Documentation for SimPy¶
Contents:
Overview¶
SimPy is a process-based discrete-event simulation framework based on standard Python.
Processes in SimPy are defined by Python generator functions and may, for example, be used to model active components like customers, vehicles or agents. SimPy also provides various types of shared resources to model limited capacity congestion points (like servers, checkout counters and tunnels).
Simulations can be performed “as fast as possible”, in real time (wall clock time) or by manually stepping through the events.
Though it is theoretically possible to do continuous simulations with SimPy, it has no features that help you with that. On the other hand, SimPy is overkill for simulations with a fixed step size where your processes don’t interact with each other or with shared resources.
A short example simulating two clocks ticking in different time intervals looks like this:
>>> import simpy
>>>
>>> def clock(env, name, tick):
... while True:
... print(name, env.now)
... yield env.timeout(tick)
...
>>> env = simpy.Environment()
>>> env.process(clock(env, 'fast', 0.5))
<Process(clock) object at 0x...>
>>> env.process(clock(env, 'slow', 1))
<Process(clock) object at 0x...>
>>> env.run(until=2)
fast 0
slow 0
fast 0.5
slow 1
fast 1.0
fast 1.5
The documentation contains a tutorial, several guides explaining key concepts, a number of examples and the API reference.
SimPy is released under the MIT License. Simulation model developers are encouraged to share their SimPy modeling techniques with the SimPy community. Please post a message to the SimPy mailing list.
There is an introductory talk that explains SimPy’s concepts and provides some examples: watch the video or get the slides.
SimPy has also been reimplemented in other programming languages. See the list of ports for more details.
SimPy in 10 Minutes¶
In this section, you’ll learn the basics of SimPy in just a few minutes. Afterwards, you will be able to implement a simple simulation using SimPy and you’ll be able to make an educated decision if SimPy is what you need. We’ll also give you some hints on how to proceed to implement more complex simulations.
Installation¶
SimPy is implemented in pure Python and has no dependencies. SimPy runs on Python 2 (>= 2.7) and Python 3 (>= 3.2). PyPy is also supported. If you have pip installed, just type
$ pip install simpy
and you are done.
Installing from source¶
Alternatively, you can download SimPy and install it manually. Extract the archive, open a terminal window where you extracted SimPy and type:
$ python setup.py install
You can now optionally run SimPy’s tests to see if everything works fine. You need pytest for this. Run the following command within the source directory of SimPy:
$ py.test --pyargs simpy
Upgrading from SimPy 2¶
If you are already familiar with SimPy 2, please read the Guide Porting from SimPy 2 to 3.
What’s Next¶
Now that you’ve installed SimPy, you probably want to simulate something. The next section will introduce you to SimPy’s basic concepts.
Basic Concepts¶
SimPy is a discrete-event simulation library. The behavior of active components (like vehicles, customers or messages) is modeled with processes. All processes live in an environment. They interact with the environment and with each other via events.
Processes are described by simple Python generators. You can call
them process function or process method, depending on whether it is
a normal function or method of a class. During their lifetime, they create
events and yield
them in order to wait for them to be triggered.
When a process yields an event, the process gets suspended. SimPy resumes the process, when the event occurs (we say that the event is triggered). Multiple processes can wait for the same event. SimPy resumes them in the same order in which they yielded that event.
An important event type is the Timeout
. Events of this
type are triggered after a certain amount of (simulated) time has passed. They
allow a process to sleep (or hold its state) for the given time.
A Timeout
and all other events can be created by calling
the appropriate method of the Environment
that the process lives in
(Environment.timeout()
for example).
Our First Process¶
Our first example will be a car process. The car will alternately drive and park for a while. When it starts driving (or parking), it will print the current simulation time.
So let’s start:
>>> def car(env):
... while True:
... print('Start parking at %d' % env.now)
... parking_duration = 5
... yield env.timeout(parking_duration)
...
... print('Start driving at %d' % env.now)
... trip_duration = 2
... yield env.timeout(trip_duration)
Our car process requires a reference to an Environment
(env
) in
order to create new events. The car’s behavior is described in an infinite
loop. Remember, this function is a generator. Though it will never terminate,
it will pass the control flow back to the simulation once a yield
statement
is reached. Once the yielded event is triggered (“it occurs”), the simulation
will resume the function at this statement.
As I said before, our car switches between the states parking and driving.
It announces its new state by printing a message and the current simulation
time (as returned by the Environment.now
property). It then calls the
Environment.timeout()
factory function to create
a Timeout
event. This event describes the point in time
the car is done parking (or driving, respectively). By yielding the event,
it signals the simulation that it wants to wait for the event to occur.
Now that the behavior of our car has been modeled, lets create an instance of it and see how it behaves:
>>> import simpy
>>> env = simpy.Environment()
>>> env.process(car(env))
<Process(car) object at 0x...>
>>> env.run(until=15)
Start parking at 0
Start driving at 5
Start parking at 7
Start driving at 12
Start parking at 14
The first thing we need to do is to create an instance of Environment
.
This instance is passed into our car process function. Calling it creates
a process generator that needs to be started and added to the environment via
Environment.process()
.
Note, that at this time, none of the code of our process function is being executed. Its execution is merely scheduled at the current simulation time.
The Process
returned by process()
can be used for process interactions (we will cover that in the next section,
so we will ignore it for now).
Finally, we start the simulation by calling run()
and
passing an end time to it.
What’s Next?¶
You should now be familiar with SimPy’s terminology and basic concepts. In the next section, we will cover process interaction.
Process Interaction¶
The Process
instance that is returned by
Environment.process()
can be utilized for process interactions. The two
most common examples for this are to wait for another process to finish and to
interrupt another process while it is waiting for an event.
Waiting for a Process¶
As it happens, a SimPy Process
can be used like an event
(technically, a process actually is an event). If you yield it, you are
resumed once the process has finished. Imagine a car-wash simulation where cars
enter the car-wash and wait for the washing process to finish. Or an airport
simulation where passengers have to wait until a security check finishes.
Lets assume that the car from our last example magically became an electric vehicle. Electric vehicles usually take a lot of time charging their batteries after a trip. They have to wait until their battery is charged before they can start driving again.
We can model this with an additional charge()
process for our car.
Therefore, we refactor our car to be a class with two process methods:
run()
(which is the original car()
process function) and charge()
.
The run
process is automatically started when Car
is instantiated.
A new charge
process is started every time the vehicle starts parking. By
yielding the Process
instance that
Environment.process()
returns, the run
process starts waiting for
it to finish:
>>> class Car(object):
... def __init__(self, env):
... self.env = env
... # Start the run process everytime an instance is created.
... self.action = env.process(self.run())
...
... def run(self):
... while True:
... print('Start parking and charging at %d' % self.env.now)
... charge_duration = 5
... # We yield the process that process() returns
... # to wait for it to finish
... yield self.env.process(self.charge(charge_duration))
...
... # The charge process has finished and
... # we can start driving again.
... print('Start driving at %d' % self.env.now)
... trip_duration = 2
... yield self.env.timeout(trip_duration)
...
... def charge(self, duration):
... yield self.env.timeout(duration)
Starting the simulation is straightforward again: We create an environment,
one (or more) cars and finally call run()
.
>>> import simpy
>>> env = simpy.Environment()
>>> car = Car(env)
>>> env.run(until=15)
Start parking and charging at 0
Start driving at 5
Start parking and charging at 7
Start driving at 12
Start parking and charging at 14
Interrupting Another Process¶
Imagine, you don’t want to wait until your electric vehicle is fully charged but want to interrupt the charging process and just start driving instead.
SimPy allows you to interrupt a running process by calling its
interrupt()
method:
>>> def driver(env, car):
... yield env.timeout(3)
... car.action.interrupt()
The driver
process has a reference to the car’s action
process. After
waiting for 3 time steps, it interrupts that process.
Interrupts are thrown into process functions as
Interrupt
exceptions that can (should) be handled by
the interrupted process. The process can then decide what to do next (e.g.,
continuing to wait for the original event or yielding a new event):
>>> class Car(object):
... def __init__(self, env):
... self.env = env
... self.action = env.process(self.run())
...
... def run(self):
... while True:
... print('Start parking and charging at %d' % self.env.now)
... charge_duration = 5
... # We may get interrupted while charging the battery
... try:
... yield self.env.process(self.charge(charge_duration))
... except simpy.Interrupt:
... # When we received an interrupt, we stop charging and
... # switch to the "driving" state
... print('Was interrupted. Hope, the battery is full enough ...')
...
... print('Start driving at %d' % self.env.now)
... trip_duration = 2
... yield self.env.timeout(trip_duration)
...
... def charge(self, duration):
... yield self.env.timeout(duration)
When you compare the output of this simulation with the previous example,
you’ll notice that the car now starts driving at time 3
instead of 5
:
>>> env = simpy.Environment()
>>> car = Car(env)
>>> env.process(driver(env, car))
<Process(driver) object at 0x...>
>>> env.run(until=15)
Start parking and charging at 0
Was interrupted. Hope, the battery is full enough ...
Start driving at 3
Start parking and charging at 5
Start driving at 10
Start parking and charging at 12
What’s Next¶
We just demonstrated two basic methods for process interactions—waiting for
a process and interrupting a process. Take a look at the
Topical Guides or the Process
API
reference for more details.
In the next section we will cover the basic usage of shared resources.
How to Proceed¶
If you are not certain yet if SimPy fulfills your requirements or if you want to see more features in action, you should take a look at the various examples we provide.
If you are looking for a more detailed description of a certain aspect or feature of SimPy, the Topical Guides section might help you.
Finally, there is an API Reference that describes all functions and classes in full detail.
Topical Guides¶
This sections covers various aspects of SimPy more in-depth. It assumes that you have a basic understanding of SimPy’s capabilities and that you know what you are looking for.
SimPy basics¶
This guide describes the basic concepts of SimPy: How does it work? What are processes, events and the environment? What can I do with them?
How SimPy works¶
If you break SimPy down, it is just an asynchronous event dispatcher. You generate events and schedule them at a given simulation time. Events are sorted by priority, simulation time, and an increasing event id. An event also has a list of callbacks, which are executed when the event is triggered and processed by the event loop. Events may also have a return value.
The components involved in this are the Environment
,
events
and the process functions that you write.
Process functions implement your simulation model, that is, they define the
behavior of your simulation. They are plain Python generator functions that
yield instances of Event
.
The environment stores these events in its event list and keeps track of the current simulation time.
If a process function yields an event, SimPy adds the process to the event’s callbacks and suspends the process until the event is triggered and processed. When a process waiting for an event is resumed, it will also receive the event’s value.
Here is a very simple example that illustrates all this; the code is more verbose than it needs to be to make things extra clear. You find a compact version of it at the end of this section:
>>> import simpy
>>>
>>> def example(env):
... event = simpy.events.Timeout(env, delay=1, value=42)
... value = yield event
... print('now=%d, value=%d' % (env.now, value))
>>>
>>> env = simpy.Environment()
>>> example_gen = example(env)
>>> p = simpy.events.Process(env, example_gen)
>>>
>>> env.run()
now=1, value=42
The example()
process function above first creates
a Timeout
event. It passes the environment, a delay, and
a value to it. The Timeout schedules itself at now + delay
(that’s why the
environment is required); other event types usually schedule themselves at the
current simulation time.
The process function then yields the event and thus gets suspended. It is
resumed, when SimPy processes the Timeout event. The process function also
receives the event’s value (42) – this is, however, optional, so yield
event
would have been okay if the you were not interested in the value or if
the event had no value at all.
Finally, the process function prints the current simulation time (that is
accessible via the environment’s now
attribute)
and the Timeout’s value.
If all required process functions are defined, you can instantiate all objects
for your simulation. In most cases, you start by creating an instance of
Environment
, because you’ll need to pass it around a lot
when creating everything else.
Starting a process function involves two things:
- You have to call the process function to create a generator object. (This will not execute any code of that function yet. Please read The Python yield keyword explained, to understand why this is the case.)
- You then create an instance of
Process
and pass the environment and the generator object to it. This will schedule anInitialize
event at the current simulation time which starts the execution of the process function. The process instance is also an event that is triggered when the process function returns. The guide to events explains why this is handy.
Finally, you can start SimPy’s event loop. By default, it will run as long as
there are events in the event list, but you can also let it stop earlier by
providing an until
argument (see Simulation control).
The following guides describe the environment and its interactions with events and process functions in more detail.
“Best practice” version of the example above¶
>>> import simpy
>>>
>>> def example(env):
... value = yield env.timeout(1, value=42)
... print('now=%d, value=%d' % (env.now, value))
>>>
>>> env = simpy.Environment()
>>> p = env.process(example(env))
>>> env.run()
now=1, value=42
Environments¶
A simulation environment manages the simulation time as well as the scheduling and processing of events. It also provides means to step through or execute the simulation.
The base class for all environments is BaseEnvironment
.
“Normal” simulations usually use its subclass
Environment
. For real-time simulations, SimPy provides a
RealtimeEnvironment
(more on that in
Real-time simulations).
Simulation control¶
SimPy is very flexible in terms of simulation execution. You can run your simulation until there are no more events, until a certain simulation time is reached, or until a certain event is triggered. You can also step through the simulation event by event. Furthermore, you can mix these things as you like.
For example, you could run your simulation until an interesting event occurs. You could then step through the simulation event by event for a while; and finally run the simulation until there are no more events left and your processes have all terminated.
The most important method here is Environment.run()
:
If you call it without any argument (
env.run()
), it steps through the simulation until there are no more events left.Warning
If your processes run forever (
while True: yield env.timeout(1)
), this method will never terminate (unless you kill your script by e.g., pressing Ctrl-C).In most cases it is advisable to stop your simulation when it reaches a certain simulation time. Therefore, you can pass the desired time via the until parameter, e.g.:
env.run(until=10)
.The simulation will then stop when the internal clock reaches 10 but will not process any events scheduled for time 10. This is similar to a new environment where the clock is 0 but (obviously) no events have yet been processed.
If you want to integrate your simulation in a GUI and want to draw a process bar, you can repeatedly call this function with increasing until values and update your progress bar after each call:
for i in range(100): env.run(until=i) progressbar.update(i)
Instead of passing a number to
run()
, you can also pass any event to it.run()
will then return when the event has been processed.Assuming that the current time is 0,
env.run(until=env.timeout(5))
is equivalent toenv.run(until=5)
.You can also pass other types of events (remember, that a
Process
is an event, too):>>> import simpy >>> >>> def my_proc(env): ... yield env.timeout(1) ... return 'Monty Python’s Flying Circus' >>> >>> env = simpy.Environment() >>> proc = env.process(my_proc(env)) >>> env.run(until=proc) 'Monty Python’s Flying Circus'
To step through the simulation event by event, the environment offers
peek()
and step()
.
peek()
returns the time of the next scheduled event or infinity
(float('inf')
) if no future events are scheduled.
step()
processes the next scheduled event. It raises an
EmptySchedule
exception if no event is available.
In a typical use case, you use these methods in a loop like:
until = 10
while env.peek() < until:
env.step()
State access¶
The environment allows you to get the current simulation time via the
Environment.now
property. The simulation time is a number without unit
and is increased via Timeout
events.
By default, now
starts at 0, but you can pass an initial_time
to the
Environment
to use something else.
Note
Although the simulation time is technically unitless, you can pretend that
it is, for example, in seconds and use it like a timestamp returned by
time.time()
to calculate a date or the day of the week.
The property Environment.active_process
is comparable to
os.getpid()
and is either None
or pointing at the currently active
Process
. A process is active when its process function
is being executed. It becomes inactive (or suspended) when it yields an
event.
Thus, it only makes sense to access this property from within a process function or a function that is called by your process function:
>>> def subfunc(env):
... print(env.active_process) # will print "p1"
>>>
>>> def my_proc(env):
... while True:
... print(env.active_process) # will print "p1"
... subfunc(env)
... yield env.timeout(1)
>>>
>>> env = simpy.Environment()
>>> p1 = env.process(my_proc(env))
>>> env.active_process # None
>>> env.step()
<Process(my_proc) object at 0x...>
<Process(my_proc) object at 0x...>
>>> env.active_process # None
An exemplary use case for this is the resource system: If a process function
calls request()
to request
a resource, the resource determines the requesting process via
env.active_process
. Take a look at the code to see how we do this :-).
Event creation¶
To create events, you normally have to import simpy.events
, instantiate
the event class and pass a reference to the environment to it. To reduce the
amount of typing, the Environment
provides some shortcuts for event
creation. For example, Environment.event()
is equivalent to
simpy.events.Event(env)
.
Other shortcuts are:
More details on what the events do can be found in the guide to events.
Miscellaneous¶
Since Python 3.3, a generator function can have a return value:
def my_proc(env):
yield env.timeout(1)
return 42
In SimPy, this can be used to provide return values for processes that can be used by other processes:
def other_proc(env):
ret_val = yield env.process(my_proc(env))
assert ret_val == 42
Internally, Python passes the return value as parameter to the
StopIteration
exception that it raises when a generator is exhausted. So
in Python 2.7 and 3.2 you could replace the return 42
with a raise
StopIteration(42)
to achieve the same result.
To keep your code more readable, the environment provides the method
exit()
to do exactly this:
def my_proc(env):
yield env.timeout(1)
env.exit(42) # Py2 equivalent to "return 42"
Events¶
SimPy includes an extensive set of event types for various purposes. All of
them inherit simpy.events.Event
. The listing below shows the
hierarchy of events built into SimPy:
events.Event
|
+— events.Timeout
|
+— events.Initialize
|
+— events.Process
|
+— events.Condition
| |
| +— events.AllOf
| |
| +— events.AnyOf
.
.
.
This is the set of basic events. Events are extensible and resources, for
example, define additional events. In this guide, we’ll focus on the events in
the simpy.events
module. The guide to resources
describes the various resource events.
Event basics¶
SimPy events are very similar – if not identical — to deferreds, futures or
promises. Instances of the class Event
are used to describe any kind
of events. Events can be in one of the following states. An event
- might happen (not triggered),
- is going to happen (triggered) or
- has happened (processed).
They traverse these states exactly once in that order. Events are also tightly bound to time and time causes events to advance their state.
Initially, events are not triggered and just objects in memory.
If an event gets triggered, it is scheduled at a given time and inserted into
SimPy’s event queue. The property Event.triggered
becomes True
.
As long as the event is not processed, you can add callbacks to an event.
Callbacks are callables that accept an event as parameter and are stored in the
Event.callbacks
list.
An event becomes processed when SimPy pops it from the event queue and
calls all of its callbacks. It is now no longer possible to add callbacks. The
property Event.processed
becomes True
.
Events also have a value. The value can be set before or when the event is
triggered and can be retrieved via Event.value
or, within a process, by
yielding the event (value = yield event
).
Adding callbacks to an event¶
“What? Callbacks? I’ve never seen no callbacks!”, you might think if you have worked your way through the tutorial.
That’s on purpose. The most common way to add a callback to an event is
yielding it from your process function (yield event
). This will add the
process’ _resume() method as a callback. That’s how your process gets resumed
when it yielded an event.
However, you can add any callable object (function) to the list of callbacks as long as it accepts an event instance as its single parameter:
>>> import simpy
>>>
>>> def my_callback(event):
... print('Called back from', event)
...
>>> env = simpy.Environment()
>>> event = env.event()
>>> event.callbacks.append(my_callback)
>>> event.callbacks
[<function my_callback at 0x...>]
If an event has been processed, all of its Event.callbacks
have been
executed and the attribute is set to None
. This is to prevent you from
adding more callbacks – these would of course never get called because the
event has already happened.
Processes are smart about this, though. If you yield a processed event, _resume() will immediately resume your process with the value of the event (because there is nothing to wait for).
Triggering events¶
When events are triggered, they can either succeed or fail. For example, if an event is to be triggered at the end of a computation and everything works out fine, the event will succeed. If an exceptions occurs during that computation, the event will fail.
To trigger an event and mark it as successful, you can use
Event.succeed(value=None)
. You can optionally pass a value to it (e.g.,
the results of a computation).
To trigger an event and mark it as failed, call Event.fail(exception)
and pass an Exception
instance to it (e.g., the exception you caught
during your failed computation).
There is also a generic way to trigger an event: Event.trigger(event)
.
This will take the value and outcome (success or failure) of the event passed
to it.
All three methods return the event instance they are bound to. This allows you
to do things like yield Event(env).succeed()
.
Example usages for Event
¶
The simple mechanics outlined above provide a great flexibility in the way
events (even the basic Event
) can be used.
One example for this is that events can be shared. They can be created by a process or outside of the context of a process. They can be passed to other processes and chained:
>>> class School:
... def __init__(self, env):
... self.env = env
... self.class_ends = env.event()
... self.pupil_procs = [env.process(self.pupil()) for i in range(3)]
... self.bell_proc = env.process(self.bell())
...
... def bell(self):
... for i in range(2):
... yield self.env.timeout(45)
... self.class_ends.succeed()
... self.class_ends = self.env.event()
... print()
...
... def pupil(self):
... for i in range(2):
... print(' \o/', end='')
... yield self.class_ends
...
>>> school = School(env)
>>> env.run()
\o/ \o/ \o/
\o/ \o/ \o/
This can also be used like the passivate / reactivate known from SimPy 2. The pupils passivate when class begins and are reactivated when the bell rings.
Let time pass by: the Timeout
¶
To actually let time pass in a simulation, there is the timeout event.
A timeout has two parameters: a delay and an optional value:
Timeout(delay, value=None)
. It triggers itself during its creation and
schedules itself at now + delay
. Thus, the succeed()
and fail()
methods cannot be called again and you have to pass the event value to it when
you create the timeout.
The delay can be any kind of number, usually an int or float as long as it supports comparison and addition.
Processes are events, too¶
SimPy processes (as created by Process
or env.process()
) have the
nice property of being events, too.
That means, that a process can yield another process. It will then be resumed when the other process ends. The event’s value will be the return value of that process:
>>> def sub(env):
... yield env.timeout(1)
... return 23
...
>>> def parent(env):
... ret = yield env.process(sub(env))
... return ret
...
>>> env.run(env.process(parent(env)))
23
The example above will only work in Python >= 3.3. As a workaround for older
Python versions, you can use env.exit(23)
with the same effect.
When a process is created, it schedules an Initialize
event which will
start the execution of the process when triggered. You usually won’t have to
deal with this type of event.
If you don’t want a process to start immediately but after a certain delay, you
can use simpy.util.start_delayed()
. This method returns a helper
process that uses a timeout before actually starting a process.
The example from above, but with a delayed start of sub()
:
>>> from simpy.util import start_delayed
>>>
>>> def sub(env):
... yield env.timeout(1)
... return 23
...
>>> def parent(env):
... start = env.now
... sub_proc = yield start_delayed(env, sub(env), delay=3)
... assert env.now - start == 3
...
... ret = yield sub_proc
... return ret
...
>>> env.run(env.process(parent(env)))
23
Waiting for multiple events at once¶
Sometimes, you want to wait for more than one event at the same time. For example, you may want to wait for a resource, but not for an unlimited amount of time. Or you may want to wait until all a set of events has happened.
SimPy therefore offers the AnyOf
and AllOf
events which both
are a Condition
event.
Both take a list of events as an argument and are triggered if at least one or all of them of them are triggered.
>>> from simpy.events import AnyOf, AllOf, Event
>>> events = [Event(env) for i in range(3)]
>>> a = AnyOf(env, events) # Triggers if at least one of "events" is triggered.
>>> b = AllOf(env, events) # Triggers if all each of "events" is triggered.
The value of a condition event is an ordered dictionary with an entry for every
triggered event. In the case of AllOf
, the size of that dictionary will
always be the same as the length of the event list. The value dict of AnyOf
will have at least one entry. In both cases, the event instances are used as
keys and the event values will be the values.
As a shorthand for AllOf
and AnyOf
, you can also use the logical
operators &
(and) and |
(or):
>>> def test_condition(env):
... t1, t2 = env.timeout(1, value='spam'), env.timeout(2, value='eggs')
... ret = yield t1 | t2
... assert ret == {t1: 'spam'}
...
... t1, t2 = env.timeout(1, value='spam'), env.timeout(2, value='eggs')
... ret = yield t1 & t2
... assert ret == {t1: 'spam', t2: 'eggs'}
...
... # You can also concatenate & and |
... e1, e2, e3 = [env.timeout(i) for i in range(3)]
... yield (e1 | e2) & e3
... assert all(e.processed for e in [e1, e2, e3])
...
>>> proc = env.process(test_condition(env))
>>> env.run()
The order of condition results is identical to the order in which the condition
events were specified. This allows the following idiom for conveniently
fetching the values of multiple events specified in an and condition
(including AllOf
):
>>> def fetch_values_of_multiple_events(env):
... t1, t2 = env.timeout(1, value='spam'), env.timeout(2, value='eggs')
... r1, r2 = (yield t1 & t2).values()
... assert r1 == 'spam' and r2 == 'eggs'
...
>>> proc = env.process(fetch_values_of_multiple_events(env))
>>> env.run()
Process Interaction¶
Discrete event simulation is only made interesting by interactions between processes.
So this guide is about:
- Sleep until woken up (passivate/reactivate)
- Waiting for another process to terminate
- Interrupting another process
The first two items were already covered in the Events guide, but we’ll also include them here for the sake of completeness.
Another possibility for processes to interact are resources. They are discussed in a separate guide.
Sleep until woken up¶
Imagine you want to model an electric vehicle with an intelligent battery-charging controller. While the vehicle is driving, the controller can be passive but needs to be reactivate once the vehicle is connected to the power grid in order to charge the battery.
In SimPy 2, this pattern was known as passivate / reactivate. In SimPy 3,
you can accomplish that with a simple, shared Event
:
>>> from random import seed, randint
>>> seed(23)
>>>
>>> import simpy
>>>
>>> class EV:
... def __init__(self, env):
... self.env = env
... self.drive_proc = env.process(self.drive(env))
... self.bat_ctrl_proc = env.process(self.bat_ctrl(env))
... self.bat_ctrl_reactivate = env.event()
...
... def drive(self, env):
... while True:
... # Drive for 20-40 min
... yield env.timeout(randint(20, 40))
...
... # Park for 1–6 hours
... print('Start parking at', env.now)
... self.bat_ctrl_reactivate.succeed() # "reactivate"
... self.bat_ctrl_reactivate = env.event()
... yield env.timeout(randint(60, 360))
... print('Stop parking at', env.now)
...
... def bat_ctrl(self, env):
... while True:
... print('Bat. ctrl. passivating at', env.now)
... yield self.bat_ctrl_reactivate # "passivate"
... print('Bat. ctrl. reactivated at', env.now)
...
... # Intelligent charging behavior here …
... yield env.timeout(randint(30, 90))
...
>>> env = simpy.Environment()
>>> ev = EV(env)
>>> env.run(until=150)
Bat. ctrl. passivating at 0
Start parking at 29
Bat. ctrl. reactivated at 29
Bat. ctrl. passivating at 60
Stop parking at 131
Since bat_ctrl()
just waits for a normal event, we no longer call this
pattern passivate / reactivate in SimPy 3.
Waiting for another process to terminate¶
The example above has a problem: it may happen that the vehicles wants to park for a shorter duration than it takes to charge the battery (this is the case if both, charging and parking would take 60 to 90 minutes).
To fix this problem we have to slightly change our model. A new bat_ctrl()
will be started every time the EV starts parking. The EV then waits until the
parking duration is over and until the charging has stopped:
>>> class EV:
... def __init__(self, env):
... self.env = env
... self.drive_proc = env.process(self.drive(env))
...
... def drive(self, env):
... while True:
... # Drive for 20-40 min
... yield env.timeout(randint(20, 40))
...
... # Park for 1–6 hours
... print('Start parking at', env.now)
... charging = env.process(self.bat_ctrl(env))
... parking = env.timeout(randint(60, 360))
... yield charging & parking
... print('Stop parking at', env.now)
...
... def bat_ctrl(self, env):
... print('Bat. ctrl. started at', env.now)
... # Intelligent charging behavior here …
... yield env.timeout(randint(30, 90))
... print('Bat. ctrl. done at', env.now)
...
>>> env = simpy.Environment()
>>> ev = EV(env)
>>> env.run(until=310)
Start parking at 29
Bat. ctrl. started at 29
Bat. ctrl. done at 83
Stop parking at 305
Again, nothing new (if you’ve read the Events guide) and special is
happening. SimPy processes are events, too, so you can yield them and will thus
wait for them to get triggered. You can also wait for two events at the same
time by concatenating them with &
(see
Waiting for multiple events at once).
Interrupting another process¶
As usual, we now have another problem: Imagine, a trip is very urgent, but with the current implementation, we always need to wait until the battery is fully charged. If we could somehow interrupt that …
Fortunate coincidence, there is indeed a way to do exactly this. You can call
interrupt()
on a Process
. This will throw an
Interrupt
exception into that process, resuming it
immediately:
>>> class EV:
... def __init__(self, env):
... self.env = env
... self.drive_proc = env.process(self.drive(env))
...
... def drive(self, env):
... while True:
... # Drive for 20-40 min
... yield env.timeout(randint(20, 40))
...
... # Park for 1 hour
... print('Start parking at', env.now)
... charging = env.process(self.bat_ctrl(env))
... parking = env.timeout(60)
... yield charging | parking
... if not charging.triggered:
... # Interrupt charging if not already done.
... charging.interrupt('Need to go!')
... print('Stop parking at', env.now)
...
... def bat_ctrl(self, env):
... print('Bat. ctrl. started at', env.now)
... try:
... yield env.timeout(randint(60, 90))
... print('Bat. ctrl. done at', env.now)
... except simpy.Interrupt as i:
... # Onoes! Got interrupted before the charging was done.
... print('Bat. ctrl. interrupted at', env.now, 'msg:',
... i.cause)
...
>>> env = simpy.Environment()
>>> ev = EV(env)
>>> env.run(until=100)
Start parking at 31
Bat. ctrl. started at 31
Stop parking at 91
Bat. ctrl. interrupted at 91 msg: Need to go!
What process.interrupt()
actually does is scheduling an
Interruption
event for immediate execution. If this
event is executed it will remove the victim process’ _resume()
method from
the callbacks of the event that it is currently waiting for (see
target
). Following that it will throw the
Interrupt
exception into the process.
Since we don’t do anything special to the original target event of the process,
the interrupted process can yield the same event again after catching the
Interrupt
– Imagine someone waiting for a shop to open. The person may get
interrupted by a phone call. After finishing the call, he or she checks if the
shop already opened and either enters or continues to wait.
Real-time simulations¶
Sometimes, you might not want to perform a simulation as fast as possible but synchronous to the wall-clock time. This kind of simulation is also called real-time simulation.
Real-time simulations may be necessary
- if you have hardware-in-the-loop,
- if there is human interaction with your simulation, or
- if you want to analyze the real-time behavior of an algorithm.
To convert a simulation into a real-time simulation, you only need to replace
SimPy’s default Environment
with
a simpy.rt.RealtimeEnvironment
. Apart from the initial_time
argument, there are two additional parameters: factor and strict:
RealtimeEnvironment(initial_time=0, factor=1.0, strict=True)
.
The factor defines how much real time passes with each step of simulation
time. By default, this is one second. If you set factor=0.1
, a unit of
simulation time will only take a tenth of a second; if you set factor=60
,
it will take a minute.
Here is a simple example for converting a normal simulation to a real-time simulation with a duration of one tenth of a second per simulation time unit:
>>> import time
>>> import simpy
>>>
>>> def example(env):
... start = time.perf_counter()
... yield env.timeout(1)
... end = time.perf_counter()
... print('Duration of one simulation time unit: %.2fs' % (end - start))
>>>
>>> env = simpy.Environment()
>>> proc = env.process(example(env))
>>> env.run(until=proc)
Duration of one simulation time unit: 0.00s
>>>
>>> import simpy.rt
>>> env = simpy.rt.RealtimeEnvironment(factor=0.1)
>>> proc = env.process(example(env))
>>> env.run(until=proc)
Duration of one simulation time unit: 0.10s
If the strict parameter is set to True
(the default), the step()
and
run()
methods will raise a RuntimeError
if the computation within
a simulation time step take more time than the real-time factor allows. In the
following example, a process will perform a task that takes 0.02 seconds within
a real-time environment with a time factor of 0.01 seconds:
>>> import time
>>> import simpy.rt
>>>
>>> def slow_proc(env):
... time.sleep(0.02) # Heavy computation :-)
... yield env.timeout(1)
>>>
>>> env = simpy.rt.RealtimeEnvironment(factor=0.01)
>>> proc = env.process(slow_proc(env))
>>> try:
... env.run(until=proc)
... print('Everything alright')
... except RuntimeError:
... print('Simulation is too slow')
Simulation is too slow
To suppress the error, simply set strict=False
:
>>> env = simpy.rt.RealtimeEnvironment(factor=0.01, strict=False)
>>> proc = env.process(slow_proc(env))
>>> try:
... env.run(until=proc)
... print('Everything alright')
... except RuntimeError:
... print('Simulation is too slow')
Everything alright
That’s it. Real-time simulations are that simple with SimPy!
Monitoring¶
Monitoring is a relatively complex topic with a lot of different use-cases and lots of variations.
This guide presents some of the more common and more interesting ones. It’s purpose is to give you some hints and ideas how you can implement simulation monitoring tailored to your use-cases.
So, before you start, you need to define them:
What do you want to monitor?
When do you want to monitor?
- Regularly in defined intervals?
- When something happens?
How do you want to store the collected data?
- Store it in a simple list?
- Log it to a file?
- Write it to a database?
The following sections discuss these questions and provide some example code to help you.
Monitoring your processes¶
Monitoring your own processes is relatively easy, because you control the code. From our experience, the most common thing you might want to do is monitor the value of one or more state variables every time they change or at discrete intervals and store it somewhere (in memory, in a database, or in a file, for example).
In the simples case, you just use a list and append the required value(s) every time they change:
>>> import simpy
>>>
>>> data = [] # This list will hold all collected data
>>>
>>> def test_process(env, data):
... val = 0
... for i in range(5):
... val += env.now
... data.append(val) # Collect data
... yield env.timeout(1)
>>>
>>> env = simpy.Environment()
>>> p = env.process(test_process(env, data))
>>> env.run(p)
>>> print('Collected', data) # Lets see what we got
Collected [0, 1, 3, 6, 10]
If you want to monitor multiple variables, you can append (named)tuples to your data list.
If you want to store the data in a NumPy array or a database, you can often increase performance if you buffer the data in a plain Python list and only write larger chunks (or the complete dataset) to the database.
Resource usage¶
The use-cases for resource monitoring are numerous, for example you might want to monitor:
Utilization of a resource over time and on average, that is,
- the number of processes that are using the resource at a time
- the level of a container
- the amount of items in a store
This can be monitored either in discrete time steps or every time there is a change.
Number of processes in the (put|get)queue over time (and the average). Again, this could be monitored at discrete time steps or every time there is a change.
For PreemptiveResource, you may want to measure how often preemption occurs over time.
In contrast to your processes, you don’t have direct access to the code of the built-in resource classes. But this doesn’t prevent you from monitoring them.
Monkey-patching some of a resource’s methods allows you to gather all the data you need.
Here is an example that demonstrate how you can add callbacks to a resource that get called just before or after a get / request or a put / release event:
>>> from functools import partial, wraps
>>> import simpy
>>>
>>> def patch_resource(resource, pre=None, post=None):
... """Patch *resource* so that it calls the callable *pre* before each
... put/get/request/release operation and the callable *post* after each
... operation. The only argument to these functions is the resource
... instance.
...
... """
... def get_wrapper(func):
... # Generate a wrapper for put/get/request/release
... @wraps(func)
... def wrapper(*args, **kwargs):
... # This is the actual wrapper
... # Call "pre" callback
... if pre:
... pre(resource)
...
... # Perform actual operation
... ret = func(*args, **kwargs)
...
... # Call "post" callback
... if post:
... post(resource)
...
... return ret
... return wrapper
...
... # Replace the original operations with our wrapper
... for name in ['put', 'get', 'request', 'release']:
... if hasattr(resource, name):
... setattr(resource, name, get_wrapper(getattr(resource, name)))
>>>
>>> def monitor(data, resource):
... """This is our monitoring callback."""
... item = (
... resource._env.now, # The current simulation time
... resource.count, # The number of users
... len(resource.queue), # The number of queued processes
... )
... data.append(item)
>>>
>>> def test_process(env, res):
... with res.request() as req:
... yield req
... yield env.timeout(1)
>>>
>>> env = simpy.Environment()
>>>
>>> res = simpy.Resource(env, capacity=1)
>>> data = []
>>> # Bind *data* as first argument to monitor()
>>> # see https://docs.python.org/3/library/functools.html#functools.partial
>>> monitor = partial(monitor, data)
>>> patch_resource(res, post=monitor) # Patches (only) this resource instance
>>>
>>> p = env.process(test_process(env, res))
>>> env.run(p)
>>>
>>> print(data)
[(0, 1, 0), (1, 0, 0)]
The example above is a very generic but also very flexible way to monitor all aspects of all kinds of resources.
The other extreme would be to fit the monitoring to exactly one use case.
Imagine, for example, you only want to know how many processes are waiting for
a Resource
at a time:
>>> import simpy
>>>
>>> class MonitoredResource(simpy.Resource):
... def __init__(self, *args, **kwargs):
... super().__init__(*args, **kwargs)
... self.data = []
...
... def request(self, *args, **kwargs):
... self.data.append((self._env.now, len(self.queue)))
... return super().request(*args, **kwargs)
...
... def release(self, *args, **kwargs):
... self.data.append((self._env.now, len(self.queue)))
... return super().release(*args, **kwargs)
>>>
>>> def test_process(env, res):
... with res.request() as req:
... yield req
... yield env.timeout(1)
>>>
>>> env = simpy.Environment()
>>>
>>> res = MonitoredResource(env, capacity=1)
>>> p1 = env.process(test_process(env, res))
>>> p2 = env.process(test_process(env, res))
>>> env.run()
>>>
>>> print(res.data)
[(0, 0), (0, 0), (1, 1), (2, 0)]
In contrast to the first example, we now haven’t patched a single resource
instance but the whole class. It also removed all of the first example’s
flexibility: We only monitor Resource
typed resources, we only collect data
before the actual requests are made and we only collect the time and queue
length. At the same time, you need less than half of the code.
Event tracing¶
In order to debug or visualize a simulation, you might want to trace when events are created, triggered and processed. Maybe you also want to trace which process created an event and which processes waited for an event.
The two most interesting functions for these use-cases are
Environment.step()
, where all events get processed, and
Environment.schedule()
, where all events get scheduled and inserted
into SimPy’s event queue.
Here is an example that shows how Environment.step()
can be patched in
order to trace all processed events:
>>> from functools import partial, wraps
>>> import simpy
>>>
>>> def trace(env, callback):
... """Replace the ``step()`` method of *env* with a tracing function
... that calls *callbacks* with an events time, priority, ID and its
... instance just before it is processed.
...
... """
... def get_wrapper(env_step, callback):
... """Generate the wrapper for env.step()."""
... @wraps(env_step)
... def tracing_step():
... """Call *callback* for the next event if one exist before
... calling ``env.step()``."""
... if len(env._queue):
... t, prio, eid, event = env._queue[0]
... callback(t, prio, eid, event)
... return env_step()
... return tracing_step
...
... env.step = get_wrapper(env.step, callback)
>>>
>>> def monitor(data, t, prio, eid, event):
... data.append((t, eid, type(event)))
>>>
>>> def test_process(env):
... yield env.timeout(1)
>>>
>>> data = []
>>> # Bind *data* as first argument to monitor()
>>> # see https://docs.python.org/3/library/functools.html#functools.partial
>>> monitor = partial(monitor, data)
>>>
>>> env = simpy.Environment()
>>> trace(env, monitor)
>>>
>>> p = env.process(test_process(env))
>>> env.run(until=p)
>>>
>>> for d in data:
... print(d)
(0, 0, <class 'simpy.events.Initialize'>)
(1, 1, <class 'simpy.events.Timeout'>)
(1, 2, <class 'simpy.events.Process'>)
The example above is inspired by a pull request from Steve Pothier.
Using the same concepts, you can also patch Environment.schedule()
.
This would give you central access to the information when which event is
scheduled for what time.
In addition to that, you could also patch some or all of SimPy’s event classes, e.g., their __init__() method in order to trace when and how an event is initially being created.
Time and Scheduling¶
The aim of this section is to give you a deeper understanding of how time passes in SimPy and how it schedules and processes events.
What is time?¶
Time itself is not easy to grasp. The wikipedians describe it this way:
«Time is the indefinite continued progress of existence and events that occur in apparently irreversible succession from the past through the present to the future. Time is a component quantity of various measurements used to sequence events, to compare the duration of events or the intervals between them, and to quantify rates of change of quantities in material reality or in the conscious experience. Time is often referred to as the fourth dimension, along with the three spatial dimensions.»
What’s the problem with it?¶
Often, events (in the real world) appear to happen “at the same time”, when they are in fact happening at slightly different times. Here is an obvious example: Alice and Bob have birthday on the same day. If your time scale is in days, both birthday events happen at the same time. If you increase the resolution of you clock, e.g. to minutes, you may realise that Alice was actually born at 0:42 in the morning and Bob at 11:14 and that there’s quite a difference between the time of both events.
Doing simulation on computers suffers from similar problems. Integers (and floats, too) are discrete numbers with a lot of void in between them. Thus, events that would occur after each other in the real world (e.g., at t1 = 0.1 and t2 = 0.2) might occur at the “same” time if mapped to an integer scale (e.g., at t = 0).
On the other hand, SimPy is (like most simulation frameworks) a single-threaded, deterministic library. It processes events sequentially – one after another. If two events are scheduled at the same time, the one that is scheduled first will also be the processed first (FIFO).
That is very important for you to understand. The processes in your
modeled/simulated world may run “in parallel”, but when the simulation runs on
your CPU, all events are processed sequentially and deterministically. If you
run your simulation multiple times (and if you don’t use random
;-)),
you will always get the same results.
So keep this in mind:
- In the real world, there’s usually no at the same time.
- Discretization of the time scale can make events appear to be at the same time.
- SimPy processes events one after another, even if they have the same time.
SimPy Events and time¶
Before we continue, let’s recap the states an event can be in (see Events for details):
- untriggered: not known to the event queue
- triggered: scheduled at a time t and inserted into the event queue
- processed: removed from the event queue
SimPy’s event queue is implemented as a heap queue: “Heaps are binary trees for which every parent node has a value less than or equal to any of its children.” So if we insert events as tuples (t, event) (with t being the scheduled time) into it, the first element in the queue will by definition always be the one with the smallest t and the next one to be processed.
However, storing (t, event) tuples will not work if two events are scheduled at the same time because events are not comparable. To fix this, we also store a strictly increasing event ID with them: (t, eid, event). That way, if two events get scheduled for the same time, the one scheduled first will always be processed first.
Porting from SimPy 2 to 3¶
Porting from SimPy 2 to SimPy 3 is not overly complicated. A lot of changes merely comprise copy/paste.
This guide describes the conceptual and API changes between both SimPy versions and shows you how to change your code for SimPy 3.
Imports¶
In SimPy 2, you had to decide at import-time whether you wanted to use a normal
simulation (SimPy.Simulation
), a real-time simulation
(SimPy.SimulationRT
) or something else. You usually had to import
Simulation
(or SimulationRT
), Process
and some of the SimPy
keywords (hold
or passivate
, for example) from that package.
In SimPy 3, you usually need to import much less classes and modules (for
example, all keywords are gone). In most use cases you will now only need to
import simpy
.
SimPy 2
from Simpy.Simulation import Simulation, Process, hold
SimPy 3
import simpy
The Simulation*
classes¶
SimPy 2 encapsulated the simulation state in a Simulation*
class (e.g.,
Simulation
, SimulationRT
or SimulationTrace
). This
class also had a simulate()
method that executed a normal simulation,
a real-time simulation or something else (depending on the particular class).
There was a global Simulation
instance that was automatically created when
you imported SimPy. You could also instantiate it on your own to uses SimPy’s
object-orient API. This led to some confusion and problems, because you had to
pass the Simulation
instance around when you were using the object-oriented
API but not if you were using the procedural API.
In SimPy 3, an Environment
replaces Simulation
and
RealtimeEnvironment
replaces SimulationRT
. You always
need to instantiate an environment. There’s no more global state.
To execute a simulation, you call the environment’s
run()
method.
SimPy 2
# Procedural API
from SimPy.Simulation import initialize, simulate
initialize()
# Start processes
simulate(until=10)
# Object-oriented API
from SimPy.Simulation import Simulation
sim = Simulation()
# Start processes
sim.simulate(until=10)
SimPy3
import simpy
env = simpy.Environment()
# Start processes
env.run(until=10)
Defining a Process¶
Processes had to inherit the Process
base class in SimPy 2. Subclasses had
to implement at least a so called Process Execution Method (PEM) (which is
basically a generator function) and in most cases __init__()
. Each process
needed to know the Simulation
instance it belonged to. This reference was
passed implicitly in the procedural API and had to be passed explicitly in the
object-oriented API. Apart from some internal problems, this made it quite
cumbersome to define a simple process.
Processes were started by passing the Process
and a generator instance
created by the generator function to either the global activate()
function
or the corresponding Simulation
method.
A process in SimPy 3 is a Python generator (no matter if it’s defined on module
level or as an instance method) wrapped in a Process
instance. The generator usually requires a reference to a
Environment
to interact with, but this is completely
optional.
Processes are can be started by creating a Process
instance and passing the generator to it. The environment provides a shortcut
for this: process()
.
SimPy 2
# Procedural API
from Simpy.Simulation import Process
class MyProcess(Process):
def __init__(self, another_param):
super().__init__()
self.another_param = another_param
def generator_function(self):
"""Implement the process' behavior."""
yield something
initialize()
proc = Process('Spam')
activate(proc, proc.generator_function())
# Object-oriented API
from SimPy.Simulation import Simulation, Process
class MyProcess(Process):
def __init__(self, sim, another_param):
super().__init__(sim=sim)
self.another_param = another_param
def generator_function(self):
"""Implement the process' behaviour."""
yield something
sim = Simulation()
proc = Process(sim, 'Spam')
sim.activate(proc, proc.generator_function())
SimPy 3
import simpy
def generator_function(env, another_param):
"""Implement the process' behavior."""
yield something
env = simpy.Environment()
proc = env.process(generator_function(env, 'Spam'))
SimPy Keywords (hold
etc.)¶
In SimPy 2, processes created new events by yielding a SimPy Keyword and some
additional parameters (at least self
). These keywords had to be imported
from SimPy.Simulation*
if they were used. Internally, the keywords were
mapped to a function that generated the according event.
In SimPy 3, you directly yield events
if you want to wait for an
event to occur. You can instantiate an event directly or use the shortcuts
provided by Environment
.
Generally, whenever a process yields an event, the execution of the process is
suspended and resumed once the event has been triggered. To motivate this
understanding, some of the events were renamed. For example, the hold
keyword meant to wait until some time has passed. In terms of events this means
that a timeout has happened. Therefore hold
has been replaced by a
Timeout
event.
SimPy 2
yield hold, self, duration
yield passivate, self
yield request, self, resource
yield release, self, resource
yield waitevent, self, event
yield waitevent, self, [event_a, event_b, event_c]
yield queueevent, self, event_list
yield get, self, level, amount
yield put, self, level, amount
SimPy 3
yield env.timeout(duration) # hold: renamed
yield env.event() # passivate: renamed
yield resource.request() # Request is now bound to class Resource
resource.release() # Release no longer needs to be yielded
yield event # waitevent: just yield the event
yield env.all_of([event_a, event_b, event_c]) # waitvent
yield env.any_of([event_a, event_b, event_c]) # queuevent
yield container.get(amount) # Level is now called Container
yield container.put(amount)
yield event_a | event_b # Wait for either a or b. This is new.
yield event_a & event_b # Wait for a and b. This is new.
yield env.process(calculation(env)) # Wait for the process calculation to
# to finish.
Partially supported features¶
The following waituntil
keyword is not completely supported anymore:
yield waituntil, self, cond_func
SimPy 2 was evaluating cond_func
after every event, which was
computationally very expensive. One possible workaround is for example the
following process, which evaluates cond_func
periodically:
def waituntil(env, cond_func, delay=1):
while not cond_func():
yield env.timeout(delay)
# Usage:
yield waituntil(env, cond_func)
Interrupts¶
In SimPy 2, interrupt()
was a method of the interrupting process. The
victim of the interrupt had to be passed as an argument.
The victim was not directly notified of the interrupt but had to check if the
interrupted
flag was set. Afterwards, it had to reset the interrupt via
interruptReset()
. You could manually set the interruptCause
attribute
of the victim.
Explicitly checking for an interrupt is obviously error prone as it is too easy to be forgotten.
In SimPy 3, you call interrupt()
on the victim
process. You can optionally supply a cause. An
Interrupt
is then thrown into the victim process,
which has to handle the interrupt via try: ... except Interrupt: ...
.
SimPy 2
class Interrupter(Process):
def __init__(self, victim):
super().__init__()
self.victim = victim
def run(self):
yield hold, self, 1
self.interrupt(self.victim_proc)
self.victim_proc.interruptCause = 'Spam'
class Victim(Process):
def run(self):
yield hold, self, 10
if self.interrupted:
cause = self.interruptCause
self.interruptReset()
SimPy 3
def interrupter(env, victim_proc):
yield env.timeout(1)
victim_proc.interrupt('Spam')
def victim(env):
try:
yield env.timeout(10)
except Interrupt as interrupt:
cause = interrupt.cause
Conclusion¶
This guide is by no means complete. If you run into problems, please have a look at the other guides, the examples or the API Reference. You are also very welcome to submit improvements. Just create a pull request at bitbucket.
Examples¶
All theory is grey. In this section, we present various practical examples that demonstrate how to uses SimPy’s features.
Here is a list of examples grouped by the features they demonstrate.
Condition events¶
Interrupts¶
Monitoring¶
Resources: Container¶
Resources: Preemptive Resource¶
Resources: Resource¶
Resources: Store¶
Waiting for other processes¶
All examples¶
Bank Renege¶
Covers:
- Resources: Resource
- Condition events
A counter with a random service time and customers who renege. Based on the program bank08.py from TheBank tutorial of SimPy 2. (KGM)
This example models a bank counter and customers arriving t random times. Each customer has a certain patience. It waits to get to the counter until she’s at the end of her tether. If she gets to the counter, she uses it for a while before releasing it.
New customers are created by the source
process every few time steps.
"""
Bank renege example
Covers:
- Resources: Resource
- Condition events
Scenario:
A counter with a random service time and customers who renege. Based on the
program bank08.py from TheBank tutorial of SimPy 2. (KGM)
"""
import random
import simpy
RANDOM_SEED = 42
NEW_CUSTOMERS = 5 # Total number of customers
INTERVAL_CUSTOMERS = 10.0 # Generate new customers roughly every x seconds
MIN_PATIENCE = 1 # Min. customer patience
MAX_PATIENCE = 3 # Max. customer patience
def source(env, number, interval, counter):
"""Source generates customers randomly"""
for i in range(number):
c = customer(env, 'Customer%02d' % i, counter, time_in_bank=12.0)
env.process(c)
t = random.expovariate(1.0 / interval)
yield env.timeout(t)
def customer(env, name, counter, time_in_bank):
"""Customer arrives, is served and leaves."""
arrive = env.now
print('%7.4f %s: Here I am' % (arrive, name))
with counter.request() as req:
patience = random.uniform(MIN_PATIENCE, MAX_PATIENCE)
# Wait for the counter or abort at the end of our tether
results = yield req | env.timeout(patience)
wait = env.now - arrive
if req in results:
# We got to the counter
print('%7.4f %s: Waited %6.3f' % (env.now, name, wait))
tib = random.expovariate(1.0 / time_in_bank)
yield env.timeout(tib)
print('%7.4f %s: Finished' % (env.now, name))
else:
# We reneged
print('%7.4f %s: RENEGED after %6.3f' % (env.now, name, wait))
# Setup and start the simulation
print('Bank renege')
random.seed(RANDOM_SEED)
env = simpy.Environment()
# Start processes and run
counter = simpy.Resource(env, capacity=1)
env.process(source(env, NEW_CUSTOMERS, INTERVAL_CUSTOMERS, counter))
env.run()
The simulation’s output:
Bank renege
0.0000 Customer00: Here I am
0.0000 Customer00: Waited 0.000
3.8595 Customer00: Finished
10.2006 Customer01: Here I am
10.2006 Customer01: Waited 0.000
12.7265 Customer02: Here I am
13.9003 Customer02: RENEGED after 1.174
23.7507 Customer01: Finished
34.9993 Customer03: Here I am
34.9993 Customer03: Waited 0.000
37.9599 Customer03: Finished
40.4798 Customer04: Here I am
40.4798 Customer04: Waited 0.000
43.1401 Customer04: Finished
Carwash¶
Covers:
- Waiting for other processes
- Resources: Resource
The Carwash example is a simulation of a carwash with a limited number of machines and a number of cars that arrive at the carwash to get cleaned.
The carwash uses a Resource
to model the
limited number of washing machines. It also defines a process for washing
a car.
When a car arrives at the carwash, it requests a machine. Once it got one, it starts the carwash’s wash processes and waits for it to finish. It finally releases the machine and leaves.
The cars are generated by a setup process. After creating an intial amount of cars it creates new car processes after a random time interval as long as the simulation continues.
"""
Carwash example.
Covers:
- Waiting for other processes
- Resources: Resource
Scenario:
A carwash has a limited number of washing machines and defines
a washing processes that takes some (random) time.
Car processes arrive at the carwash at a random time. If one washing
machine is available, they start the washing process and wait for it
to finish. If not, they wait until they an use one.
"""
import random
import simpy
RANDOM_SEED = 42
NUM_MACHINES = 2 # Number of machines in the carwash
WASHTIME = 5 # Minutes it takes to clean a car
T_INTER = 7 # Create a car every ~7 minutes
SIM_TIME = 20 # Simulation time in minutes
class Carwash(object):
"""A carwash has a limited number of machines (``NUM_MACHINES``) to
clean cars in parallel.
Cars have to request one of the machines. When they got one, they
can start the washing processes and wait for it to finish (which
takes ``washtime`` minutes).
"""
def __init__(self, env, num_machines, washtime):
self.env = env
self.machine = simpy.Resource(env, num_machines)
self.washtime = washtime
def wash(self, car):
"""The washing processes. It takes a ``car`` processes and tries
to clean it."""
yield self.env.timeout(WASHTIME)
print("Carwash removed %d%% of %s's dirt." %
(random.randint(50, 99), car))
def car(env, name, cw):
"""The car process (each car has a ``name``) arrives at the carwash
(``cw``) and requests a cleaning machine.
It then starts the washing process, waits for it to finish and
leaves to never come back ...
"""
print('%s arrives at the carwash at %.2f.' % (name, env.now))
with cw.machine.request() as request:
yield request
print('%s enters the carwash at %.2f.' % (name, env.now))
yield env.process(cw.wash(name))
print('%s leaves the carwash at %.2f.' % (name, env.now))
def setup(env, num_machines, washtime, t_inter):
"""Create a carwash, a number of initial cars and keep creating cars
approx. every ``t_inter`` minutes."""
# Create the carwash
carwash = Carwash(env, num_machines, washtime)
# Create 4 initial cars
for i in range(4):
env.process(car(env, 'Car %d' % i, carwash))
# Create more cars while the simulation is running
while True:
yield env.timeout(random.randint(t_inter - 2, t_inter + 2))
i += 1
env.process(car(env, 'Car %d' % i, carwash))
# Setup and start the simulation
print('Carwash')
print('Check out http://youtu.be/fXXmeP9TvBg while simulating ... ;-)')
random.seed(RANDOM_SEED) # This helps reproducing the results
# Create an environment and start the setup process
env = simpy.Environment()
env.process(setup(env, NUM_MACHINES, WASHTIME, T_INTER))
# Execute!
env.run(until=SIM_TIME)
The simulation’s output:
Carwash
Check out http://youtu.be/fXXmeP9TvBg while simulating ... ;-)
Car 0 arrives at the carwash at 0.00.
Car 1 arrives at the carwash at 0.00.
Car 2 arrives at the carwash at 0.00.
Car 3 arrives at the carwash at 0.00.
Car 0 enters the carwash at 0.00.
Car 1 enters the carwash at 0.00.
Car 4 arrives at the carwash at 5.00.
Carwash removed 97% of Car 0's dirt.
Carwash removed 67% of Car 1's dirt.
Car 0 leaves the carwash at 5.00.
Car 1 leaves the carwash at 5.00.
Car 2 enters the carwash at 5.00.
Car 3 enters the carwash at 5.00.
Car 5 arrives at the carwash at 10.00.
Carwash removed 64% of Car 2's dirt.
Carwash removed 58% of Car 3's dirt.
Car 2 leaves the carwash at 10.00.
Car 3 leaves the carwash at 10.00.
Car 4 enters the carwash at 10.00.
Car 5 enters the carwash at 10.00.
Carwash removed 97% of Car 4's dirt.
Carwash removed 56% of Car 5's dirt.
Car 4 leaves the carwash at 15.00.
Car 5 leaves the carwash at 15.00.
Car 6 arrives at the carwash at 16.00.
Car 6 enters the carwash at 16.00.
Machine Shop¶
Covers:
- Interrupts
- Resources: PreemptiveResource
This example comprises a workshop with n identical machines. A stream of jobs (enough to keep the machines busy) arrives. Each machine breaks down periodically. Repairs are carried out by one repairman. The repairman has other, less important tasks to perform, too. Broken machines preempt theses tasks. The repairman continues them when he is done with the machine repair. The workshop works continuously.
A machine has two processes: working implements the actual behaviour of the machine (producing parts). break_machine periodically interrupts the working process to simulate the machine failure.
The repairman’s other job is also a process (implemented by other_job). The
repairman itself is a PreemptiveResource
with a capacity of 1. The machine repairing has a priority of 1, while the
other job has a priority of 2 (the smaller the number, the higher the
priority).
"""
Machine shop example
Covers:
- Interrupts
- Resources: PreemptiveResource
Scenario:
A workshop has *n* identical machines. A stream of jobs (enough to
keep the machines busy) arrives. Each machine breaks down
periodically. Repairs are carried out by one repairman. The repairman
has other, less important tasks to perform, too. Broken machines
preempt theses tasks. The repairman continues them when he is done
with the machine repair. The workshop works continuously.
"""
import random
import simpy
RANDOM_SEED = 42
PT_MEAN = 10.0 # Avg. processing time in minutes
PT_SIGMA = 2.0 # Sigma of processing time
MTTF = 300.0 # Mean time to failure in minutes
BREAK_MEAN = 1 / MTTF # Param. for expovariate distribution
REPAIR_TIME = 30.0 # Time it takes to repair a machine in minutes
JOB_DURATION = 30.0 # Duration of other jobs in minutes
NUM_MACHINES = 10 # Number of machines in the machine shop
WEEKS = 4 # Simulation time in weeks
SIM_TIME = WEEKS * 7 * 24 * 60 # Simulation time in minutes
def time_per_part():
"""Return actual processing time for a concrete part."""
return random.normalvariate(PT_MEAN, PT_SIGMA)
def time_to_failure():
"""Return time until next failure for a machine."""
return random.expovariate(BREAK_MEAN)
class Machine(object):
"""A machine produces parts and my get broken every now and then.
If it breaks, it requests a *repairman* and continues the production
after the it is repaired.
A machine has a *name* and a numberof *parts_made* thus far.
"""
def __init__(self, env, name, repairman):
self.env = env
self.name = name
self.parts_made = 0
self.broken = False
# Start "working" and "break_machine" processes for this machine.
self.process = env.process(self.working(repairman))
env.process(self.break_machine())
def working(self, repairman):
"""Produce parts as long as the simulation runs.
While making a part, the machine may break multiple times.
Request a repairman when this happens.
"""
while True:
# Start making a new part
done_in = time_per_part()
while done_in:
try:
# Working on the part
start = self.env.now
yield self.env.timeout(done_in)
done_in = 0 # Set to 0 to exit while loop.
except simpy.Interrupt:
self.broken = True
done_in -= self.env.now - start # How much time left?
# Request a repairman. This will preempt its "other_job".
with repairman.request(priority=1) as req:
yield req
yield self.env.timeout(REPAIR_TIME)
self.broken = False
# Part is done.
self.parts_made += 1
def break_machine(self):
"""Break the machine every now and then."""
while True:
yield self.env.timeout(time_to_failure())
if not self.broken:
# Only break the machine if it is currently working.
self.process.interrupt()
def other_jobs(env, repairman):
"""The repairman's other (unimportant) job."""
while True:
# Start a new job
done_in = JOB_DURATION
while done_in:
# Retry the job until it is done.
# It's priority is lower than that of machine repairs.
with repairman.request(priority=2) as req:
yield req
try:
start = env.now
yield env.timeout(done_in)
done_in = 0
except simpy.Interrupt:
done_in -= env.now - start
# Setup and start the simulation
print('Machine shop')
random.seed(RANDOM_SEED) # This helps reproducing the results
# Create an environment and start the setup process
env = simpy.Environment()
repairman = simpy.PreemptiveResource(env, capacity=1)
machines = [Machine(env, 'Machine %d' % i, repairman)
for i in range(NUM_MACHINES)]
env.process(other_jobs(env, repairman))
# Execute!
env.run(until=SIM_TIME)
# Analyis/results
print('Machine shop results after %s weeks' % WEEKS)
for machine in machines:
print('%s made %d parts.' % (machine.name, machine.parts_made))
The simulation’s output:
Machine shop
Machine shop results after 4 weeks
Machine 0 made 3251 parts.
Machine 1 made 3273 parts.
Machine 2 made 3242 parts.
Machine 3 made 3343 parts.
Machine 4 made 3387 parts.
Machine 5 made 3244 parts.
Machine 6 made 3269 parts.
Machine 7 made 3185 parts.
Machine 8 made 3302 parts.
Machine 9 made 3279 parts.
Movie Renege¶
Covers:
- Resources: Resource
- Condition events
- Shared events
This examples models a movie theater with one ticket counter selling tickets for three movies (next show only). People arrive at random times and triy to buy a random number (1–6) tickets for a random movie. When a movie is sold out, all people waiting to buy a ticket for that movie renege (leave the queue).
The movie theater is just a container for all the related data (movies, the
counter, tickets left, collected data, …). The counter is
a Resource
with a capacity of one.
The moviegoer process starts waiting until either it’s his turn (it acquires the counter resource) or until the sold out signal is triggered. If the latter is the case it reneges (leaves the queue). If it gets to the counter, it tries to buy some tickets. This might not be successful, e.g. if the process tries to buy 5 tickets but only 3 are left. If less then two tickets are left after the ticket purchase, the sold out signal is triggered.
Moviegoers are generated by the customer arrivals process. It also chooses a movie and the number of tickets for the moviegoer.
"""
Movie renege example
Covers:
- Resources: Resource
- Condition events
- Shared events
Scenario:
A movie theatre has one ticket counter selling tickets for three
movies (next show only). When a movie is sold out, all people waiting
to buy tickets for that movie renege (leave queue).
"""
import collections
import random
import simpy
RANDOM_SEED = 42
TICKETS = 50 # Number of tickets per movie
SIM_TIME = 120 # Simulate until
def moviegoer(env, movie, num_tickets, theater):
"""A moviegoer tries to by a number of tickets (*num_tickets*) for
a certain *movie* in a *theater*.
If the movie becomes sold out, she leaves the theater. If she gets
to the counter, she tries to buy a number of tickets. If not enough
tickets are left, she argues with the teller and leaves.
If at most one ticket is left after the moviegoer bought her
tickets, the *sold out* event for this movie is triggered causing
all remaining moviegoers to leave.
"""
with theater.counter.request() as my_turn:
# Wait until its our turn or until the movie is sold out
result = yield my_turn | theater.sold_out[movie]
# Check if it's our turn of if movie is sold out
if my_turn not in result:
theater.num_renegers[movie] += 1
env.exit()
# Check if enough tickets left.
if theater.available[movie] < num_tickets:
# Moviegoer leaves after some discussion
yield env.timeout(0.5)
env.exit()
# Buy tickets
theater.available[movie] -= num_tickets
if theater.available[movie] < 2:
# Trigger the "sold out" event for the movie
theater.sold_out[movie].succeed()
theater.when_sold_out[movie] = env.now
theater.available[movie] = 0
yield env.timeout(1)
def customer_arrivals(env, theater):
"""Create new *moviegoers* until the sim time reaches 120."""
while True:
yield env.timeout(random.expovariate(1 / 0.5))
movie = random.choice(theater.movies)
num_tickets = random.randint(1, 6)
if theater.available[movie]:
env.process(moviegoer(env, movie, num_tickets, theater))
Theater = collections.namedtuple('Theater', 'counter, movies, available, '
'sold_out, when_sold_out, '
'num_renegers')
# Setup and start the simulation
print('Movie renege')
random.seed(RANDOM_SEED)
env = simpy.Environment()
# Create movie theater
counter = simpy.Resource(env, capacity=1)
movies = ['Python Unchained', 'Kill Process', 'Pulp Implementation']
available = {movie: TICKETS for movie in movies}
sold_out = {movie: env.event() for movie in movies}
when_sold_out = {movie: None for movie in movies}
num_renegers = {movie: 0 for movie in movies}
theater = Theater(counter, movies, available, sold_out, when_sold_out,
num_renegers)
# Start process and run
env.process(customer_arrivals(env, theater))
env.run(until=SIM_TIME)
# Analysis/results
for movie in movies:
if theater.sold_out[movie]:
print('Movie "%s" sold out %.1f minutes after ticket counter '
'opening.' % (movie, theater.when_sold_out[movie]))
print(' Number of people leaving queue when film sold out: %s' %
theater.num_renegers[movie])
The simulation’s output:
Movie renege
Movie "Python Unchained" sold out 38.0 minutes after ticket counter opening.
Number of people leaving queue when film sold out: 16
Movie "Kill Process" sold out 43.0 minutes after ticket counter opening.
Number of people leaving queue when film sold out: 5
Movie "Pulp Implementation" sold out 28.0 minutes after ticket counter opening.
Number of people leaving queue when film sold out: 5
Gas Station Refueling¶
Covers:
- Resources: Resource
- Resources: Container
- Waiting for other processes
This examples models a gas station and cars that arrive at the station for refueling.
The gas station has a limited number of fuel pumps and a fuel tank that is
shared between the fuel pumps. The gas station is thus modeled as
Resource
. The shared fuel tank is modeled
with a Container
.
Vehicles arriving at the gas station first request a fuel pump from the station. Once they acquire one, they try to take the desired amount of fuel from the fuel pump. They leave when they are done.
The gas stations fuel level is regularly monitored by gas station control. When the level drops below a certain threshold, a tank truck is called to refuel the gas station itself.
"""
Gas Station Refueling example
Covers:
- Resources: Resource
- Resources: Container
- Waiting for other processes
Scenario:
A gas station has a limited number of gas pumps that share a common
fuel reservoir. Cars randomly arrive at the gas station, request one
of the fuel pumps and start refueling from that reservoir.
A gas station control process observes the gas station's fuel level
and calls a tank truck for refueling if the station's level drops
below a threshold.
"""
import itertools
import random
import simpy
RANDOM_SEED = 42
GAS_STATION_SIZE = 200 # liters
THRESHOLD = 10 # Threshold for calling the tank truck (in %)
FUEL_TANK_SIZE = 50 # liters
FUEL_TANK_LEVEL = [5, 25] # Min/max levels of fuel tanks (in liters)
REFUELING_SPEED = 2 # liters / second
TANK_TRUCK_TIME = 300 # Seconds it takes the tank truck to arrive
T_INTER = [30, 300] # Create a car every [min, max] seconds
SIM_TIME = 1000 # Simulation time in seconds
def car(name, env, gas_station, fuel_pump):
"""A car arrives at the gas station for refueling.
It requests one of the gas station's fuel pumps and tries to get the
desired amount of gas from it. If the stations reservoir is
depleted, the car has to wait for the tank truck to arrive.
"""
fuel_tank_level = random.randint(*FUEL_TANK_LEVEL)
print('%s arriving at gas station at %.1f' % (name, env.now))
with gas_station.request() as req:
start = env.now
# Request one of the gas pumps
yield req
# Get the required amount of fuel
liters_required = FUEL_TANK_SIZE - fuel_tank_level
yield fuel_pump.get(liters_required)
# The "actual" refueling process takes some time
yield env.timeout(liters_required / REFUELING_SPEED)
print('%s finished refueling in %.1f seconds.' % (name,
env.now - start))
def gas_station_control(env, fuel_pump):
"""Periodically check the level of the *fuel_pump* and call the tank
truck if the level falls below a threshold."""
while True:
if fuel_pump.level / fuel_pump.capacity * 100 < THRESHOLD:
# We need to call the tank truck now!
print('Calling tank truck at %d' % env.now)
# Wait for the tank truck to arrive and refuel the station
yield env.process(tank_truck(env, fuel_pump))
yield env.timeout(10) # Check every 10 seconds
def tank_truck(env, fuel_pump):
"""Arrives at the gas station after a certain delay and refuels it."""
yield env.timeout(TANK_TRUCK_TIME)
print('Tank truck arriving at time %d' % env.now)
ammount = fuel_pump.capacity - fuel_pump.level
print('Tank truck refuelling %.1f liters.' % ammount)
yield fuel_pump.put(ammount)
def car_generator(env, gas_station, fuel_pump):
"""Generate new cars that arrive at the gas station."""
for i in itertools.count():
yield env.timeout(random.randint(*T_INTER))
env.process(car('Car %d' % i, env, gas_station, fuel_pump))
# Setup and start the simulation
print('Gas Station refuelling')
random.seed(RANDOM_SEED)
# Create environment and start processes
env = simpy.Environment()
gas_station = simpy.Resource(env, 2)
fuel_pump = simpy.Container(env, GAS_STATION_SIZE, init=GAS_STATION_SIZE)
env.process(gas_station_control(env, fuel_pump))
env.process(car_generator(env, gas_station, fuel_pump))
# Execute!
env.run(until=SIM_TIME)
The simulation’s output:
Gas Station refuelling
Car 0 arriving at gas station at 87.0
Car 0 finished refueling in 18.5 seconds.
Car 1 arriving at gas station at 129.0
Car 1 finished refueling in 19.0 seconds.
Car 2 arriving at gas station at 284.0
Car 2 finished refueling in 21.0 seconds.
Car 3 arriving at gas station at 385.0
Car 3 finished refueling in 13.5 seconds.
Car 4 arriving at gas station at 459.0
Calling tank truck at 460
Car 4 finished refueling in 22.0 seconds.
Car 5 arriving at gas station at 705.0
Car 6 arriving at gas station at 750.0
Tank truck arriving at time 760
Tank truck refuelling 188.0 liters.
Car 6 finished refueling in 29.0 seconds.
Car 5 finished refueling in 76.5 seconds.
Car 7 arriving at gas station at 891.0
Car 7 finished refueling in 13.0 seconds.
Process Communication¶
Covers:
- Resources: Store
This example shows how to interconnect simulation model elements together using “resources.Store” for one-to-one, and many-to-one asynchronous processes. For one-to-many a simple BroadCastPipe class is constructed from Store.
- When Useful:
When a consumer process does not always wait on a generating process and these processes run asynchronously. This example shows how to create a buffer and also tell is the consumer process was late yielding to the event from a generating process.
This is also useful when some information needs to be broadcast to many receiving processes
Finally, using pipes can simplify how processes are interconnected to each other in a simulation model.
- Example By:
- Keith Smith
"""
Process communication example
Covers:
- Resources: Store
Scenario:
This example shows how to interconnect simulation model elements
together using :class:`~simpy.resources.store.Store` for one-to-one,
and many-to-one asynchronous processes. For one-to-many a simple
BroadCastPipe class is constructed from Store.
When Useful:
When a consumer process does not always wait on a generating process
and these processes run asynchronously. This example shows how to
create a buffer and also tell is the consumer process was late
yielding to the event from a generating process.
This is also useful when some information needs to be broadcast to
many receiving processes
Finally, using pipes can simplify how processes are interconnected to
each other in a simulation model.
Example By:
Keith Smith
"""
import random
import simpy
RANDOM_SEED = 42
SIM_TIME = 100
class BroadcastPipe(object):
"""A Broadcast pipe that allows one process to send messages to many.
This construct is useful when message consumers are running at
different rates than message generators and provides an event
buffering to the consuming processes.
The parameters are used to create a new
:class:`~simpy.resources.store.Store` instance each time
:meth:`get_output_conn()` is called.
"""
def __init__(self, env, capacity=simpy.core.Infinity):
self.env = env
self.capacity = capacity
self.pipes = []
def put(self, value):
"""Broadcast a *value* to all receivers."""
if not self.pipes:
raise RuntimeError('There are no output pipes.')
events = [store.put(value) for store in self.pipes]
return self.env.all_of(events) # Condition event for all "events"
def get_output_conn(self):
"""Get a new output connection for this broadcast pipe.
The return value is a :class:`~simpy.resources.store.Store`.
"""
pipe = simpy.Store(self.env, capacity=self.capacity)
self.pipes.append(pipe)
return pipe
def message_generator(name, env, out_pipe):
"""A process which randomly generates messages."""
while True:
# wait for next transmission
yield env.timeout(random.randint(6, 10))
# messages are time stamped to later check if the consumer was
# late getting them. Note, using event.triggered to do this may
# result in failure due to FIFO nature of simulation yields.
# (i.e. if at the same env.now, message_generator puts a message
# in the pipe first and then message_consumer gets from pipe,
# the event.triggered will be True in the other order it will be
# False
msg = (env.now, '%s says hello at %d' % (name, env.now))
out_pipe.put(msg)
def message_consumer(name, env, in_pipe):
"""A process which consumes messages."""
while True:
# Get event for message pipe
msg = yield in_pipe.get()
if msg[0] < env.now:
# if message was already put into pipe, then
# message_consumer was late getting to it. Depending on what
# is being modeled this, may, or may not have some
# significance
print('LATE Getting Message: at time %d: %s received message: %s' %
(env.now, name, msg[1]))
else:
# message_consumer is synchronized with message_generator
print('at time %d: %s received message: %s.' %
(env.now, name, msg[1]))
# Process does some other work, which may result in missing messages
yield env.timeout(random.randint(4, 8))
# Setup and start the simulation
print('Process communication')
random.seed(RANDOM_SEED)
env = simpy.Environment()
# For one-to-one or many-to-one type pipes, use Store
pipe = simpy.Store(env)
env.process(message_generator('Generator A', env, pipe))
env.process(message_consumer('Consumer A', env, pipe))
print('\nOne-to-one pipe communication\n')
env.run(until=SIM_TIME)
# For one-to many use BroadcastPipe
# (Note: could also be used for one-to-one,many-to-one or many-to-many)
env = simpy.Environment()
bc_pipe = BroadcastPipe(env)
env.process(message_generator('Generator A', env, bc_pipe))
env.process(message_consumer('Consumer A', env, bc_pipe.get_output_conn()))
env.process(message_consumer('Consumer B', env, bc_pipe.get_output_conn()))
print('\nOne-to-many pipe communication\n')
env.run(until=SIM_TIME)
The simulation’s output:
Process communication
One-to-one pipe communication
at time 6: Consumer A received message: Generator A says hello at 6.
at time 12: Consumer A received message: Generator A says hello at 12.
at time 19: Consumer A received message: Generator A says hello at 19.
at time 26: Consumer A received message: Generator A says hello at 26.
at time 36: Consumer A received message: Generator A says hello at 36.
at time 46: Consumer A received message: Generator A says hello at 46.
at time 52: Consumer A received message: Generator A says hello at 52.
at time 58: Consumer A received message: Generator A says hello at 58.
LATE Getting Message: at time 66: Consumer A received message: Generator A says hello at 65
at time 75: Consumer A received message: Generator A says hello at 75.
at time 85: Consumer A received message: Generator A says hello at 85.
at time 95: Consumer A received message: Generator A says hello at 95.
One-to-many pipe communication
at time 10: Consumer A received message: Generator A says hello at 10.
at time 10: Consumer B received message: Generator A says hello at 10.
at time 18: Consumer A received message: Generator A says hello at 18.
at time 18: Consumer B received message: Generator A says hello at 18.
at time 27: Consumer A received message: Generator A says hello at 27.
at time 27: Consumer B received message: Generator A says hello at 27.
at time 34: Consumer A received message: Generator A says hello at 34.
at time 34: Consumer B received message: Generator A says hello at 34.
at time 40: Consumer A received message: Generator A says hello at 40.
LATE Getting Message: at time 41: Consumer B received message: Generator A says hello at 40
at time 46: Consumer A received message: Generator A says hello at 46.
LATE Getting Message: at time 47: Consumer B received message: Generator A says hello at 46
at time 56: Consumer A received message: Generator A says hello at 56.
at time 56: Consumer B received message: Generator A says hello at 56.
at time 65: Consumer A received message: Generator A says hello at 65.
at time 65: Consumer B received message: Generator A says hello at 65.
at time 74: Consumer A received message: Generator A says hello at 74.
at time 74: Consumer B received message: Generator A says hello at 74.
at time 82: Consumer A received message: Generator A says hello at 82.
at time 82: Consumer B received message: Generator A says hello at 82.
at time 92: Consumer A received message: Generator A says hello at 92.
at time 92: Consumer B received message: Generator A says hello at 92.
at time 98: Consumer B received message: Generator A says hello at 98.
at time 98: Consumer A received message: Generator A says hello at 98.
Event Latency¶
Covers:
- Resources: Store
This example shows how to separate the time delay of events between processes from the processes themselves.
- When Useful:
When modeling physical things such as cables, RF propagation, etc. it better encapsulation to keep this propagation mechanism outside of the sending and receiving processes.
Can also be used to interconnect processes sending messages
- Example by:
- Keith Smith
"""
Event Latency example
Covers:
- Resources: Store
Scenario:
This example shows how to separate the time delay of events between
processes from the processes themselves.
When Useful:
When modeling physical things such as cables, RF propagation, etc. it
better encapsulation to keep this propagation mechanism outside of the
sending and receiving processes.
Can also be used to interconnect processes sending messages
Example by:
Keith Smith
"""
import simpy
SIM_DURATION = 100
class Cable(object):
"""This class represents the propagation through a cable."""
def __init__(self, env, delay):
self.env = env
self.delay = delay
self.store = simpy.Store(env)
def latency(self, value):
yield self.env.timeout(self.delay)
self.store.put(value)
def put(self, value):
self.env.process(self.latency(value))
def get(self):
return self.store.get()
def sender(env, cable):
"""A process which randomly generates messages."""
while True:
# wait for next transmission
yield env.timeout(5)
cable.put('Sender sent this at %d' % env.now)
def receiver(env, cable):
"""A process which consumes messages."""
while True:
# Get event for message pipe
msg = yield cable.get()
print('Received this at %d while %s' % (env.now, msg))
# Setup and start the simulation
print('Event Latency')
env = simpy.Environment()
cable = Cable(env, 10)
env.process(sender(env, cable))
env.process(receiver(env, cable))
env.run(until=SIM_DURATION)
The simulation’s output:
Event Latency
Received this at 15 while Sender sent this at 5
Received this at 20 while Sender sent this at 10
Received this at 25 while Sender sent this at 15
Received this at 30 while Sender sent this at 20
Received this at 35 while Sender sent this at 25
Received this at 40 while Sender sent this at 30
Received this at 45 while Sender sent this at 35
Received this at 50 while Sender sent this at 40
Received this at 55 while Sender sent this at 45
Received this at 60 while Sender sent this at 50
Received this at 65 while Sender sent this at 55
Received this at 70 while Sender sent this at 60
Received this at 75 while Sender sent this at 65
Received this at 80 while Sender sent this at 70
Received this at 85 while Sender sent this at 75
Received this at 90 while Sender sent this at 80
Received this at 95 while Sender sent this at 85
You have ideas for better examples? Please send them to our mailing list or make a pull request on bitbucket.
API Reference¶
The API reference provides detailed descriptions of SimPy’s classes and functions. It should be helpful if you plan to extend SimPy with custom components.
simpy
¶
The simpy
module aggregates SimPy’s most used components into a single
namespace. This is purely for convenience. You can of course also access
everything (and more!) via their actual submodules.
The following tables list all of the available components in this module.
Environments¶
Environment ([initial_time]) |
Execution environment for an event-based simulation. |
RealtimeEnvironment ([initial_time, factor, …]) |
Execution environment for an event-based simulation which is synchronized with the real-time (also known as wall-clock time). |
Events¶
Event (env) |
An event that may happen at some point in time. |
Timeout (env, delay[, value]) |
A Event that gets triggered after a delay has passed. |
Process (env, generator) |
Process an event yielding generator. |
AllOf (env, events) |
A Condition event that is triggered if all of a list of events have been successfully triggered. |
AnyOf (env, events) |
A Condition event that is triggered if any of a list of events has been successfully triggered. |
Interrupt (cause) |
Exception thrown into a process if it is interrupted (see interrupt() ). |
Resources¶
Resource (env[, capacity]) |
Resource with capacity of usage slots that can be requested by processes. |
PriorityResource (env[, capacity]) |
A Resource supporting prioritized requests. |
PreemptiveResource (env[, capacity]) |
A PriorityResource with preemption. |
Container (env[, capacity, init]) |
Resource containing up to capacity of matter which may either be continuous (like water) or discrete (like apples). |
Store (env[, capacity]) |
Resource with capacity slots for storing arbitrary objects. |
PriorityItem |
Wrap an arbitrary item with an orderable priority. |
PriorityStore (env[, capacity]) |
Resource with capacity slots for storing objects in priority order. |
FilterStore (env[, capacity]) |
Resource with capacity slots for storing arbitrary objects supporting filtered get requests. |
Exceptions¶
SimPyException |
Base class for all SimPy specific exceptions. |
Interrupt (cause) |
Exception thrown into a process if it is interrupted (see interrupt() ). |
StopProcess (value) |
Raised to stop a SimPy process (similar to StopIteration ). |
simpy.core
— SimPy’s core components¶
Core components for event-discrete simulation environments.
-
class
simpy.core.
BaseEnvironment
¶ Base class for event processing environments.
An implementation must at least provide the means to access the current time of the environment (see
now
) and to schedule (seeschedule()
) events as well as processing them (seestep()
.The class is meant to be subclassed for different execution environments. For example, SimPy defines a
Environment
for simulations with a virtual time and and aRealtimeEnvironment
that schedules and executes events in real (e.g., wallclock) time.-
now
¶ The current time of the environment.
-
active_process
¶ The currently active process of the environment.
-
schedule
(event, priority=1, delay=0)¶ Schedule an event with a given priority and a delay.
-
step
()¶ Processes the next event.
-
run
(until=None)¶ Executes
step()
until the given criterion until is met.- If it is
None
(which is the default), this method will return when there are no further events to be processed. - If it is an
Event
, the method will continue stepping until this event has been triggered and will return its value. Raises aRuntimeError
if there are no further events to be processed and the until event was not triggered. - If it is a number, the method will continue stepping until the environment’s time reaches until.
- If it is
-
exit
(value=None)¶ Stop the current process, optionally providing a
value
.This is a convenience function provided for Python versions prior to 3.3. From Python 3.3, you can instead use
return value
in a process.
-
-
class
simpy.core.
Environment
(initial_time=0)¶ Execution environment for an event-based simulation. The passing of time is simulated by stepping from event to event.
You can provide an initial_time for the environment. By default, it starts at
0
.This class also provides aliases for common event types, for example
process
,timeout
andevent
.-
now
¶ The current simulation time.
-
active_process
¶ The currently active process of the environment.
-
event
()¶ Return a new
Event
instance. Yielding this event suspends a process until another process triggers the event.
-
exit
(value=None)¶ Stop the current process, optionally providing a
value
.This is a convenience function provided for Python versions prior to 3.3. From Python 3.3, you can instead use
return value
in a process.
-
schedule
(event, priority=1, delay=0)¶ Schedule an event with a given priority and a delay.
-
step
()¶ Process the next event.
Raise an
EmptySchedule
if no further events are available.
-
run
(until=None)¶ Executes
step()
until the given criterion until is met.- If it is
None
(which is the default), this method will return when there are no further events to be processed. - If it is an
Event
, the method will continue stepping until this event has been triggered and will return its value. Raises aRuntimeError
if there are no further events to be processed and the until event was not triggered. - If it is a number, the method will continue stepping until the environment’s time reaches until.
- If it is
-
-
class
simpy.core.
BoundClass
(cls)¶ Allows classes to behave like methods.
The
__get__()
descriptor is basically identical tofunction.__get__()
and binds the first argument of thecls
to the descriptor instance.-
static
bind_early
(instance)¶ Bind all
BoundClass
attributes of the instance’s class to the instance itself to increase performance.
-
static
-
class
simpy.core.
EmptySchedule
¶ Thrown by an
Environment
if there are no further events to be processed.
-
simpy.core.
Infinity
= inf¶ Convenience alias for infinity
simpy.exceptions
— Exception types used by SimPy¶
SimPy specific exeptions.
-
exception
simpy.exceptions.
SimPyException
¶ Base class for all SimPy specific exceptions.
-
exception
simpy.exceptions.
Interrupt
(cause)¶ Exception thrown into a process if it is interrupted (see
interrupt()
).cause
provides the reason for the interrupt, if any.If a process is interrupted concurrently, all interrupts will be thrown into the process in the same order as they occurred.
-
cause
¶ The cause of the interrupt or
None
if no cause was provided.
-
-
exception
simpy.exceptions.
StopProcess
(value)¶ Raised to stop a SimPy process (similar to
StopIteration
).In Python 2, a
return value
inside generator functions is not allowed. The fall-back was raisingStopIteration(value)
instead. However, this is deprecated now, so we need a custom exception type for this.-
value
¶ The process’ return value.
-
simpy.events
— Core event types¶
This module contains the basic event types used in SimPy.
The base class for all events is Event
. Though it can be directly
used, there are several specialized subclasses of it.
Event (env) |
An event that may happen at some point in time. |
Timeout (env, delay[, value]) |
A Event that gets triggered after a delay has passed. |
Process (env, generator) |
Process an event yielding generator. |
AnyOf (env, events) |
A Condition event that is triggered if any of a list of events has been successfully triggered. |
AllOf (env, events) |
A Condition event that is triggered if all of a list of events have been successfully triggered. |
-
simpy.events.
PENDING
= object()¶ Unique object to identify pending values of events.
-
simpy.events.
URGENT
= 0¶ Priority of interrupts and process initialization events.
-
simpy.events.
NORMAL
= 1¶ Default priority used by events.
-
class
simpy.events.
Event
(env)¶ An event that may happen at some point in time.
An event
- may happen (
triggered
isFalse
), - is going to happen (
triggered
isTrue
) or - has happened (
processed
isTrue
).
Every event is bound to an environment env and is initially not triggered. Events are scheduled for processing by the environment after they are triggered by either
succeed()
,fail()
ortrigger()
. These methods also set the ok flag and the value of the event.An event has a list of
callbacks
. A callback can be any callable. Once an event gets processed, all callbacks will be invoked with the event as the single argument. Callbacks can check if the event was successful by examining ok and do further processing with the value it has produced.Failed events are never silently ignored and will raise an exception upon being processed. If a callback handles an exception, it must set
defused
toTrue
to prevent this.This class also implements
__and__()
(&
) and__or__()
(|
). If you concatenate two events using one of these operators, aCondition
event is generated that lets you wait for both or one of them.-
env
= None¶ The
Environment
the event lives in.
-
callbacks
= None¶ List of functions that are called when the event is processed.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
- may happen (
-
class
simpy.events.
Timeout
(env, delay, value=None)¶ A
Event
that gets triggered after a delay has passed.This event is automatically triggered when it is created.
-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
-
class
simpy.events.
Initialize
(env, process)¶ Initializes a process. Only used internally by
Process
.This event is automatically triggered when it is created.
-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
-
class
simpy.events.
Interruption
(process, cause)¶ Immediately schedules an
Interrupt
exception with the given cause to be thrown into process.This event is automatically triggered when it is created.
-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
-
class
simpy.events.
Process
(env, generator)¶ Process an event yielding generator.
A generator (also known as a coroutine) can suspend its execution by yielding an event.
Process
will take care of resuming the generator with the value of that event once it has happened. The exception of failed events is thrown into the generator.Process
itself is an event, too. It is triggered, once the generator returns or raises an exception. The value of the process is the return value of the generator or the exception, respectively.Note
Python version prior to 3.3 do not support return statements in generators. You can use :meth:~simpy.core.Environment.exit() as a workaround.
Processes can be interrupted during their execution by
interrupt()
.-
target
¶ The event that the process is currently waiting for.
Returns
None
if the process is dead or it is currently being interrupted.
-
is_alive
¶ True
until the process generator exits.
-
interrupt
(cause=None)¶ Interupt this process optionally providing a cause.
A process cannot be interrupted if it already terminated. A process can also not interrupt itself. Raise a
RuntimeError
in these cases.
-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
-
class
simpy.events.
Condition
(env, evaluate, events)¶ An event that gets triggered once the condition function evaluate returns
True
on the given list of events.The value of the condition event is an instance of
ConditionValue
which allows convenient access to the input events and their values. TheConditionValue
will only contain entries for those events that occurred before the condition is processed.If one of the events fails, the condition also fails and forwards the exception of the failing event.
The evaluate function receives the list of target events and the number of processed events in this list:
evaluate(events, processed_count)
. If it returnsTrue
, the condition is triggered. TheCondition.all_events()
andCondition.any_events()
functions are used to implement and (&
) and or (|
) for events.Condition events can be nested.
-
static
all_events
(events, count)¶ An evaluation function that returns
True
if all events have been triggered.
-
static
any_events
(events, count)¶ An evaluation function that returns
True
if at least one of events has been triggered.
-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
static
-
class
simpy.events.
AllOf
(env, events)¶ A
Condition
event that is triggered if all of a list of events have been successfully triggered. Fails immediately if any of events failed.-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
-
class
simpy.events.
AnyOf
(env, events)¶ A
Condition
event that is triggered if any of a list of events has been successfully triggered. Fails immediately if any of events failed.-
defused
¶ Becomes
True
when the failed event’s exception is “defused”.When an event fails (i.e. with
fail()
), the failed event’s value is an exception that will be re-raised when theEnvironment
processes the event (i.e. instep()
).It is also possible for the failed event’s exception to be defused by setting
defused
toTrue
from an event callback. Doing so prevents the event’s exception from being re-raised when the event is processed by theEnvironment
.
-
fail
(exception)¶ Set exception as the events value, mark it as failed and schedule it for processing by the environment. Returns the event instance.
Raises
ValueError
if exception is not anException
.Raises
RuntimeError
if this event has already been triggered.
-
ok
¶ Becomes
True
when the event has been triggered successfully.A “successful” event is one triggered with
succeed()
.Raises: AttributeError – if accessed before the event is triggered.
-
processed
¶ Becomes
True
if the event has been processed (e.g., its callbacks have been invoked).
-
succeed
(value=None)¶ Set the event’s value, mark it as successful and schedule it for processing by the environment. Returns the event instance.
Raises
RuntimeError
if this event has already been triggerd.
-
trigger
(event)¶ Trigger the event with the state and value of the provided event. Return self (this event instance).
This method can be used directly as a callback function to trigger chain reactions.
-
triggered
¶ Becomes
True
if the event has been triggered and its callbacks are about to be invoked.
-
value
¶ The value of the event if it is available.
The value is available when the event has been triggered.
Raises
AttributeError
if the value is not yet available.
-
simpy.resources
— Shared resource primitives¶
SimPy implements three types of resources that can be used to synchronize processes or to model congestion points:
resource |
Shared resources supporting priorities and preemption. |
container |
Resource for sharing homogeneous matter between processes, either continuous (like water) or discrete (like apples). |
store |
Shared resources for storing a possibly unlimited amount of objects supporting requests for specific objects. |
They are derived from the base classes defined in the
base
module. These classes are also meant to support
the implementation of custom resource types.
Resources — simpy.resources.resource
¶
Shared resources supporting priorities and preemption.
These resources can be used to limit the number of processes using them concurrently. A process needs to request the usage right to a resource. Once the usage right is not needed anymore it has to be released. A gas station can be modelled as a resource with a limited amount of fuel-pumps. Vehicles arrive at the gas station and request to use a fuel-pump. If all fuel-pumps are in use, the vehicle needs to wait until one of the users has finished refueling and releases its fuel-pump.
These resources can be used by a limited number of processes at a time. Processes request these resources to become a user and have to release them once they are done. For example, a gas station with a limited number of fuel pumps can be modeled with a Resource. Arriving vehicles request a fuel-pump. Once one is available they refuel. When they are done, the release the fuel-pump and leave the gas station.
Requesting a resource is modelled as “putting a process’ token into the
resources” and releasing a resources correspondingly as “getting a process’
token out of the resource”. Thus, calling request()
/release()
is
equivalent to calling put()
/get()
. Note, that releasing a resource will
always succeed immediately, no matter if a process is actually using a resource
or not.
Besides Resource
, there is a PriorityResource
, where
processes can define a request priority, and a PreemptiveResource
whose resource users can be preempted by requests with a higher priority.
-
class
simpy.resources.resource.
Resource
(env, capacity=1)¶ Resource with capacity of usage slots that can be requested by processes.
If all slots are taken, requests are enqueued. Once a usage request is released, a pending request will be triggered.
The env parameter is the
Environment
instance the resource is bound to.-
count
¶ Number of users currently using the resource.
-
-
class
simpy.resources.resource.
PriorityResource
(env, capacity=1)¶ A
Resource
supporting prioritized requests.Pending requests in the
queue
are sorted in ascending order by their priority (that means lower values are more important).-
PutQueue
¶ Type of the put queue. See
put_queue
for details.alias of
SortedQueue
-
GetQueue
¶ alias of
__builtin__.list
-
request
¶ Request a usage slot with the given priority.
alias of
PriorityRequest
-
-
class
simpy.resources.resource.
PreemptiveResource
(env, capacity=1)¶ A
PriorityResource
with preemption.If a request is preempted, the process of that request will receive an
Interrupt
with aPreempted
instance as cause.
-
class
simpy.resources.resource.
Preempted
(by, usage_since, resource)¶ Cause of an preemption
Interrupt
containing information about the preemption.-
by
= None¶ The preempting
simpy.events.Process
.
-
usage_since
= None¶ The simulation time at which the preempted process started to use the resource.
-
resource
= None¶ The resource which was lost, i.e., caused the preemption.
-
-
class
simpy.resources.resource.
Request
(resource)¶ Request usage of the resource. The event is triggered once access is granted. Subclass of
simpy.resources.base.Put
.If the maximum capacity of users has not yet been reached, the request is triggered immediately. If the maximum capacity has been reached, the request is triggered once an earlier usage request on the resource is released.
The request is automatically released when the request was created within a
with
statement.
-
class
simpy.resources.resource.
PriorityRequest
(resource, priority=0, preempt=True)¶ Request the usage of resource with a given priority. If the resource supports preemption and preempt is
True
other usage requests of the resource may be preempted (seePreemptiveResource
for details).This event type inherits
Request
and adds some additional attributes needed byPriorityResource
andPreemptiveResource
-
priority
= None¶ The priority of this request. A smaller number means higher priority.
-
preempt
= None¶ Indicates whether the request should preempt a resource user or not (
PriorityResource
ignores this flag).
-
time
= None¶ The time at which the request was made.
-
usage_since
= None¶ The time at which the request succeeded.
-
key
= None¶ Key for sorting events. Consists of the priority (lower value is more important), the time at which the request was made (earlier requests are more important) and finally the preemption flag (preempt requests are more important).
-
-
class
simpy.resources.resource.
Release
(resource, request)¶ Releases the usage of resource granted by request. This event is triggered immediately. Subclass of
simpy.resources.base.Get
.
-
class
simpy.resources.resource.
SortedQueue
(maxlen=None)¶ Queue for sorting events by their
key
attribute.-
maxlen
= None¶ Maximum length of the queue.
-
append
(item)¶ Sort item into the queue.
Raise a
RuntimeError
if the queue is full.
-
Containers — simpy.resources.container
¶
Resource for sharing homogeneous matter between processes, either continuous (like water) or discrete (like apples).
A Container
can be used to model the fuel tank of a gasoline station.
Tankers increase and refuelled cars decrease the amount of gas in the station’s
fuel tanks.
-
class
simpy.resources.container.
Container
(env, capacity=inf, init=0)¶ Resource containing up to capacity of matter which may either be continuous (like water) or discrete (like apples). It supports requests to put or get matter into/from the container.
The env parameter is the
Environment
instance the container is bound to.The capacity defines the size of the container. By default, a container is of unlimited size. The initial amount of matter is specified by init and defaults to
0
.Raise a
ValueError
ifcapacity <= 0
,init < 0
orinit > capacity
.-
level
¶ The current amount of the matter in the container.
-
put
¶ Request to put amount of matter into the container.
alias of
ContainerPut
-
get
¶ Request to get amount of matter out of the container.
alias of
ContainerGet
-
-
class
simpy.resources.container.
ContainerPut
(container, amount)¶ Request to put amount of matter into the container. The request will be triggered once there is enough space in the container available.
Raise a
ValueError
ifamount <= 0
.-
amount
= None¶ The amount of matter to be put into the container.
-
-
class
simpy.resources.container.
ContainerGet
(container, amount)¶ Request to get amount of matter from the container. The request will be triggered once there is enough matter available in the container.
Raise a
ValueError
ifamount <= 0
.-
amount
= None¶ The amount of matter to be taken out of the container.
-
Stores — simpy.resources.store
¶
Shared resources for storing a possibly unlimited amount of objects supporting requests for specific objects.
The Store
operates in a FIFO (first-in, first-out) order. Objects are
retrieved from the store in the order they were put in. The get requests of a
FilterStore
can be customized by a filter to only retrieve objects
matching a given criterion.
-
class
simpy.resources.store.
Store
(env, capacity=inf)¶ Resource with capacity slots for storing arbitrary objects. By default, the capacity is unlimited and objects are put and retrieved from the store in a first-in first-out order.
The env parameter is the
Environment
instance the container is bound to.-
items
= None¶ List of the items available in the store.
-
-
class
simpy.resources.store.
PriorityItem
(priority, item)¶ Wrap an arbitrary item with an orderable priority.
Pairs a priority with an arbitrary item. Comparisons of PriorityItem instances only consider the priority attribute, thus supporting use of unorderable items in a
PriorityStore
instance.-
item
¶ Alias for field number 1
-
priority
¶ Alias for field number 0
-
-
class
simpy.resources.store.
PriorityStore
(env, capacity=inf)¶ Resource with capacity slots for storing objects in priority order.
Unlike
Store
which provides first-in first-out discipline,PriorityStore
maintains items in sorted order such that the smallest items value are retreived first from the store.All items in a PriorityStore instance must be orderable; which is to say that items must implement
__lt__()
. To use unorderable items with PriorityStore, usePriorityItem
.
-
class
simpy.resources.store.
FilterStore
(env, capacity=inf)¶ Resource with capacity slots for storing arbitrary objects supporting filtered get requests. Like the
Store
, the capacity is unlimited by default and objects are put and retrieved from the store in a first-in first-out order.Get requests can be customized with a filter function to only trigger for items for which said filter function returns
True
.Note
In contrast to
Store
, get requests of aFilterStore
won’t necessarily be triggered in the same order they were issued.Example: The store is empty. Process 1 tries to get an item of type a, Process 2 an item of type b. Another process puts one item of type b into the store. Though Process 2 made his request after Process 1, it will receive that new item because Process 1 doesn’t want it.
-
get
¶ Request a to get an item, for which filter returns
True
, out of the store.alias of
FilterStoreGet
-
-
class
simpy.resources.store.
StorePut
(store, item)¶ Request to put item into the store. The request is triggered once there is space for the item in the store.
-
item
= None¶ The item to put into the store.
-
-
class
simpy.resources.store.
StoreGet
(resource)¶ Request to get an item from the store. The request is triggered once there is an item available in the store.
-
class
simpy.resources.store.
FilterStoreGet
(resource, filter=<function <lambda>>)¶ Request to get an item from the store matching the filter. The request is triggered once there is such an item available in the store.
filter is a function receiving one item. It should return
True
for items matching the filter criterion. The default function returnsTrue
for all items, which makes the request to behave exactly likeStoreGet
.-
filter
= None¶ The filter function to filter items in the store.
-
Base classes — simpy.resources.base
¶
Base classes of for SimPy’s shared resource types.
BaseResource
defines the abstract base resource. It supports get and
put requests, which return Put
and Get
events respectively.
These events are triggered once the request has been completed.
-
class
simpy.resources.base.
BaseResource
(env, capacity)¶ Abstract base class for a shared resource.
You can
put()
something into the resources orget()
something out of it. Both methods return an event that is triggered once the operation is completed. If aput()
request cannot complete immediately (for example if the resource has reached a capacity limit) it is enqueued in theput_queue
for later processing. Likewise forget()
requests.Subclasses can customize the resource by:
- providing custom
PutQueue
andGetQueue
types, - providing custom
Put
respectivelyGet
events, - and implementing the request processing behaviour through the methods
_do_get()
and_do_put()
.
-
PutQueue
¶ alias of
__builtin__.list
-
GetQueue
¶ alias of
__builtin__.list
-
put_queue
= None¶ Queue of pending put requests.
-
get_queue
= None¶ Queue of pending get requests.
-
capacity
¶ Maximum capacity of the resource.
- providing custom
-
class
simpy.resources.base.
Put
(resource)¶ Generic event for requesting to put something into the resource.
This event (and all of its subclasses) can act as context manager and can be used with the
with
statement to automatically cancel the request if an exception (like ansimpy.exceptions.Interrupt
for example) occurs:with res.put(item) as request: yield request
-
class
simpy.resources.base.
Get
(resource)¶ Generic event for requesting to get something from the resource.
This event (and all of its subclasses) can act as context manager and can be used with the
with
statement to automatically cancel the request if an exception (like ansimpy.exceptions.Interrupt
for example) occurs:with res.get() as request: item = yield request
simpy.rt
— Real-time simulation¶
Execution environment for events that synchronizes passing of time with the real-time (aka wall-clock time).
-
class
simpy.rt.
RealtimeEnvironment
(initial_time=0, factor=1.0, strict=True)¶ Execution environment for an event-based simulation which is synchronized with the real-time (also known as wall-clock time). A time step will take factor seconds of real time (one second by default). A step from
0
to3
with afactor=0.5
will, for example, take at least 1.5 seconds.The
step()
method will raise aRuntimeError
if a time step took too long to compute. This behaviour can be disabled by setting strict toFalse
.-
now
¶ The current simulation time.
-
active_process
¶ The currently active process of the environment.
-
factor
¶ Scaling factor of the real-time.
-
strict
¶ Running mode of the environment.
step()
will raise aRuntimeError
if this is set toTrue
and the processing of events takes too long.
-
event
()¶ Return a new
Event
instance. Yielding this event suspends a process until another process triggers the event.
-
exit
(value=None)¶ Stop the current process, optionally providing a
value
.This is a convenience function provided for Python versions prior to 3.3. From Python 3.3, you can instead use
return value
in a process.
-
schedule
(event, priority=1, delay=0)¶ Schedule an event with a given priority and a delay.
-
step
()¶ Process the next event after enough real-time has passed for the event to happen.
The delay is scaled according to the real-time
factor
. Withstrict
mode enabled, aRuntimeError
will be raised, if the event is processed too slowly.
-
sync
()¶ Synchronize the internal time with the current wall-clock time.
This can be useful to prevent
step()
from raising an error if a lot of time passes between creating the RealtimeEnvironment and callingrun()
orstep()
.
-
run
(until=None)¶ Executes
step()
until the given criterion until is met.- If it is
None
(which is the default), this method will return when there are no further events to be processed. - If it is an
Event
, the method will continue stepping until this event has been triggered and will return its value. Raises aRuntimeError
if there are no further events to be processed and the until event was not triggered. - If it is a number, the method will continue stepping until the environment’s time reaches until.
- If it is
-
simpy.util
— Utility functions for SimPy¶
A collection of utility functions:
start_delayed (env, generator, delay) |
Return a helper process that starts another process for generator after a certain delay. |
-
simpy.util.
start_delayed
(env, generator, delay)¶ Return a helper process that starts another process for generator after a certain delay.
process()
starts a process at the current simulation time. This helper allows you to start a process after a delay of delay simulation time units:>>> from simpy import Environment >>> from simpy.util import start_delayed >>> def my_process(env, x): ... print('%s, %s' % (env.now, x)) ... yield env.timeout(1) ... >>> env = Environment() >>> proc = start_delayed(env, my_process(env, 3), 5) >>> env.run() 5, 3
Raise a
ValueError
ifdelay <= 0
.
About SimPy¶
This sections is all about the non-technical stuff. How did SimPy evolve? Who was responsible for it? And what the heck were they thinking when they made it?
SimPy History & Change Log¶
SimPy was originally based on ideas from Simula and Simscript but uses standard Python. It combines two previous packages, SiPy, in Simula-Style (Klaus Müller) and SimPy, in Simscript style (Tony Vignaux and Chang Chui).
SimPy was based on efficient implementation of co-routines using Python’s generators capability.
SimPy 3 introduced a completely new and easier-to-use API, but still relied on Python’s generators as they proved to work very well.
The package has been hosted on Sourceforge.net since September 15th, 2002. In June 2012, the project moved to Bitbucket.org.
3.0.9 – 2016-06-12¶
- [NEW]
PriorityStore
resource and performance benchmarks were implemented by Peter Grayson. - [FIX] Support for identifying nested preemptions was added by Cristian Klein.
3.0.8 – 2015-06-23¶
- [NEW] Added a monitoring guide to the documentation.
- [FIX] Improved packaging (thanks to Larissa Reis).
- [FIX] Fixed and improved various test cases.
3.0.7 - 2015-03-01¶
- [FIX] State of resources and requests were inconsistent before the request has been processed (issue #62).
- [FIX] Empty conditions were never triggered (regression in 3.0.6, issue #63).
- [FIX]
Environment.run()
will fail if the until event does not get triggered (issue #64). - [FIX] Callback modification during event processing is now prohibited (thanks to Andreas Beham).
3.0.6 - 2015-01-30¶
- [NEW] Guide to SimPy resources.
- [CHANGE] Improve performance of condition events.
- [CHANGE] Improve performance of filter store (thanks to Christoph Körner).
- [CHANGE] Exception tracebacks are now more compact.
- [FIX]
AllOf
conditions handle already processed events correctly (issue #52). - [FIX] Add
sync()
toRealtimeEnvironment
to reset its internal wall-clock reference time (issue #42). - [FIX] Only send copies of exceptions into processes to prevent traceback modifications.
- [FIX] Documentation improvements.
3.0.5 – 2014-05-14¶
- [CHANGE] Move interruption and all of the safety checks into a new event (pull request #30)
- [FIX]
FilterStore.get()
now behaves correctly (issue #49). - [FIX] Documentation improvements.
3.0.4 – 2014-04-07¶
- [NEW] Verified, that SimPy works on Python 3.4.
- [NEW] Guide to SimPy events
- [CHANGE] The result dictionary for condition events (
AllOF
/&
andAnyOf
/|
) now is an OrderedDict sorted in the same way as the original events list. - [CHANGE] Condition events now also except processed events.
- [FIX]
Resource.request()
directly afterResource.release()
no longer successful. The process now has to wait as supposed to. - [FIX]
Event.fail()
now accept all exceptions derived fromBaseException
instead of onlyException
.
3.0.3 – 2014-03-06¶
- [NEW] Guide to SimPy basics.
- [NEW] Guide to SimPy Environments.
- [FIX] Timing problems with real time simulation on Windows (issue #46).
- [FIX] Installation problems on Windows due to Unicode errors (issue #41).
- [FIX] Minor documentation issues.
3.0.2 – 2013-10-24¶
- [FIX] The default capacity for
Container
andFilterStore
is now alsoinf
.
3.0.1 – 2013-10-24¶
- [FIX] Documentation and default parameters of
Store
didn’t match. Its default capacity is nowinf
.
3.0 – 2013-10-11¶
SimPy 3 has been completely rewritten from scratch. Our main goals were to simplify the API and code base as well as making SimPy more flexible and extensible. Some of the most important changes are:
- Stronger focus on events. Processes yield event instances and are suspended until the event is triggered. An example for an event is a timeout (formerly known as hold), but even processes are now events, too (you can wait until a process terminates).
- Events can be combined with
&
(and) and|
(or) to create condition events. - Process can now be defined by any generator function. You don’t have to
subclass
Process
anymore. - No more global simulation state. Every simulation stores its state in an
environment which is comparable to the old
Simulation
class. - Improved resource system with newly added resource types.
- Removed plotting and GUI capabilities. Pyside and matplotlib are much better with this.
- Greatly improved test suite. Its cleaner, and the tests are shorter and more numerous.
- Completely overhauled documentation.
There is a guide for porting from SimPy 2 to SimPy 3. If you want to stick
to SimPy 2 for a while, change your requirements to 'SimPy>=2.3,<3'
.
All in all, SimPy has become a framework for asynchronous programming based on coroutines. It brings more than ten years of experience and scientific know-how in the field of event-discrete simulation to the world of asynchronous programming and should thus be a solid foundation for everything based on an event loop.
You can find information about older versions on the history page
2.3.1 – 2012-01-28¶
- [NEW] More improvements on the documentation.
- [FIX] Syntax error in tkconsole.py when installing on Py3.2.
- [FIX] Added mock to the dep. list in SimPy.test().
2.3 – 2011-12-24¶
- [NEW] Support for Python 3.2. Support for Python <= 2.5 has been dropped.
- [NEW] SimPy.test() method to run the tests on the installed version of SimPy.
- [NEW] Tutorials/examples were integrated into the test suite.
- [CHANGE] Even more code clean-up (e.g., removed prints throughout the code, removed if-main-blocks, …).
- [CHANGE] Many documentation improvements.
2.2 – 2011-09-27¶
- [CHANGE] Restructured package layout to be conform to the Hitchhiker’s Guide to packaging
- [CHANGE] Tests have been ported to pytest.
- [CHANGE] Documentation improvements and clean-ups.
- [FIX] Fixed incorrect behavior of Store._put, thanks to Johannes Koomer for the fix.
2.1 – 2010-06-03¶
- [NEW] A function step has been added to the API. When called, it executes the next scheduled event. (step is actually a method of Simulation.)
- [NEW] Another new function is peek. It returns the time of the next event. By using peek and step together, one can easily write e.g. an interactive program to step through a simulation event by event.
- [NEW] A simple interactive debugger
stepping.py
has been added. It allows stepping through a simulation, with options to skip to a certain time, skip to the next event of a given process, or viewing the event list. - [NEW] Versions of the Bank tutorials (documents and programs) using the advanced- [NEW] object-oriented API have been added.
- [NEW] A new document describes tools for gaining insight into and debugging SimPy models.
- [CHANGE] Major re-structuring of SimPy code, resulting in much less SimPy code – great for the maintainers.
- [CHANGE] Checks have been added which test whether entities belong to the same Simulation instance.
- [CHANGE] The Monitor and Tally methods timeAverage and timeVariance now calculate only with the observed time-series. No value is assumed for the period prior to the first observation.
- [CHANGE] Changed class Lister so that circular references between objects no longer lead to stack overflow and crash.
- [FIX] Functions allEventNotices and allEventTimes are working again.
- [FIX] Error messages for methods in SimPy.Lib work again.
2.0.1 – 2009-04-06¶
- [NEW] Tests for real time behavior (testRT_Behavior.py and testRT_Behavior_OO.py in folder SimPy).
- [FIX] Repaired a number of coding errors in several models in the SimPyModels folder.
- [FIX] Repaired SimulationRT.py bug introduced by recoding for the OO API.
- [FIX] Repaired errors in sample programs in documents:
- Simulation with SimPy - In Depth Manual
- SimPy’s Object Oriented API Manual
- Simulation With Real Time Synchronization Manual
- SimPlot Manual
- Publication-quality Plot Production With Matplotlib Manual
2.0.0 – 2009-01-26¶
This is a major release with changes to the SimPy application programming interface (API) and the formatting of the documentation.
API changes¶
In addition to its existing API, SimPy now also has an object oriented API. The additional API
- allows running SimPy in parallel on multiple processors or multi-core CPUs,
- supports better structuring of SimPy programs,
- allows subclassing of class Simulation and thus provides users with the capability of creating new simulation modes/libraries like SimulationTrace, and
- reduces the total amount of SimPy code, thereby making it easier to maintain.
Note that the OO API is in addition to the old API. SimPy 2.0 is fully backward compatible.
Documentation format changes¶
SimPy’s documentation has been restructured and processed by the Sphinx documentation generation tool. This has generated one coherent, well structured document which can be easily browsed. A seach capability is included.
March 2008: Version 1.9.1¶
This is a bug-fix release which cures the following bugs:
- Excessive production of circular garbage, due to a circular reference between Process instances and event notices. This led to large memory requirements.
- Runtime error for preempts of proceeses holding multiple Resource objects.
It also adds a Short Manual, describing only the basic facilities of SimPy.
December 2007: Version 1.9¶
This is a major release with added functionality/new user API calls and bug fixes.
Major changes¶
- The event list handling has been changed to improve the runtime performance of large SimPy models (models with thousands of processes). The use of dictionaries for timestamps has been stopped. Thanks are due to Prof. Norm Matloff and a team of his students who did a study on improving SimPy performance. This was one of their recommendations. Thanks, Norm and guys! Furthermore, in version 1.9 the ‘heapq’ sorting package replaces ‘bisect’. Finally, cancelling events no longer removes them, but rather marks them. When their event time comes, they are ignored. This was Tony Vignaux’ idea!
- The Manual has been edited and given an easier-to-read layout.
- The Bank2 tutorial has been extended by models which use more advanced SimPy commands/constructs.
Bug fixes¶
- The tracing of ‘activate’ statements has been enabled.
Additions¶
- A method returning the time-weighted variance of observations has been added to classes Monitor and Tally.
- A shortcut activation method called “start” has been added to class Process.
January 2007: Version 1.8¶
Major Changes¶
- SimPy 1.8 and future releases will not run under the obsolete Python 2.2 version. They require Python 2.3 or later.
- The Manual has been thoroughly edited, restructured and rewritten. It is now also provided in PDF format.
- The Cheatsheet has been totally rewritten in a tabular format. It is provided in both XLS (MS Excel spreadsheet) and PDF format.
- The version of SimPy.Simulation(RT/Trace/Step) is now accessible by the variable ‘version’.
- The __str__ method of Histogram was changed to return a table format.
Bug fixes¶
- Repaired a bug in yield waituntil runtime code.
- Introduced check for capacity parameter of a Level or a Store being a number > 0.
- Added code so that self.eventsFired gets set correctly after an event fires in a compound yield get/put with a waitevent clause (reneging case).
- Repaired a bug in prettyprinting of Store objects.
Additions¶
- New compound yield statements support time-out or event-based reneging in get and put operations on Store and Level instances.
- yield get on a Store instance can now have a filter function.
- All Monitor and Tally instances are automatically registered in list allMonitors and allTallies, respectively.
- The new function startCollection allows activation of Monitors and Tallies at a specified time.
- A printHistogram method was added to Tally and Monitor which generates a table-form histogram.
- In SimPy.SimulationRT: A function for allowing changing the ratio wall clock time to simulation time has been added.
June 2006: Version 1.7.1¶
This is a maintenance release. The API has not been changed/added to.
- Repair of a bug in the _get methods of Store and Level which could lead to synchronization problems (blocking of producer processes, despite space being available in the buffer).
- Repair of Level __init__ method to allow initialBuffered to be of either float or int type.
- Addition of type test for Level get parameter ‘nrToGet’ to limit it to positive int or float.
- To improve pretty-printed output of ‘Level’ objects, changed attribute ‘_nrBuffered’ to ‘nrBuffered’ (synonym for ‘amount’ property).
- To improve pretty-printed output of ‘Store’ objects, added attribute ‘buffered’ (which refers to ‘_theBuffer’ attribute).
February 2006: Version 1.7¶
This is a major release.
- Addition of an abstract class Buffer, with two sub-classes Store and Level Buffers are used for modelling inter-process synchronization in producer/ consumer and multi-process cooperation scenarios.
- Addition of two new yield statements:
- yield put for putting items into a buffer, and
- yield get for getting items from a buffer.
- The Manual has undergone a major re-write/edit.
- All scripts have been restructured for compatibility with IronPython 1 beta2. This was doen by moving all import statements to the beginning of the scripts. After the removal of the first (shebang) line, all scripts (with the exception of plotting and GUI scripts) can run successfully under this new Python implementation.
September 2005: Version 1.6.1¶
This is a minor release.
- Addition of Tally data collection class as alternative to Monitor. It is intended for collecting very large data sets more efficiently in storage space and time than Monitor.
- Change of Resource to work with Tally (new Resource API is backwards-compatible with 1.6).
- Addition of function setHistogram to class Monitor for initializing histograms.
- New function allEventNotices() for debugging/teaching purposes. It returns a prettyprinted string with event times and names of process instances.
- Addition of function allEventTimes (returns event times of all scheduled events).
15 June 2005: Version 1.6¶
- Addition of two compound yield statement forms to support the modelling of processes reneging from resource queues.
- Addition of two test/demo files showing the use of the new reneging statements.
- Addition of test for prior simulation initialization in method activate().
- Repair of bug in monitoring thw waitQ of a resource when preemption occurs.
- Major restructuring/editing to Manual and Cheatsheet.
1 February 2005: Version 1.5.1¶
MAJOR LICENSE CHANGE:
Starting with this version 1.5.1, SimPy is being release under the GNU Lesser General Public License (LGPL), instead of the GNU GPL. This change has been made to encourage commercial firms to use SimPy in for-profit work.
Minor re-release
No additional/changed functionality
Includes unit test file’MonitorTest.py’ which had been accidentally deleted from 1.5
Provides updated version of ‘Bank.html’ tutorial.
Provides an additional tutorial (‘Bank2.html’) which shows how to use the new synchronization constructs introduced in SimPy 1.5.
More logical, cleaner version numbering in files.
1 December 2004: Version 1.5¶
- No new functionality/API changes relative to 1.5 alpha
- Repaired bug related to waiting/queuing for multiple events
- SimulationRT: Improved synchronization with wallclock time on Unix/Linux
25 September 2004: Version 1.5alpha¶
New functionality/API additions
- SimEvents and signalling synchronization constructs, with ‘yield waitevent’ and ‘yield queueevent’ commands.
- A general “wait until” synchronization construct, with the ‘yield waituntil’ command.
No changes to 1.4.x API, i.e., existing code will work as before.
19 May 2004: Version 1.4.2¶
Sub-release to repair two bugs:
- The unittest for monitored Resource queues does not fail anymore.
- SimulationTrace now works correctly with “yield hold,self” form.
No functional or API changes
29 February 2004: Version 1.4.1¶
Sub-release to repair two bugs:
- The (optional) monitoring of the activeQ in Resource now works correctly.
- The “cellphone.py” example is now implemented correctly.
No functional or API changes
1 February 2004: Version 1.4¶
- Released on SourceForge.net
22 December 2003: Version 1.4 alpha¶
New functionality/API changes
- All classes in the SimPy API are now new style classes, i.e., they inherit from object either directly or indirectly.
- Module Monitor.py has been merged into module Simulation.py and all SimulationXXX.py modules. Import of Simulation or any SimulationXXX module now also imports Monitor.
- Some Monitor methods/attributes have changed. See Manual!
- Monitor now inherits from list.
- A class Histogram has been added to Simulation.py and all SimulationXXX.py modules.
- A module SimulationRT has been added which allows synchronization between simulated and wallclock time.
- A moduleSimulationStep which allows the execution of a simulation model event-by-event, with the facility to execute application code after each event.
- A Tk/Tkinter-based module SimGUI has been added which provides a SimPy GUI framework.
- A Tk/Tkinter-based module SimPlot has been added which provides for plot output from SimPy programs.
22 June 2003: Version 1.3¶
- No functional or API changes
- Reduction of sourcecode linelength in Simulation.py to <= 80 characters
June 2003: Version 1.3 alpha¶
Significantly improved performance
Significant increase in number of quasi-parallel processes SimPy can handle
New functionality/API changes:
- Addition of SimulationTrace, an event trace utility
- Addition of Lister, a prettyprinter for instance attributes
- No API changes
Internal changes:
- Implementation of a proposal by Simon Frost: storing the keys of the event set dictionary in a binary search tree using bisect. Thank you, Simon! SimPy 1.3 is dedicated to you!
Update of Manual to address tracing.
Update of Interfacing doc to address output visualization using Scientific Python gplt package.
29 April 2003: Version 1.2¶
- No changes in API.
- Internal changes:
- Defined “True” and “False” in Simulation.py to support Python 2.2.
22 October 2002¶
- Re-release of 0.5 Beta on SourceForge.net to replace corrupted file __init__.py.
- No code changes whatever!
18 October 2002¶
Version 0.5 Beta-release, intended to get testing by application developers and system integrators in preparation of first full (production) release. Released on SourceForge.net on 20 October 2002.
More models
Documentation enhanced by a manual, a tutorial (“The Bank”) and installation instructions.
Major changes to the API:
Introduced ‘simulate(until=0)’ instead of ‘scheduler(till=0)’. Left ‘scheduler()’ in for backward compatibility, but marked as deprecated.
Added attribute “name” to class Process. Process constructor is now:
def __init__(self,name="a_process")
Backward compatible if keyword parameters used.
Changed Resource constructor to:
def __init__(self,capacity=1,name="a_resource",unitName="units")
Backward compatible if keyword parameters used.
27 September 2002¶
- Version 0.2 Alpha-release, intended to attract feedback from users
- Extended list of models
- Upodated documentation
17 September 2002¶
- Version 0.1.2 published on SourceForge; fully working, pre-alpha code
- Implements simulation, shared resources with queuing (FIFO), and monitors for data gathering/analysis.
- Contains basic documentation (cheatsheet) and simulation models for test and demonstration.
Acknowledgments¶
SimPy 2 has been primarily developed by Stefan Scherfke and Ontje Lünsdorf, starting from SimPy 1.9. Their work has resulted in a most elegant combination of an object oriented API with the existing API, maintaining full backward compatibility. It has been quite easy to integrate their product into the existing SimPy code and documentation environment.
Thanks, guys, for this great job! SimPy 2.0 is dedicated to you!
SimPy was originally created by Klaus Müller and Tony Vignaux. They pushed its development for several years and built the SimPy community. Without them, there would be no SimPy 3.
Thanks, guys, for this great job! SimPy 3.0 is dedicated to you!
The many contributions of the SimPy user and developer communities are of course also gratefully acknowledged.
Ports and comparable libraries¶
Reimplementations of SimPy and libraries similar to SimPy are available in the following languages:
Defense of Design¶
This document explains why SimPy is designed the way it is and how its design evolved over time.
Original Design of SimPy 1¶
SimPy 1 was heavily inspired by Simula 67 and Simscript. The basic entity of the framework was a process. A process described a temporal sequence of actions.
In SimPy 1, you implemented a process by sub-classing Process
. The instance
of such a subclass carried both, process and simulation internal information,
whereas the latter wasn’t of any use to the process itself. The sequence of
actions of the process was specified in a method of the subclass, called the
process execution method (or PEM in short). A PEM interacted with the
simulation by yielding one of several keywords defined in the simulation
package.
The simulation itself was executed via module level functions. The simulation state was stored in the global scope. This made it very easy to implement and execute a simulation (despite from having to inherit from Process and instantianting the processes before starting their PEMs). However, having all simulation state global makes it hard to parallelize multiple simulations.
SimPy 1 also followed the “batteries included” approach, providing shared resources, monitoring, plotting, GUIs and multiple types of simulations (“normal”, real-time, manual stepping, with tracing).
The following code fragment shows how a simple simulation could be implemented in SimPy 1:
from SimPy.Simulation import Process, hold, initialize, activate, simulate
class MyProcess(Process):
def pem(self, repeat):
for i in range(repeat):
yield hold, self, 1
initialize()
proc = MyProcess()
activate(proc, proc.pem(3))
simulate(until=10)
sim = Simulation()
proc = MyProcess(sim=sim)
sim.activate(proc, proc.pem(3))
sim.simulate(until=10)
Changes in SimPy 2¶
SimPy 2 mostly sticked with SimPy 1’s design, but added an object orient API
for the execution of simulations, allowing them to be executed in parallel.
Since processes and the simulation state were so closely coupled, you now
needed to pass the Simulation
instance into your process to “bind” them to
that instance. Additionally, you still had to activate the process. If you
forgot to pass the simulation instance, the process would use a global instance
thereby breaking your program. SimPy 2’s OO-API looked like this:
from SimPy.Simulation import Simulation, Process, hold
class MyProcess(Process):
def pem(self, repeat):
for i in range(repeat):
yield hold, self, 1
sim = Simulation()
proc = MyProcess(sim=sim)
sim.activate(proc, proc.pem(3))
sim.simulate(until=10)
Changes and Decisions in SimPy 3¶
The original goals for SimPy 3 were to simplify and PEP8-ify its API and to clean up and modularize its internals. We knew from the beginning that our goals would not be achievable without breaking backwards compatibility with SimPy 2. However, we didn’t expect the API changes to become as extensive as they ended up to be.
We also removed some of the included batteries, namely SimPy’s plotting and GUI capabilities, since dedicated libraries like matplotlib or PySide do a much better job here.
However, by far the most changes are—from the end user’s view—mostly
syntactical. Thus, porting from 2 to 3 usually just means replacing a line of
SimPy 2 code with its SimPy3 equivalent (e.g., replacing yield hold, self,
1
with yield env.timeout(1)
).
In short, the most notable changes in SimPy 3 are:
- No more sub-classing of
Process
required. PEMs can even be simple module level functions. - The simulation state is now stored in an
Environment
which can also be used by a PEM to interact with the simulation. - PEMs now yield event objects. This implicates interesting new features and allows an easy extension with new event types.
These changes are causing the above example to now look like this:
from simpy import Environment, simulate
def pem(env, repeat):
for i in range(repeat):
yield env.timeout(i)
env = Environment()
env.process(pem(env, 7))
simulate(env, until=10)
The following sections describe these changes in detail:
No More Sub-classing of Process
¶
In SimPy 3, every Python generator can be used as a PEM, no matter if it is
a module level function or a method of an object. This reduces the amount of
code required for simple processes. The Process
class still exists, but you
don’t need to instantiate it by yourself, though. More on that later.
Processes Live in an Environment¶
Process and simulation state are decoupled. An Environment
holds the
simulation state and serves as base API for processes to create new events.
This allows you to implement advanced use cases by extending the Process
or
Environment
class without affecting other components.
For the same reason, the simulate()
method now is a module level function
that takes an environment to simulate.
Stronger Focus on Events¶
In former versions, PEMs needed to yield one of SimPy’s built-in keywords (like
hold
) to interact with the simulation. These keywords had to be imported
separately and were bound to some internal functions that were tightly
integrated with the Simulation
and Process
making it very hard to
extend SimPy with new functionality.
In SimPy 3, PEMs just need to yield events. There are various built-in event
types, but you can also create custom ones by making a subclass of
a BaseEvent
. Most events are generated by factory methods of
Environment
. For example, Environment.timeout()
creates a Timeout
event that replaces the hold
keyword.
The Process
is now also an event. You can now yield another process and
wait for it to finish. For example, think of a car-wash simulation were
“washing” is a process that the car processes can wait for once they enter the
washing station.
Creating Events via the Environment or Resources¶
The Environment
and resources have methods to create new events, e.g.
Environment.timeout()
or Resource.request()
. Each of these methods maps
to a certain event type. It creates a new instance of it and returns it, e.g.:
def event(self):
return Event()
To simplify things, we wanted to use the event classes directly as methods:
class Environment(object)
event = Event
This was, unfortunately, not directly possible and we had to wrap the classes
to behave like bound methods. Therefore, we introduced a BoundClass
:
class BoundClass(object):
"""Allows classes to behave like methods. The ``__get__()`` descriptor
is basically identical to ``function.__get__()`` and binds the first
argument of the ``cls`` to the descriptor instance.
"""
def __init__(self, cls):
self.cls = cls
def __get__(self, obj, type=None):
if obj is None:
return self.cls
return types.MethodType(self.cls, obj)
class Environment(object):
event = BoundClass(Event)
These methods are called a lot, so we added the event classes as
types.MethodType
to the instance of Environment
(or the resources,
respectively):
class Environment(object):
def __init__(self):
self.event = types.MethodType(Event, self)
It turned out the the class attributes (the BoundClass
instances) were now
quite useless, so we removed them allthough it was actually the “right” way to
to add classes as methods to another class.
Release Process¶
This process describes the steps to execute in order to release a new version of SimPy.
Preparations¶
Close all tickets for the next version.
Update the minium required versions of dependencies in
setup.py
. Update the exact version of all entries inrequirements.txt
.Run tox from the project root. All tests for all supported versions must pass:
$ tox [...] ________ summary ________ py27: commands succeeded py32: commands succeeded py33: commands succeeded pypy: commands succeeded congratulations :)
Note
Tox will use the
requirements.txt
to setup the venvs, so make sure you’ve updated it!Build the docs (HTML is enough). Make sure there are no errors and undefined references.
$ cd docs/ $ make clean html $ cd ..
Check if all authors are listed in
AUTHORS.txt
.Update the change logs (
CHANGES.txt
anddocs/about/history.rst
). Only keep changes for the current major release inCHANGES.txt
and reference the history page from there.Commit all changes:
$ hg ci -m 'Updated change log for the upcoming release.'
Update the version number in
simpy/__init__.py
andsetup.py
and commit:$ hg ci -m 'Bump version from x.y.z to a.b.c'
Warning
Do not yet tag and push the changes so that you can safely do a rollback if one of the next step fails and you need change something!
Write a draft for the announcement mail with a list of changes, acknowledgements and installation instructions. Everyone in the team should agree with it.
Build and release¶
Test the release process. Build a source distribution and a wheel package and test them:
$ python setup.py sdist bdist_wheel $ ls dist/ simpy-a.b.c-py2.py3-none-any.whl simpy-a.b.c.tar.gz
Try installing them:
$ rm -rf /tmp/simpy-sdist # ensure clean state if ran repeatedly $ virtualenv /tmp/simpy-sdist $ /tmp/simpy-sdist/bin/pip install dist/simpy-a.b.c.tar.gz
and
$ rm -rf /tmp/simpy-wheel # ensure clean state if ran repeatedly $ virtualenv /tmp/simpy-wheel $ /tmp/simpy-wheel/bin/pip install dist/simpy-a.b.c-py2.py3-none-any.whl
Create or check your accounts for the test server <https://testpypi.python.org/pypi> and PyPI. Update your
~/.pypirc
with your current credentials:[distutils] index-servers = pypi test [test] repository = https://testpypi.python.org/pypi username = <your test user name goes here> password = <your test password goes here> [pypi] repository = http://pypi.python.org/pypi username = <your production user name goes here> password = <your production password goes here>
Upload the distributions for the new version to the test server and test the installation again:
$ twine upload -r test dist/simpy*a.b.c* $ pip install -i https://testpypi.python.org/pypi simpy
Check if the package is displayed correctly: https://testpypi.python.org/pypi/simpy
Finally upload the package to PyPI and test its installation one last time:
$ twine upload -r pypi dist/simpy*a.b.c* $ pip install -U simpy
Check if the package is displayed correctly: https://pypi.python.org/pypi/simpy
Post release¶
Push your changes:
$ hg tag a.b.c $ hg push ssh://hg@bitbucket.org/simpy/simpy
Activate the documentation build for the new version.
Send the prepared email to the mailing list and post it on Google+.
Update Wikipedia entries.
Update Python Wiki
Post something to Planet Python (e.g., via Stefan’s blog).
License¶
The MIT License (MIT)
Copyright (c) 2013 Ontje Lünsdorf and Stefan Scherfke (also see AUTHORS.txt)
Permission is hereby granted, free of charge, to any person obtaining a copy of
this software and associated documentation files (the "Software"), to deal in
the Software without restriction, including without limitation the rights to
use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
the Software, and to permit persons to whom the Software is furnished to do so,
subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.