Struct KompactSystem 

pub struct KompactSystem { /* private fields */ }

A Kompact system is a collection of components and services

An instance of KompactSystem is created from a KompactConfig via the build function.

It is possible to run more than one Kompact system in a single process. This allows different settings to be used for different component groups, for example. It can also be used for testing communication in unit or integration tests.

Note

For some schedulers it can happen that components may switch from one system’s scheduler to the other’s when running multiple systems in the same process and communicating between them via channels or actor references. Generally, this shouldn’t be an issue, but it can invalidate assumptions on thread assignments, so it’s important to be aware of. If this behaviour needs to be avoided at all costs, one can either use a scheduler that doesn’t use thread-local variables to determine the target queue (e.g., crossbeam_channel_pool), or limit cross-system communication to network-only, incurring the associated serialisation/deserialisations costs.

Example

Build a system with default settings with:

use kompact::prelude::*;

let system = KompactConfig::default().build().wait().expect("system");

Implementations

impl KompactSystem

pub fn logger(&self) -> &Logger<Arc<Fuse<Async>>>

Get a reference to the system-wide Kompact logger

Example

info!(system.logger(), "Hello World from the system logger!");

pub fn config(&self) -> &Config

Get a reference to the system configuration

Use load_config_str or load_config_file to load values into the config object.

Example

use kompact::prelude::*;
let default_values = r#"a = 7"#;
let mut conf = KompactConfig::default();
conf.load_config_str(default_values);
let system = conf.build().wait().expect("system");
assert_eq!(Some(7i64), system.config()["a"].as_i64());

pub fn config_owned(&self) -> Arc<Config>

Get a owned reference to the system configuration

pub fn now(&self) -> Instant

Returns the current time according to the system timer.

pub fn create<C, F>(&self, f: F) -> Arc<Component<C>>

where F: FnOnce() -> C, C: ComponentDefinition + 'static,

Create a new component

Uses f to create an instance of a ComponentDefinition, which is the initialised to form a Component. Since components are shared between threads, the created component is wrapped into an Arc.

Newly created components are not started automatically. Use start or start_notify to start a newly created component, once it is connected properly.

If you need address this component via the network, see the register function.

Example

let c = system.create(TestComponent1::new);

pub fn create_unsupervised<C, F>(&self, f: F) -> Arc<Component<C>>

where F: FnOnce() -> C, C: ComponentDefinition + 'static,

Create a new system component

You must use this instead of create to create a new system component. During system initialisation the supervisor is not available, yet, so normal create calls will panic!

pub fn create_dedicated<C, F>(&self, f: F) -> Arc<Component<C>>

where F: FnOnce() -> C, C: ComponentDefinition + 'static,

Create a new component, which runs on its own dedicated thread.

Uses f to create an instance of a ComponentDefinition, which is the initialised to form a Component. Since components are shared between threads, the created component is wrapped into an Arc.

A dedicated thread is assigned to this component, which sleeps when the component has no work.

Newly created components are not started automatically. Use start or start_notify to start a newly created component, once it is connected properly.

If you need address this component via the network, see the register function.

pub fn create_dedicated_unsupervised<C, F>(&self, f: F) -> Arc<Component<C>>

where F: FnOnce() -> C, C: ComponentDefinition + 'static,

Create a new system component, which runs on its own dedicated thread.

A dedicated thread is assigned to this component, which sleeps when the component has no work.

You must use this instead of create_dedicated to create a new system component on its own thread. During system initialisation the supervisor is not available, yet, so normal create calls will panic!

pub fn register( &self, c: &dyn UniqueRegistrable, ) -> KFuture<Result<ActorPath, RegistrationError>>

Attempts to register c with the dispatcher using its unique id

The returned future will contain the unique id ActorPath for the given component, once it is completed by the dispatcher.

Once the future completes, the component can be addressed via the network, even if it has not been started, yet (in which case messages will simply be queued up).

Example

use kompact::prelude::*;
use std::time::Duration;
let system = KompactConfig::default().build().wait().expect("KompactSystem");
let c = system.create(TestComponent1::new);
system.register(&c).wait_expect(Duration::from_millis(1000), "Failed to register TestComponent1");

pub fn create_and_register<C, F>( &self, f: F, ) -> (Arc<Component<C>>, KFuture<Result<ActorPath, RegistrationError>>)

where F: FnOnce() -> C, C: ComponentDefinition + 'static,

Creates a new component and registers it with the dispatcher

This function is simply a convenience shortcut for create followed by register, as this combination is very common in networked Kompact systems.

Example

use kompact::prelude::*;
use std::time::Duration;
let system = KompactConfig::default().build().wait().expect("KompactSystem");
let (c, registration_future) = system.create_and_register(TestComponent1::new);
registration_future.wait_expect(Duration::from_millis(1000), "Failed to register TestComponent1");

pub fn register_by_alias<A>( &self, c: &dyn DynActorRefFactory, alias: A, ) -> KFuture<Result<ActorPath, RegistrationError>>

where A: Into<String>,

Attempts to register the provided component with a human-readable alias.

The returned future will contain the named ActorPath for the given alias, once it is completed by the dispatcher.

Alias registration will fail if a previous registration already exists. Use update_alias_registration to override an existing registration.

Note

While aliases are easier to read, lookup by unique ids is significantly more efficient. However, named aliases allow services to be taken over by another component when the original registrant failed, something that is not possible with unique paths. Thus, this kind of addressing lends itself to lookup-service style components, for example.

Example

use kompact::prelude::*;
use std::time::Duration;
let system = KompactConfig::default().build().wait().expect("KompactSystem");
let (c, unique_registration_future) = system.create_and_register(TestComponent1::new);
unique_registration_future.wait_expect(Duration::from_millis(1000), "Failed to register TestComponent1");
let alias_registration_future = system.register_by_alias(&c, "test");
alias_registration_future.wait_expect(Duration::from_millis(1000), "Failed to register TestComponent1 by alias");

pub fn update_alias_registration<A>( &self, c: &dyn DynActorRefFactory, alias: A, ) -> KFuture<Result<ActorPath, RegistrationError>>

where A: Into<String>,

Attempts to register the provided component with a human-readable alias.

The returned future will contain the named ActorPath for the given alias, once it is completed by the dispatcher.

This registration will replace any previous registration, if it exists.

Note

While aliases are easier to read, lookup by unique ids is significantly more efficient. However, named aliases allow services to be taken over by another component when the original registrant failed, something that is not possible with unique paths. Thus, this kind of addressing lends itself to lookup-service style components, for example.

Example

use kompact::prelude::*;
use std::time::Duration;
let system = KompactConfig::default().build().wait().expect("KompactSystem");
let (c, unique_registration_future) = system.create_and_register(TestComponent1::new);
unique_registration_future.wait_expect(Duration::from_millis(1000), "Failed to register TestComponent1");
let alias_registration_future = system.update_alias_registration(&c, "test");
alias_registration_future.wait_expect(Duration::from_millis(1000), "Failed to register TestComponent1 by alias");
let alias_reregistration_future = system.update_alias_registration(&c, "test");
alias_reregistration_future.wait_expect(Duration::from_millis(1000), "Failed to override TestComponent1 registration by alias");

pub fn set_routing_policy<P>( &self, policy: P, path: &str, update: bool, ) -> KFuture<Result<ActorPath, RegistrationError>>

where P: Into<StorePolicy>,

Attempts to set the routing policy at path

Setting a routing policy at a path “a” will include all actors paths registered under paths of the form “a/…” to be included as members of the routing group that the policy applies to.

Having an explicit routing policy at a path will cause routing over group members, even if no routing marker (e.g., “a/*” or “a/?” is given).

Overriding an existing actor or policy at the given path will fail, unless update is set to true.

Provided routing policies can be found in the routing::groups module.

Example

use kompact::prelude::*;
use kompact::routing::groups::*;
use std::time::Duration;
let system = KompactConfig::default().build().wait().expect("KompactSystem");
let policy_registration_future = system.set_routing_policy(BroadcastRouting::default(), "broadcast-me", false);
let broadcast_path = policy_registration_future.wait_expect(Duration::from_millis(1000), "Failed to set broadcast policy");
let c1 = system.create(TestComponent1::new);
let c2 = system.create(TestComponent1::new);
let alias_registration_future1 = system.register_by_alias(&c1, "broadcast-me/test1");
let alias_registration_future2 = system.register_by_alias(&c2, "broadcast-me/something/test2");
alias_registration_future1.wait_expect(Duration::from_millis(1000), "Failed to register TestComponent1 by alias");
alias_registration_future2.wait_expect(Duration::from_millis(1000), "Failed to register TestComponent2 by alias");
// sending to broadcast_path now will send to c1 and c2

pub fn start(&self, c: &Arc<impl AbstractComponent + ?Sized>)

Start a component

A component only handles events/messages once it is started. In particular, a component that isn’t started shouldn’t be scheduled and thus access to its definition should always succeed, for example via on_definition.

Example

let c = system.create(TestComponent1::new);
system.start(&c);

pub fn with_dispatcher_definition<F>(&self, f: F)

where F: FnOnce(&mut (dyn Any + 'static)),

Allows backend crates to interact with the concrete dispatcher definition.

The callback receives the dispatcher definition erased as Any. Backend crates can downcast it to their concrete dispatcher type in order to wire backend-specific ports or query backend-specific state.

pub fn start_notify( &self, c: &Arc<impl AbstractComponent + ?Sized>, ) -> KFuture<()>

Start a component and complete a future once it has started

When the returned future completes, the component is guaranteed to have started. However, it is not guaranteed to be in an active state, as it could already have been stopped or could have failed since.

A component only handles events/messages once it is started. In particular, a component that isn’t started shouldn’t be scheduled and thus access to its definition should always succeed, for example via on_definition.

Example

use std::time::Duration;
let c = system.create(TestComponent1::new);
system.start_notify(&c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never started!");

pub fn stop(&self, c: &Arc<impl AbstractComponent + ?Sized>)

Stop a component

A component does not handle any events/messages while it is stopped, but it does not get deallocated either. It can be started again later with start or start_notify.

A component that is stopped shouldn’t be scheduled and thus access to its definition should always succeed, for example via on_definition.

Example

use std::time::Duration;
let c = system.create(TestComponent1::new);
system.start_notify(&c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never started!");
system.stop(&c);

pub fn stop_notify( &self, c: &Arc<impl AbstractComponent + ?Sized>, ) -> KFuture<()>

Stop a component and complete a future once it has stopped

When the returned future completes, the component is guaranteed to have stopped. However, it is not guaranteed to be in a passive state, as it could already have been started again since.

A component does not handle any events/messages while it is stopped, but it does not get deallocated either. It can be started again later with start or start_notify.

A component that is stopped shouldn’t be scheduled and thus access to its definition should always succeed, for example via on_definition.

Example

use std::time::Duration;
let c = system.create(TestComponent1::new);
system.start_notify(&c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never started!");
system.stop_notify(&c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never stopped!");
system.start_notify(&c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never re-started!");

pub fn kill(&self, c: Arc<impl AbstractComponent + ?Sized>)

Stop and deallocate a component

The supervisor will attempt to deallocate c once it is stopped. However, if there are still outstanding references somewhere else in the system this will fail, of course. In that case the supervisor leaves a debug message in the logging output, so that this circumstance can be discovered if necessary.

Example

use std::time::Duration;
let c = system.create(TestComponent1::new);
system.start_notify(&c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never started!");
system.kill(c);

pub fn kill_notify( &self, c: Arc<impl AbstractComponent + ?Sized>, ) -> KFuture<()>

Stop and deallocate a component, and complete a future once it has stopped

The supervisor will attempt to deallocate c once it is stopped. However, if there are still outstanding references somewhere else in the system this will fail, of course. In that case the supervisor leaves a debug message in the logging output, so that this circumstance can be discovered if necessary.

Note

The completion of the future indicates that the component has been stopped, not that it has been deallocated.

If, for some reason, you really need to know when it has been deallocated, you need to hold on to a copy of the component, use try_unwrap and then call drop once you are successful.

Example

use std::time::Duration;
let c = system.create(TestComponent1::new);
system.start_notify(&c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never started!");
system.kill_notify(c)
      .wait_timeout(Duration::from_millis(1000))
      .expect("TestComponent1 never stopped!");

pub fn trigger_i<P>( &self, event: <P as Port>::Indication, port: &RequiredRef<P>, )

where P: Port + 'static,

Trigger an indication event on a shared port

This can be used to send events to component without connecting a channel. You only nneed to acquire port reference via share.

pub fn trigger_r<P>(&self, msg: <P as Port>::Request, port: &ProvidedRef<P>)

where P: Port + 'static,

Trigger a request event on a shared port

This can be used to send events to component without connecting a channel. You only nneed to acquire port reference via share.

pub fn throughput(&self) -> usize

Return the configured thoughput value

See also throughput.

pub fn max_messages(&self) -> usize

Return the configured maximum number of messages per scheduling

This value is based on throughput and msg_priority.

pub fn terminated(&self) -> impl Future<Output = ()> + Unpin + 'static

Return a future that completes when the Kompact system is terminated.

The returned future observes system state and completes once the system is destroyed or faulty. Async callers should .await it, while synchronous callers can use BlockingFutureExt::wait.

Note

Don’t use this method for any measurements, as its implementation currently does not include any notification logic. It simply checks its internal state every second or so, so there might be quite some delay until a shutdown is detected.

pub fn await_termination(self)

Wait for the Kompact system to be terminated

Suspends this thread until the system is terminated from some other thread, such as its own threadpool, for example.

Use terminated for the future-returning API.

Note

Don’t use this method for any measurements, as its implementation currently does not include any notification logic. It simply checks its internal state every second or so, so there might be quite some delay until a shutdown is detected.

pub fn shutdown( self, ) -> impl Future<Output = Result<(), ShutdownError>> + Send + Unpin + 'static

Shutdown the Kompact system.

Stops all components and then stops the scheduler.

This function may fail to stop in time (or at all), if components hang on to scheduler threads indefinitely.

This function returns a future. Async callers should .await it, while synchronous callers can use BlockingFutureExt::wait. The shutdown sequence awaits component shutdown without entering a nested executor. Scheduler implementations can also provide a non-blocking shutdown future via Scheduler::shutdown_notify.

Example

use kompact::prelude::*;

let system = KompactConfig::default().build().wait().expect("system");
system.shutdown().wait().expect("shutdown");

pub fn kill_system(self) -> Result<(), ShutdownError>

Kill the Kompact system

Similar to shutdown() except the Network will not be gracefully disconnected, making this shutdown more immediate and brutal.

Remote systems will perceive this system as crashed.

pub fn shutdown_async(&self)

Start shutting down the Kompact system from within a component.

This is a fire-and-forget helper for lifecycle and message handlers that cannot consume their system handle. Use shutdown when you own the system and want to wait for shutdown to finish.

Shutdown may fail to stop in time (or at all), if components hang on to scheduler threads indefinitely.

Example

use kompact::prelude::*;

#[derive(ComponentDefinition, Actor)]
struct Stopper {
   ctx: ComponentContext<Self>,
}
impl Stopper {
    fn new() -> Stopper {
        Stopper {
            ctx: ComponentContext::uninitialised(),
        }
    }
}
impl ComponentLifecycle for Stopper {
   fn on_start(&mut self) -> HandlerResult {
       self.ctx().system().shutdown_async();
       Handled::OK
    }
}
let system = KompactConfig::default().build().wait().expect("system");
let c = system.create(Stopper::new);
system.start(&c);
system.await_termination();

pub fn system_path(&self) -> SystemPath

Return the system path of this Kompact system

The system path forms a prefix for every ActorPath.

pub fn actor_path_for( &self, component: &Arc<impl AbstractComponent + ?Sized>, ) -> ActorPath

Generate an unique path for the given component

Produces a unique id ActorPath for component using the system path of this system.

The returned ActorPath is only useful if the component was first registered, otherwise all messages sent to it will land in the system’s deadletter box.

Note

If you pass in a component from a different system, you will get a perfectly valid ActorPath instance, but which can not receive any messages, unless you also registered the component with this system’s dispatcher. Suffice to say, crossing system boundaries in this manner is not recommended.

pub fn spawn<R>( &self, future: impl Future<Output = R> + Send + 'static, ) -> Receiver<R>

where R: Send + 'static,

Run a Future on this system’s executor pool and return a handle to the result

Handles can be awaited like any other future.

Note

The current API is not as efficient as calling FuturesExecutor::spawn directly, due to some trait object indirection in Kompact systems. Thus, if performance is important, it is recommended to maintain a (non trait-object) handle to the actual Executor pool being used and call its spawn function instead. This API is really just a somewhat roundabout convenience for doing the same.

pub fn deadletter_path(&self) -> ActorPath

The remote path for the deadletter box

Trait Implementations

impl ActorPathFactory for KompactSystem

Available on crate feature distributed only.

fn actor_path(&self) -> ActorPath

Returns the associated actor path.

impl ActorRefFactory for KompactSystem

fn actor_ref(&self) -> ActorRef<!>

Returns a reference to the deadletter box

type Message = !

The type of messages carried by references produced by this factory

impl Clone for KompactSystem

fn clone(&self) -> KompactSystem

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Dispatching for KompactSystem

Available on crate feature distributed only.

fn dispatcher_ref(&self) -> ActorRefStrong<DispatchEnvelope>

Returns the associated dispatcher reference.

impl DistributedSystemHandle for KompactSystem

Available on crate feature distributed only.

fn register( &self, c: &dyn UniqueRegistrable, ) -> KFuture<Result<ActorPath, RegistrationError>>

Attempts to register c with the dispatcher using its unique id.

fn create_and_register<C, F>( &self, f: F, ) -> (Arc<Component<C>>, KFuture<Result<ActorPath, RegistrationError>>)

where F: FnOnce() -> C, C: ComponentDefinition + 'static,

Creates a new component and registers it with the dispatcher.

fn register_by_alias<A>( &self, c: &dyn DynActorRefFactory, alias: A, ) -> KFuture<Result<ActorPath, RegistrationError>>

where A: Into<String>,

Attempts to register the provided component with a human-readable alias.

fn update_alias_registration<A>( &self, c: &dyn DynActorRefFactory, alias: A, ) -> KFuture<Result<ActorPath, RegistrationError>>

where A: Into<String>,

Attempts to update the registration for the given alias.

fn set_routing_policy<P>( &self, policy: P, path: &str, update: bool, ) -> KFuture<Result<ActorPath, RegistrationError>>

where P: Into<StorePolicy>,

Attempts to set the routing policy at path.

fn system_path(&self) -> SystemPath

Return the system path of this Kompact system.

impl NetworkSystemExt for KompactSystem

fn connect_network_status_port( &self, required: &mut RequiredPort<NetworkStatusPort>, )

impl SystemHandle for KompactSystem

fn now(&self) -> Instant

Returns the current time according to the system timer.

fn create<C, F>(&self, f: F) -> Arc<Component<C>>

where F: FnOnce() -> C, C: ComponentDefinition + 'static,

Create a new component

fn start(&self, c: &Arc<impl AbstractComponent + ?Sized>)

Start a component

fn start_notify(&self, c: &Arc<impl AbstractComponent + ?Sized>) -> KFuture<()>

Start a component and complete a future once it has started

fn stop(&self, c: &Arc<impl AbstractComponent + ?Sized>)

Stop a component

fn stop_notify(&self, c: &Arc<impl AbstractComponent + ?Sized>) -> KFuture<()>

Stop a component and complete a future once it has stopped

fn kill(&self, c: Arc<impl AbstractComponent + ?Sized>)

Stop and deallocate a component

fn kill_notify(&self, c: Arc<impl AbstractComponent + ?Sized>) -> KFuture<()>

Stop and deallocate a component, and complete a future once it has stopped

fn throughput(&self) -> usize

Return the configured thoughput value

fn max_messages(&self) -> usize

Return the configured maximum number of messages per scheduling

fn shutdown_async(&self)

Start shutting down the Kompact system from within a component.

fn deadletter_ref(&self) -> ActorRef<!>

Returns a reference to the system’s deadletter box

fn spawn<R>( &self, future: impl Future<Output = R> + Send + 'static, ) -> Receiver<R>

where R: Send + 'static,

Run a Future on this system’s executor pool and return a handle to the result

impl TimerRefFactory for KompactSystem

fn timer_ref(&self) -> TimerRef<Uuid, ActorRefState, ActorRefState>

Returns the timer reference for associated with this factory

fn now(&self) -> Instant

Returns the current time according to the underlying timer implementation.