0.40 < elapsed < 0.60 # Total run should take ~1/2 second
assert len(when) == 0 # And no events should fire
def test_scaled_preemption():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
when = []
sch.enterabs( beg + 2.0, 0, lambda: when.append(timer()), () )
threading.Timer( 0.5, lambda: sch.enterabs( beg + 1.0, 0, lambda: when.append(timer()), () )).start()
sch.run()
now = timer()
elapsed = now - beg; assert 1.90 < elapsed < 2.10 # Total run should take ~2 seconds
delay00 = when[0] - beg; assert 0.90 < delay00 < 1.10 # The Timer-scheduled event, now 1s ago
delay01 = when[1] - beg; assert 1.90 < delay01 < 2.10 # The original event, just expired
def test_scaled_cancellation():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
when = []
e = sch.enterabs(beg + 2.0, 0, lambda: when.append(timer()), ())
threading.Timer(0.5, lambda: sch.cancel(e)).start()
sch.run()
now = timer()
elapsed = now - beg
assert 0.40 < elapsed < 0.60 # Total run should take ~1/2 second
assert len(when) == 0 # And no events should fire
def test_scaled_basic():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
when = []
sch.enter(1.0, 0, lambda: when.append(timer()), ())
sch.run()
now = timer()
elapsed = now - beg
assert 0.90 < elapsed < 1.10 # Total run should take ~1 second
delayed = when[0] - beg
assert 0.90 < delayed < 1.10
def test_scaled_preemption():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
when = []
sch.enterabs(beg + 2.0, 0, lambda: when.append(timer()), ())
threading.Timer(
0.5, lambda: sch.enterabs(beg + 1.0, 0, lambda: when.append(timer()),
())).start()
sch.run()
now = timer()
elapsed = now - beg
assert 1.90 < elapsed < 2.10 # Total run should take ~2 seconds
delay00 = when[0] - beg
assert 0.90 < delay00 < 1.10 # The Timer-scheduled event, now 1s ago
delay01 = when[1] - beg
assert 1.90 < delay01 < 2.10 # The original event, just expired
def test_scaled_multiple():
sch = sched.scaled_scheduler(timer, sleep)
beg = timer()
dur = 15.0
lat = 0.10
log = [] # [(timeout, 'name', priority, delay, latency), ...]
# Schedule a bunch of events, for immediate execution at the same
# instant, and 0.0 delay on reschedule. This will force the scheduler
# to select them based on priority scaling of their age
mix = [
('a -1/2', 0.50),
('b 1-', 1),
('c 1-1/4', 1.25),
('d 2-', 2.00),
('e 1-', 1.00),
('f 1-1/2', 1.50),
('g 0-1/4', 0.25),
('h 3-1/4', 3.25),
('i -1/8', 0.125),
]
for name, prio in mix:
sch.enterabs(beg,
prio,
reschedule, (log, beg + dur, sch, name, prio, 0.0, lat),
kwargs={'event_time': beg})
siz = len(sch.queue)
sch.run()
# Given a certain function latency 'lat', and a total duration of
# 'dur', we can compute roughly how many times reschedule should have
# triggered... Of course,
print("\ntime name overdue")
for expiry, name, prio, delay, lat, overdue in log:
print("%7.3f %-10s %7.3fs" % (expiry - beg, name, overdue))
runs = len(log)
assert dur / lat * .9 < runs - siz < dur / lat * 1.1
# Count up the incidence of each type of event, and make sure their
# distribution reflects their relative priority.
prio_sum = 0.
for name, prio in mix:
prio_sum += prio
ovr = {}
rel = {}
for now, name, prio, delay, latency, overdue in log:
try:
rel[name] += 1
except:
rel[name] = 1
try:
ovr[name] += overdue
except:
ovr[name] = overdue
for name, prio in mix:
strength = prio / prio_sum
incidence = rel[name] / len(log)
error = (incidence - strength) / strength
overdue = ovr[name] / rel[name] # Average overdue in seconds
print(
"%3d x %-10s prio == %7f; %4.2fs overdue; strengh == % 5.1f%% vs. % 5.1f%% incidence (% 5.1f%% err.)"
% (rel[name], name, prio, overdue, strength * 100, incidence * 100,
error * 100))
# Within 5 percentage points of where they should be... This is a very loose metric,
# because (especially for very low priority events), even one extra scheduling out of 100
# may result in a very high error "percentage".
assert abs(incidence - strength) < .05
