hydro_lang/live_collections/stream/networking.rs
1//! Networking APIs for [`Stream`].
2
3use std::marker::PhantomData;
4
5use serde::Serialize;
6use serde::de::DeserializeOwned;
7use stageleft::{q, quote_type};
8use syn::parse_quote;
9
10use super::{ExactlyOnce, MinOrder, Ordering, Stream, TotalOrder};
11use crate::compile::builder::ExternalPortId;
12use crate::compile::ir::{
13 DebugInstantiate, HydroIrOpMetadata, HydroNode, HydroRoot, NetworkRecv, NetworkSend,
14};
15use crate::live_collections::boundedness::{Boundedness, Unbounded};
16use crate::live_collections::keyed_singleton::{KeyedSingleton, MonotonicKeys};
17use crate::live_collections::keyed_stream::KeyedStream;
18use crate::live_collections::sliced::sliced;
19use crate::live_collections::stream::Retries;
20#[cfg(feature = "sim")]
21use crate::location::LocationKey;
22use crate::location::cluster::{ClusterIds, Consistency, NoConsistency};
23#[cfg(stageleft_runtime)]
24use crate::location::dynamic::DynLocation;
25use crate::location::external_process::ExternalBincodeStream;
26use crate::location::{Cluster, External, Location, MemberId, MembershipEvent, Process};
27use crate::networking::{NetworkFor, TCP};
28use crate::nondet::{NonDet, nondet};
29use crate::properties::manual_proof;
30#[cfg(feature = "sim")]
31use crate::sim::SimReceiver;
32use crate::staging_util::get_this_crate;
33
34// same as the one in `hydro_std`, but internal use only
35fn track_membership<'a, C, L: Location<'a>>(
36 membership: KeyedStream<MemberId<C>, MembershipEvent, L, Unbounded>,
37) -> KeyedSingleton<MemberId<C>, bool, L, MonotonicKeys> {
38 membership.fold(
39 q!(|| false),
40 q!(|present, event| {
41 match event {
42 MembershipEvent::Joined => *present = true,
43 MembershipEvent::Left => *present = false,
44 }
45 }),
46 )
47}
48
49fn serialize_bincode_with_type(is_demux: bool, t_type: &syn::Type) -> syn::Expr {
50 let root = get_this_crate();
51
52 if is_demux {
53 parse_quote! {
54 #root::runtime_support::stageleft::runtime_support::fn1_type_hint::<(#root::__staged::location::MemberId<_>, #t_type), _>(
55 |(id, data)| {
56 (id.into_tagless(), #root::runtime_support::bincode::serialize(&data).unwrap().into())
57 }
58 )
59 }
60 } else {
61 parse_quote! {
62 #root::runtime_support::stageleft::runtime_support::fn1_type_hint::<#t_type, _>(
63 |data| {
64 #root::runtime_support::bincode::serialize(&data).unwrap().into()
65 }
66 )
67 }
68 }
69}
70
71pub(crate) fn serialize_bincode<T: Serialize>(is_demux: bool) -> syn::Expr {
72 serialize_bincode_with_type(is_demux, "e_type::<T>())
73}
74
75fn deserialize_bincode_with_type(tagged: Option<&syn::Type>, t_type: &syn::Type) -> syn::Expr {
76 let root = get_this_crate();
77 if let Some(c_type) = tagged {
78 parse_quote! {
79 |res| {
80 let (id, b) = res.unwrap();
81 (#root::__staged::location::MemberId::<#c_type>::from_tagless(id as #root::__staged::location::TaglessMemberId), #root::runtime_support::bincode::deserialize::<#t_type>(&b).unwrap())
82 }
83 }
84 } else {
85 parse_quote! {
86 |res| {
87 #root::runtime_support::bincode::deserialize::<#t_type>(&res.unwrap()).unwrap()
88 }
89 }
90 }
91}
92
93pub(crate) fn deserialize_bincode<T: DeserializeOwned>(tagged: Option<&syn::Type>) -> syn::Expr {
94 deserialize_bincode_with_type(tagged, "e_type::<T>())
95}
96
97impl<'a, T, L, B: Boundedness, O: Ordering, R: Retries> Stream<T, Process<'a, L>, B, O, R> {
98 #[deprecated = "use Stream::send(..., TCP.fail_stop().bincode()) instead"]
99 /// "Moves" elements of this stream to a new distributed location by sending them over the network,
100 /// using [`bincode`] to serialize/deserialize messages.
101 ///
102 /// The returned stream captures the elements received at the destination, where values will
103 /// asynchronously arrive over the network. Sending from a [`Process`] to another [`Process`]
104 /// preserves ordering and retries guarantees by using a single TCP channel to send the values. The
105 /// recipient is guaranteed to receive a _prefix_ or the sent messages; if the TCP connection is
106 /// dropped no further messages will be sent.
107 ///
108 /// # Example
109 /// ```rust
110 /// # #[cfg(feature = "deploy")] {
111 /// # use hydro_lang::prelude::*;
112 /// # use futures::StreamExt;
113 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p_out| {
114 /// let p1 = flow.process::<()>();
115 /// let numbers: Stream<_, Process<_>, Bounded> = p1.source_iter(q!(vec![1, 2, 3]));
116 /// let p2 = flow.process::<()>();
117 /// let on_p2: Stream<_, Process<_>, Unbounded> = numbers.send_bincode(&p2);
118 /// // 1, 2, 3
119 /// # on_p2.send_bincode(&p_out)
120 /// # }, |mut stream| async move {
121 /// # for w in 1..=3 {
122 /// # assert_eq!(stream.next().await, Some(w));
123 /// # }
124 /// # }));
125 /// # }
126 /// ```
127 pub fn send_bincode<L2>(
128 self,
129 other: &Process<'a, L2>,
130 ) -> Stream<T, Process<'a, L2>, Unbounded, O, R>
131 where
132 T: Serialize + DeserializeOwned,
133 {
134 self.send(other, TCP.fail_stop().bincode())
135 }
136
137 /// "Moves" elements of this stream to a new distributed location by sending them over the network,
138 /// using the configuration in `via` to set up the message transport.
139 ///
140 /// The returned stream captures the elements received at the destination, where values will
141 /// asynchronously arrive over the network. Sending from a [`Process`] to another [`Process`]
142 /// preserves ordering and retries guarantees when using a single TCP channel to send the values.
143 /// The recipient is guaranteed to receive a _prefix_ or the sent messages; if the connection is
144 /// dropped no further messages will be sent.
145 ///
146 /// # Example
147 /// ```rust
148 /// # #[cfg(feature = "deploy")] {
149 /// # use hydro_lang::prelude::*;
150 /// # use futures::StreamExt;
151 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p_out| {
152 /// let p1 = flow.process::<()>();
153 /// let numbers: Stream<_, Process<_>, Bounded> = p1.source_iter(q!(vec![1, 2, 3]));
154 /// let p2 = flow.process::<()>();
155 /// let on_p2: Stream<_, Process<_>, Unbounded> = numbers.send(&p2, TCP.fail_stop().bincode());
156 /// // 1, 2, 3
157 /// # on_p2.send(&p_out, TCP.fail_stop().bincode())
158 /// # }, |mut stream| async move {
159 /// # for w in 1..=3 {
160 /// # assert_eq!(stream.next().await, Some(w));
161 /// # }
162 /// # }));
163 /// # }
164 /// ```
165 pub fn send<L2, N: NetworkFor<T>>(
166 self,
167 to: &Process<'a, L2>,
168 via: N,
169 ) -> Stream<T, Process<'a, L2>, Unbounded, <O as MinOrder<N::OrderingGuarantee>>::Min, R>
170 where
171 O: MinOrder<N::OrderingGuarantee>,
172 {
173 let name = via.name();
174 if to.multiversioned() && name.is_none() {
175 panic!(
176 "Cannot send to a multiversioned location without a channel name. Please provide a name for the network."
177 );
178 }
179
180 let (serialize, deserialize) = if N::is_embedded() {
181 (
182 NetworkSend::Embedded {
183 tag: None,
184 element_type: quote_type::<T>().into(),
185 },
186 NetworkRecv::Embedded {
187 tag: None,
188 element_type: quote_type::<T>().into(),
189 },
190 )
191 } else {
192 (
193 NetworkSend::Custom {
194 serialize_fn: Some(N::serialize_thunk(false).into()),
195 },
196 NetworkRecv::Custom {
197 deserialize_fn: Some(N::deserialize_thunk(None).into()),
198 },
199 )
200 };
201
202 Stream::new(
203 to.clone(),
204 HydroNode::Network {
205 name: name.map(ToOwned::to_owned),
206 networking_info: N::networking_info(),
207 serialize,
208 deserialize,
209 instantiate_fn: DebugInstantiate::Building,
210 input: Box::new(self.ir_node.replace(HydroNode::Placeholder)),
211 metadata: to.new_node_metadata(Stream::<
212 T,
213 Process<'a, L2>,
214 Unbounded,
215 <O as MinOrder<N::OrderingGuarantee>>::Min,
216 R,
217 >::collection_kind()),
218 },
219 )
220 }
221
222 #[deprecated = "use Stream::broadcast(..., TCP.fail_stop().bincode()) instead"]
223 /// Broadcasts elements of this stream to all members of a cluster by sending them over the network,
224 /// using [`bincode`] to serialize/deserialize messages.
225 ///
226 /// Each element in the stream will be sent to **every** member of the cluster based on the latest
227 /// membership information. This is a common pattern in distributed systems for broadcasting data to
228 /// all nodes in a cluster. Unlike [`Stream::demux_bincode`], which requires `(MemberId, T)` tuples to
229 /// target specific members, `broadcast_bincode` takes a stream of **only data elements** and sends
230 /// each element to all cluster members.
231 ///
232 /// # Non-Determinism
233 /// The set of cluster members may asynchronously change over time. Each element is only broadcast
234 /// to the current cluster members _at that point in time_. Depending on when we are notified of
235 /// membership changes, we will broadcast each element to different members.
236 ///
237 /// # Example
238 /// ```rust
239 /// # #[cfg(feature = "deploy")] {
240 /// # use hydro_lang::prelude::*;
241 /// # use futures::StreamExt;
242 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
243 /// let p1 = flow.process::<()>();
244 /// let workers: Cluster<()> = flow.cluster::<()>();
245 /// let numbers: Stream<_, Process<_>, _> = p1.source_iter(q!(vec![123]));
246 /// let on_worker: Stream<_, Cluster<_>, _> = numbers.broadcast_bincode(&workers, nondet!(/** assuming stable membership */));
247 /// # on_worker.send_bincode(&p2).entries()
248 /// // if there are 4 members in the cluster, each receives one element
249 /// // - MemberId::<()>(0): [123]
250 /// // - MemberId::<()>(1): [123]
251 /// // - MemberId::<()>(2): [123]
252 /// // - MemberId::<()>(3): [123]
253 /// # }, |mut stream| async move {
254 /// # let mut results = Vec::new();
255 /// # for w in 0..4 {
256 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
257 /// # }
258 /// # results.sort();
259 /// # assert_eq!(results, vec!["(MemberId::<()>(0), 123)", "(MemberId::<()>(1), 123)", "(MemberId::<()>(2), 123)", "(MemberId::<()>(3), 123)"]);
260 /// # }));
261 /// # }
262 /// ```
263 pub fn broadcast_bincode<L2: 'a>(
264 self,
265 other: &Cluster<'a, L2>,
266 nondet_membership: NonDet,
267 ) -> Stream<T, Cluster<'a, L2>, Unbounded, O, R>
268 where
269 T: Clone + Serialize + DeserializeOwned,
270 {
271 self.broadcast(other, TCP.fail_stop().bincode(), nondet_membership)
272 }
273
274 /// Broadcasts elements of this stream to all members of a cluster by sending them over the network,
275 /// using the configuration in `via` to set up the message transport.
276 ///
277 /// Each element in the stream will be sent to **every** member of the cluster based on the latest
278 /// membership information. This is a common pattern in distributed systems for broadcasting data to
279 /// all nodes in a cluster. Unlike [`Stream::demux`], which requires `(MemberId, T)` tuples to
280 /// target specific members, `broadcast` takes a stream of **only data elements** and sends
281 /// each element to all cluster members.
282 ///
283 /// # Non-Determinism
284 /// The set of cluster members may asynchronously change over time. Each element is only broadcast
285 /// to the current cluster members _at that point in time_. Depending on when we are notified of
286 /// membership changes, we will broadcast each element to different members.
287 ///
288 /// # Example
289 /// ```rust
290 /// # #[cfg(feature = "deploy")] {
291 /// # use hydro_lang::prelude::*;
292 /// # use futures::StreamExt;
293 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
294 /// let p1 = flow.process::<()>();
295 /// let workers: Cluster<()> = flow.cluster::<()>();
296 /// let numbers: Stream<_, Process<_>, _> = p1.source_iter(q!(vec![123]));
297 /// let on_worker: Stream<_, Cluster<_>, _> = numbers.broadcast(&workers, TCP.fail_stop().bincode(), nondet!(/** assuming stable membership */));
298 /// # on_worker.send(&p2, TCP.fail_stop().bincode()).entries()
299 /// // if there are 4 members in the cluster, each receives one element
300 /// // - MemberId::<()>(0): [123]
301 /// // - MemberId::<()>(1): [123]
302 /// // - MemberId::<()>(2): [123]
303 /// // - MemberId::<()>(3): [123]
304 /// # }, |mut stream| async move {
305 /// # let mut results = Vec::new();
306 /// # for w in 0..4 {
307 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
308 /// # }
309 /// # results.sort();
310 /// # assert_eq!(results, vec!["(MemberId::<()>(0), 123)", "(MemberId::<()>(1), 123)", "(MemberId::<()>(2), 123)", "(MemberId::<()>(3), 123)"]);
311 /// # }));
312 /// # }
313 /// ```
314 pub fn broadcast<L2: 'a, N: NetworkFor<T>>(
315 self,
316 to: &Cluster<'a, L2>,
317 via: N,
318 nondet_membership: NonDet,
319 ) -> Stream<T, Cluster<'a, L2>, Unbounded, <O as MinOrder<N::OrderingGuarantee>>::Min, R>
320 where
321 T: Clone,
322 O: MinOrder<N::OrderingGuarantee>,
323 {
324 // TODO(#1875): the membership snapshot below is over a `KeyedSingleton`, and keyed
325 // sim hooks do not exist yet. Once they do, expose a composite hook payload here
326 // (`NonDet<(Option<KeyedSnapshotHook<..>>, Option<BatchHook<T, O, R>>)>`) so tests
327 // can script the membership snapshot and the element batching independently.
328 let ids = track_membership(self.location.source_cluster_membership_stream(
329 to,
330 nondet!(/** dropped prefixes don't affect broadcast */),
331 ));
332 sliced! {
333 let members_snapshot = use::snapshot(ids, nondet!(
334 /// membership timing is captured by the caller's guard
335 nondet_membership
336 ));
337 let elements = use::batch(self, nondet!(
338 /// batching timing is captured by the caller's guard
339 nondet_membership
340 ));
341
342 let current_members = members_snapshot.filter(q!(|b| *b));
343 elements.repeat_with_keys(current_members)
344 }
345 .demux(to, via)
346 }
347
348 /// Broadcasts elements of this stream to all members of a cluster,
349 /// assuming membership is closed (fixed at deploy time).
350 ///
351 /// Unlike [`Stream::broadcast`], this does not require a [`NonDet`] guard.
352 /// The membership set is obtained from deploy metadata via
353 /// [`ClusterIds`], producing a
354 /// `Bounded` stream. The cross-product of data × members is fully
355 /// deterministic.
356 ///
357 /// The consistency guarantee of the output depends on the network's failure policy
358 /// ([`NetworkFor::ConsistencyGuarantee`]). Policies like `fail_stop` and
359 /// `lossy_delayed_forever` guarantee that every live member eventually materializes the same
360 /// elements, so the output is
361 /// [`EventualConsistency`](crate::location::cluster::EventualConsistency). A plain `lossy`
362 /// policy can drop individual messages for some members while delivering them to others, so
363 /// replicas may permanently diverge and the output only has
364 /// [`NoConsistency`].
365 ///
366 /// This is only available in deployment targets with static cluster
367 /// membership (legacy Hydro Deploy and simulation). There are no late
368 /// joiners in that context, so broadcast receivers are guaranteed to
369 /// get data from the start of the stream. On dynamic targets
370 /// (e.g. ECS), use [`Stream::broadcast`] instead.
371 ///
372 /// # Example
373 /// ```rust
374 /// # #[cfg(feature = "deploy")] {
375 /// # use hydro_lang::prelude::*;
376 /// # use futures::StreamExt;
377 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
378 /// let p1 = flow.process::<()>();
379 /// let workers: Cluster<()> = flow.cluster::<()>();
380 /// let numbers: Stream<_, Process<_>, _> = p1.source_iter(q!(vec![123]));
381 /// let on_worker = numbers.broadcast_closed(&workers, TCP.fail_stop().bincode());
382 /// # on_worker.send(&p2, TCP.fail_stop().bincode()).entries()
383 /// // each of the 4 cluster members receives 123
384 /// # }, |mut stream| async move {
385 /// # let mut results = Vec::new();
386 /// # for _ in 0..4 {
387 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
388 /// # }
389 /// # results.sort();
390 /// # assert_eq!(results, vec!["(MemberId::<()>(0), 123)", "(MemberId::<()>(1), 123)", "(MemberId::<()>(2), 123)", "(MemberId::<()>(3), 123)"]);
391 /// # }));
392 /// # }
393 /// ```
394 pub fn broadcast_closed<L2: 'a, N: NetworkFor<T>>(
395 self,
396 to: &Cluster<'a, L2>,
397 via: N,
398 ) -> Stream<
399 T,
400 Cluster<'a, L2, N::ConsistencyGuarantee>,
401 Unbounded,
402 <O as MinOrder<N::OrderingGuarantee>>::Min,
403 R,
404 >
405 where
406 T: Clone,
407 O: MinOrder<N::OrderingGuarantee>,
408 {
409 let cluster_ids = ClusterIds {
410 key: to.key,
411 _phantom: PhantomData,
412 };
413 let member_ids = self.location.source_iter(q!(cluster_ids
414 .iter()
415 .map(|id| MemberId::from_tagless(id.clone()))));
416
417 // Late joiners will receive no data from this broadcast, which is
418 // future-monotone and eventually consistent (a safe under-approximation).
419 self.cross_product(member_ids)
420 .map(q!(|(data, member_id)| (member_id, data)))
421 .into_keyed()
422 .demux(to, via)
423 .assert_has_consistency_of_trusted(manual_proof!(
424 /// With a network whose failure policy delivers the same messages to every live
425 /// member (tracked by `NetworkFor::ConsistencyGuarantee`), a closed broadcast
426 /// will materialize the same elements on each member.
427 ))
428 }
429
430 /// Sends the elements of this stream to an external (non-Hydro) process, using [`bincode`]
431 /// serialization. The external process can receive these elements by establishing a TCP
432 /// connection and decoding using [`tokio_util::codec::LengthDelimitedCodec`].
433 ///
434 /// # Example
435 /// ```rust
436 /// # #[cfg(feature = "deploy")] {
437 /// # use hydro_lang::prelude::*;
438 /// # use futures::StreamExt;
439 /// # tokio_test::block_on(async move {
440 /// let mut flow = FlowBuilder::new();
441 /// let process = flow.process::<()>();
442 /// let numbers: Stream<_, Process<_>, Bounded> = process.source_iter(q!(vec![1, 2, 3]));
443 /// let external = flow.external::<()>();
444 /// let external_handle = numbers.send_bincode_external(&external);
445 ///
446 /// let mut deployment = hydro_deploy::Deployment::new();
447 /// let nodes = flow
448 /// .with_process(&process, deployment.Localhost())
449 /// .with_external(&external, deployment.Localhost())
450 /// .deploy(&mut deployment);
451 ///
452 /// deployment.deploy().await.unwrap();
453 /// // establish the TCP connection
454 /// let mut external_recv_stream = nodes.connect(external_handle).await;
455 /// deployment.start().await.unwrap();
456 ///
457 /// for w in 1..=3 {
458 /// assert_eq!(external_recv_stream.next().await, Some(w));
459 /// }
460 /// # });
461 /// # }
462 /// ```
463 pub fn send_bincode_external<L2>(
464 self,
465 other: &External<'_, L2>,
466 ) -> ExternalBincodeStream<T, O, R>
467 where
468 T: Serialize + DeserializeOwned,
469 {
470 let external_port_id =
471 self.register_serialized_external_port(other, serialize_bincode::<T>(false));
472
473 ExternalBincodeStream {
474 process_key: other.key,
475 port_id: external_port_id,
476 _phantom: PhantomData,
477 }
478 }
479
480 // TODO: Add a codec-parameterized external stream handle once deployment supports custom codecs.
481 fn register_serialized_external_port<L2>(
482 self,
483 other: &External<'_, L2>,
484 serialize_pipeline: syn::Expr,
485 ) -> ExternalPortId {
486 let mut flow_state_borrow = self.location.flow_state().borrow_mut();
487
488 let external_port_id = flow_state_borrow.next_external_port();
489
490 flow_state_borrow.push_root(HydroRoot::SendExternal {
491 to_external_key: other.key,
492 to_port_id: external_port_id,
493 to_many: false,
494 unpaired: true,
495 serialize_fn: Some(serialize_pipeline.into()),
496 instantiate_fn: DebugInstantiate::Building,
497 input: Box::new(self.ir_node.replace(HydroNode::Placeholder)),
498 op_metadata: HydroIrOpMetadata::new(),
499 });
500
501 external_port_id
502 }
503
504 #[cfg(feature = "sim")]
505 /// Sets up a bincode-encoded simulation output port for this stream, allowing test code to
506 /// receive elements sent to this stream during simulation. Use [`Stream::sim_output_with`] to
507 /// select another codec.
508 pub fn sim_output(self) -> SimReceiver<T, O, R>
509 where
510 T: Serialize + DeserializeOwned,
511 {
512 self.sim_output_with(crate::sim::codec::BincodeCodec)
513 }
514
515 #[cfg(feature = "sim")]
516 /// Sets up a simulation output port using `codec`, allowing test code to receive elements
517 /// sent to this stream during simulation. Custom codecs implement
518 /// [`SimCodec`](crate::sim::codec::SimCodec), which documents where they must be defined.
519 pub fn sim_output_with<C>(self, _codec: C) -> SimReceiver<T, O, R>
520 where
521 C: crate::sim::codec::SimCodec<T>,
522 {
523 let external_location: External<'a, ()> = External {
524 key: LocationKey::FIRST,
525 flow_state: self.location.flow_state().clone(),
526 _phantom: PhantomData,
527 };
528
529 let external_port_id = self.register_serialized_external_port(
530 &external_location,
531 crate::sim::codec::staged_serialize::<T, C>(),
532 );
533
534 SimReceiver(external_port_id, PhantomData, C::decode)
535 }
536}
537
538impl<'a, T, L: Location<'a>, B: Boundedness> Stream<T, L, B, TotalOrder, ExactlyOnce> {
539 /// Creates an external output for embedded deployment mode.
540 ///
541 /// The `name` parameter specifies the name of the field in the generated
542 /// `EmbeddedOutputs` struct that will receive elements from this stream.
543 /// The generated function will accept an `EmbeddedOutputs` struct with an
544 /// `impl FnMut(T)` field with this name.
545 pub fn embedded_output(self, name: impl Into<String>) {
546 let ident = syn::Ident::new(&name.into(), proc_macro2::Span::call_site());
547
548 self.location
549 .flow_state()
550 .borrow_mut()
551 .push_root(HydroRoot::EmbeddedOutput {
552 ident,
553 input: Box::new(self.ir_node.replace(HydroNode::Placeholder)),
554 op_metadata: HydroIrOpMetadata::new(),
555 });
556 }
557}
558
559impl<'a, T, L, L2, B: Boundedness, O: Ordering, R: Retries>
560 Stream<(MemberId<L2>, T), Process<'a, L>, B, O, R>
561{
562 #[deprecated = "use Stream::demux(..., TCP.fail_stop().bincode()) instead"]
563 /// Sends elements of this stream to specific members of a cluster, identified by a [`MemberId`],
564 /// using [`bincode`] to serialize/deserialize messages.
565 ///
566 /// Each element in the stream must be a tuple `(MemberId<L2>, T)` where the first element
567 /// specifies which cluster member should receive the data. Unlike [`Stream::broadcast_bincode`],
568 /// this API allows precise targeting of specific cluster members rather than broadcasting to
569 /// all members.
570 ///
571 /// # Example
572 /// ```rust
573 /// # #[cfg(feature = "deploy")] {
574 /// # use hydro_lang::prelude::*;
575 /// # use futures::StreamExt;
576 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
577 /// let p1 = flow.process::<()>();
578 /// let workers: Cluster<()> = flow.cluster::<()>();
579 /// let numbers: Stream<_, Process<_>, _> = p1.source_iter(q!(vec![0, 1, 2, 3]));
580 /// let on_worker: Stream<_, Cluster<_>, _> = numbers
581 /// .map(q!(|x| (hydro_lang::location::MemberId::from_raw_id(x), x)))
582 /// .demux_bincode(&workers);
583 /// # on_worker.send_bincode(&p2).entries()
584 /// // if there are 4 members in the cluster, each receives one element
585 /// // - MemberId::<()>(0): [0]
586 /// // - MemberId::<()>(1): [1]
587 /// // - MemberId::<()>(2): [2]
588 /// // - MemberId::<()>(3): [3]
589 /// # }, |mut stream| async move {
590 /// # let mut results = Vec::new();
591 /// # for w in 0..4 {
592 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
593 /// # }
594 /// # results.sort();
595 /// # assert_eq!(results, vec!["(MemberId::<()>(0), 0)", "(MemberId::<()>(1), 1)", "(MemberId::<()>(2), 2)", "(MemberId::<()>(3), 3)"]);
596 /// # }));
597 /// # }
598 /// ```
599 pub fn demux_bincode(
600 self,
601 other: &Cluster<'a, L2>,
602 ) -> Stream<T, Cluster<'a, L2>, Unbounded, O, R>
603 where
604 T: Serialize + DeserializeOwned,
605 {
606 self.demux(other, TCP.fail_stop().bincode())
607 }
608
609 /// Sends elements of this stream to specific members of a cluster, identified by a [`MemberId`],
610 /// using the configuration in `via` to set up the message transport.
611 ///
612 /// Each element in the stream must be a tuple `(MemberId<L2>, T)` where the first element
613 /// specifies which cluster member should receive the data. Unlike [`Stream::broadcast`],
614 /// this API allows precise targeting of specific cluster members rather than broadcasting to
615 /// all members.
616 ///
617 /// # Example
618 /// ```rust
619 /// # #[cfg(feature = "deploy")] {
620 /// # use hydro_lang::prelude::*;
621 /// # use futures::StreamExt;
622 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
623 /// let p1 = flow.process::<()>();
624 /// let workers: Cluster<()> = flow.cluster::<()>();
625 /// let numbers: Stream<_, Process<_>, _> = p1.source_iter(q!(vec![0, 1, 2, 3]));
626 /// let on_worker: Stream<_, Cluster<_>, _> = numbers
627 /// .map(q!(|x| (hydro_lang::location::MemberId::from_raw_id(x), x)))
628 /// .demux(&workers, TCP.fail_stop().bincode());
629 /// # on_worker.send(&p2, TCP.fail_stop().bincode()).entries()
630 /// // if there are 4 members in the cluster, each receives one element
631 /// // - MemberId::<()>(0): [0]
632 /// // - MemberId::<()>(1): [1]
633 /// // - MemberId::<()>(2): [2]
634 /// // - MemberId::<()>(3): [3]
635 /// # }, |mut stream| async move {
636 /// # let mut results = Vec::new();
637 /// # for w in 0..4 {
638 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
639 /// # }
640 /// # results.sort();
641 /// # assert_eq!(results, vec!["(MemberId::<()>(0), 0)", "(MemberId::<()>(1), 1)", "(MemberId::<()>(2), 2)", "(MemberId::<()>(3), 3)"]);
642 /// # }));
643 /// # }
644 /// ```
645 pub fn demux<N: NetworkFor<T>>(
646 self,
647 to: &Cluster<'a, L2>,
648 via: N,
649 ) -> Stream<
650 T,
651 Cluster<'a, L2, NoConsistency>,
652 Unbounded,
653 <O as MinOrder<N::OrderingGuarantee>>::Min,
654 R,
655 >
656 where
657 O: MinOrder<N::OrderingGuarantee>,
658 {
659 self.into_keyed().demux(to, via)
660 }
661}
662
663impl<'a, T, L, B: Boundedness> Stream<T, Process<'a, L>, B, TotalOrder, ExactlyOnce> {
664 #[deprecated = "use Stream::round_robin(..., TCP.fail_stop().bincode()) instead"]
665 /// Distributes elements of this stream to cluster members in a round-robin fashion, using
666 /// [`bincode`] to serialize/deserialize messages.
667 ///
668 /// This provides load balancing by evenly distributing work across cluster members. The
669 /// distribution is deterministic based on element order - the first element goes to member 0,
670 /// the second to member 1, and so on, wrapping around when reaching the end of the member list.
671 ///
672 /// # Non-Determinism
673 /// The set of cluster members may asynchronously change over time. Each element is distributed
674 /// based on the current cluster membership _at that point in time_. Depending on when cluster
675 /// members join and leave, the round-robin pattern will change. Furthermore, even when the
676 /// membership is stable, the order of members in the round-robin pattern may change across runs.
677 ///
678 /// # Ordering Requirements
679 /// This method is only available on streams with [`TotalOrder`] and [`ExactlyOnce`], since the
680 /// order of messages and retries affects the round-robin pattern.
681 ///
682 /// # Example
683 /// ```rust
684 /// # #[cfg(feature = "deploy")] {
685 /// # use hydro_lang::prelude::*;
686 /// # use hydro_lang::live_collections::stream::{TotalOrder, ExactlyOnce};
687 /// # use futures::StreamExt;
688 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
689 /// let p1 = flow.process::<()>();
690 /// let workers: Cluster<()> = flow.cluster::<()>();
691 /// let numbers: Stream<_, Process<_>, _, TotalOrder, ExactlyOnce> = p1.source_iter(q!(vec![1, 2, 3, 4]));
692 /// let on_worker: Stream<_, Cluster<_>, _> = numbers.round_robin_bincode(&workers, nondet!(/** assuming stable membership */));
693 /// on_worker.send_bincode(&p2)
694 /// # .first().values() // we use first to assert that each member gets one element
695 /// // with 4 cluster members, elements are distributed (with a non-deterministic round-robin order):
696 /// // - MemberId::<()>(?): [1]
697 /// // - MemberId::<()>(?): [2]
698 /// // - MemberId::<()>(?): [3]
699 /// // - MemberId::<()>(?): [4]
700 /// # }, |mut stream| async move {
701 /// # let mut results = Vec::new();
702 /// # for w in 0..4 {
703 /// # results.push(stream.next().await.unwrap());
704 /// # }
705 /// # results.sort();
706 /// # assert_eq!(results, vec![1, 2, 3, 4]);
707 /// # }));
708 /// # }
709 /// ```
710 pub fn round_robin_bincode<L2: 'a>(
711 self,
712 other: &Cluster<'a, L2>,
713 nondet_membership: NonDet,
714 ) -> Stream<T, Cluster<'a, L2>, Unbounded, TotalOrder, ExactlyOnce>
715 where
716 T: Serialize + DeserializeOwned,
717 {
718 self.round_robin(other, TCP.fail_stop().bincode(), nondet_membership)
719 }
720
721 /// Distributes elements of this stream to cluster members in a round-robin fashion, using
722 /// the configuration in `via` to set up the message transport.
723 ///
724 /// This provides load balancing by evenly distributing work across cluster members. The
725 /// distribution is deterministic based on element order - the first element goes to member 0,
726 /// the second to member 1, and so on, wrapping around when reaching the end of the member list.
727 ///
728 /// # Non-Determinism
729 /// The set of cluster members may asynchronously change over time. Each element is distributed
730 /// based on the current cluster membership _at that point in time_. Depending on when cluster
731 /// members join and leave, the round-robin pattern will change. Furthermore, even when the
732 /// membership is stable, the order of members in the round-robin pattern may change across runs.
733 ///
734 /// # Ordering Requirements
735 /// This method is only available on streams with [`TotalOrder`] and [`ExactlyOnce`], since the
736 /// order of messages and retries affects the round-robin pattern.
737 ///
738 /// # Example
739 /// ```rust
740 /// # #[cfg(feature = "deploy")] {
741 /// # use hydro_lang::prelude::*;
742 /// # use hydro_lang::live_collections::stream::{TotalOrder, ExactlyOnce};
743 /// # use futures::StreamExt;
744 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
745 /// let p1 = flow.process::<()>();
746 /// let workers: Cluster<()> = flow.cluster::<()>();
747 /// let numbers: Stream<_, Process<_>, _, TotalOrder, ExactlyOnce> = p1.source_iter(q!(vec![1, 2, 3, 4]));
748 /// let on_worker: Stream<_, Cluster<_>, _> = numbers.round_robin(&workers, TCP.fail_stop().bincode(), nondet!(/** assuming stable membership */));
749 /// on_worker.send(&p2, TCP.fail_stop().bincode())
750 /// # .first().values() // we use first to assert that each member gets one element
751 /// // with 4 cluster members, elements are distributed (with a non-deterministic round-robin order):
752 /// // - MemberId::<()>(?): [1]
753 /// // - MemberId::<()>(?): [2]
754 /// // - MemberId::<()>(?): [3]
755 /// // - MemberId::<()>(?): [4]
756 /// # }, |mut stream| async move {
757 /// # let mut results = Vec::new();
758 /// # for w in 0..4 {
759 /// # results.push(stream.next().await.unwrap());
760 /// # }
761 /// # results.sort();
762 /// # assert_eq!(results, vec![1, 2, 3, 4]);
763 /// # }));
764 /// # }
765 /// ```
766 pub fn round_robin<L2: 'a, N: NetworkFor<T>>(
767 self,
768 to: &Cluster<'a, L2>,
769 via: N,
770 nondet_membership: NonDet,
771 ) -> Stream<T, Cluster<'a, L2>, Unbounded, N::OrderingGuarantee, ExactlyOnce> {
772 // TODO(#1875): the membership snapshot below is over a `KeyedSingleton`, and keyed
773 // sim hooks do not exist yet. Once they do, expose a composite hook payload here
774 // (`NonDet<(Option<KeyedSnapshotHook<..>>, Option<BatchHook<T, O, R>>)>`) so tests
775 // can script the membership snapshot and the element batching independently.
776 let ids = track_membership(self.location.source_cluster_membership_stream(
777 to,
778 nondet!(/** dropped prefixes don't affect broadcast */),
779 ));
780 sliced! {
781 let members_snapshot = use::snapshot(ids, nondet!(
782 /// membership timing is captured by the caller's guard
783 nondet_membership
784 ));
785 let elements = use::batch(self.enumerate(), nondet!(
786 /// batching timing is captured by the caller's guard
787 nondet_membership
788 ));
789
790 let current_members = members_snapshot
791 .filter(q!(|b| *b))
792 .keys()
793 .assume_ordering::<TotalOrder>(nondet!(/** membership timing is captured by the caller guard */ nondet_membership))
794 .collect_vec();
795
796 elements
797 .cross_singleton(current_members)
798 .filter_map(q!(|(data, members)| {
799 if members.is_empty() {
800 None
801 } else {
802 Some((members[data.0 % members.len()].clone(), data.1))
803 }
804 }))
805 }
806 .demux(to, via)
807 }
808}
809
810impl<'a, T, L, B: Boundedness, C: Consistency>
811 Stream<T, Cluster<'a, L, C>, B, TotalOrder, ExactlyOnce>
812{
813 #[deprecated = "use Stream::round_robin(..., TCP.fail_stop().bincode()) instead"]
814 /// Distributes elements of this stream to cluster members in a round-robin fashion, using
815 /// [`bincode`] to serialize/deserialize messages.
816 ///
817 /// This provides load balancing by evenly distributing work across cluster members. The
818 /// distribution is deterministic based on element order - the first element goes to member 0,
819 /// the second to member 1, and so on, wrapping around when reaching the end of the member list.
820 ///
821 /// # Non-Determinism
822 /// The set of cluster members may asynchronously change over time. Each element is distributed
823 /// based on the current cluster membership _at that point in time_. Depending on when cluster
824 /// members join and leave, the round-robin pattern will change. Furthermore, even when the
825 /// membership is stable, the order of members in the round-robin pattern may change across runs.
826 ///
827 /// # Ordering Requirements
828 /// This method is only available on streams with [`TotalOrder`] and [`ExactlyOnce`], since the
829 /// order of messages and retries affects the round-robin pattern.
830 ///
831 /// # Example
832 /// ```rust
833 /// # #[cfg(feature = "deploy")] {
834 /// # use hydro_lang::prelude::*;
835 /// # use hydro_lang::live_collections::stream::{TotalOrder, ExactlyOnce, NoOrder};
836 /// # use hydro_lang::location::MemberId;
837 /// # use futures::StreamExt;
838 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
839 /// let p1 = flow.process::<()>();
840 /// let workers1: Cluster<()> = flow.cluster::<()>();
841 /// let workers2: Cluster<()> = flow.cluster::<()>();
842 /// let numbers: Stream<_, Process<_>, _, TotalOrder, ExactlyOnce> = p1.source_iter(q!(0..=16));
843 /// let on_worker1: Stream<_, Cluster<_>, _> = numbers.round_robin_bincode(&workers1, nondet!(/** assuming stable membership */));
844 /// let on_worker2: Stream<_, Cluster<_>, _> = on_worker1.round_robin_bincode(&workers2, nondet!(/** assuming stable membership */)).entries().assume_ordering(nondet!(/** assuming stable membership */));
845 /// on_worker2.send_bincode(&p2)
846 /// # .entries()
847 /// # .map(q!(|(w2, (w1, v))| ((w2, w1), v)))
848 /// # }, |mut stream| async move {
849 /// # let mut results = Vec::new();
850 /// # let mut locations = std::collections::HashSet::new();
851 /// # for w in 0..=16 {
852 /// # let (location, v) = stream.next().await.unwrap();
853 /// # locations.insert(location);
854 /// # results.push(v);
855 /// # }
856 /// # results.sort();
857 /// # assert_eq!(results, (0..=16).collect::<Vec<_>>());
858 /// # assert_eq!(locations.len(), 16);
859 /// # }));
860 /// # }
861 /// ```
862 pub fn round_robin_bincode<L2: 'a>(
863 self,
864 other: &Cluster<'a, L2>,
865 nondet_membership: NonDet,
866 ) -> KeyedStream<MemberId<L>, T, Cluster<'a, L2>, Unbounded, TotalOrder, ExactlyOnce>
867 where
868 T: Serialize + DeserializeOwned,
869 {
870 self.round_robin(other, TCP.fail_stop().bincode(), nondet_membership)
871 }
872
873 /// Distributes elements of this stream to cluster members in a round-robin fashion, using
874 /// the configuration in `via` to set up the message transport.
875 ///
876 /// This provides load balancing by evenly distributing work across cluster members. The
877 /// distribution is deterministic based on element order - the first element goes to member 0,
878 /// the second to member 1, and so on, wrapping around when reaching the end of the member list.
879 ///
880 /// # Non-Determinism
881 /// The set of cluster members may asynchronously change over time. Each element is distributed
882 /// based on the current cluster membership _at that point in time_. Depending on when cluster
883 /// members join and leave, the round-robin pattern will change. Furthermore, even when the
884 /// membership is stable, the order of members in the round-robin pattern may change across runs.
885 ///
886 /// # Ordering Requirements
887 /// This method is only available on streams with [`TotalOrder`] and [`ExactlyOnce`], since the
888 /// order of messages and retries affects the round-robin pattern.
889 ///
890 /// # Example
891 /// ```rust
892 /// # #[cfg(feature = "deploy")] {
893 /// # use hydro_lang::prelude::*;
894 /// # use hydro_lang::live_collections::stream::{TotalOrder, ExactlyOnce, NoOrder};
895 /// # use hydro_lang::location::MemberId;
896 /// # use futures::StreamExt;
897 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
898 /// let p1 = flow.process::<()>();
899 /// let workers1: Cluster<()> = flow.cluster::<()>();
900 /// let workers2: Cluster<()> = flow.cluster::<()>();
901 /// let numbers: Stream<_, Process<_>, _, TotalOrder, ExactlyOnce> = p1.source_iter(q!(0..=16));
902 /// let on_worker1: Stream<_, Cluster<_>, _> = numbers.round_robin(&workers1, TCP.fail_stop().bincode(), nondet!(/** assuming stable membership */));
903 /// let on_worker2: Stream<_, Cluster<_>, _> = on_worker1.round_robin(&workers2, TCP.fail_stop().bincode(), nondet!(/** assuming stable membership */)).entries().assume_ordering(nondet!(/** assuming stable membership */));
904 /// on_worker2.send(&p2, TCP.fail_stop().bincode())
905 /// # .entries()
906 /// # .map(q!(|(w2, (w1, v))| ((w2, w1), v)))
907 /// # }, |mut stream| async move {
908 /// # let mut results = Vec::new();
909 /// # let mut locations = std::collections::HashSet::new();
910 /// # for w in 0..=16 {
911 /// # let (location, v) = stream.next().await.unwrap();
912 /// # locations.insert(location);
913 /// # results.push(v);
914 /// # }
915 /// # results.sort();
916 /// # assert_eq!(results, (0..=16).collect::<Vec<_>>());
917 /// # assert_eq!(locations.len(), 16);
918 /// # }));
919 /// # }
920 /// ```
921 pub fn round_robin<L2: 'a, N: NetworkFor<T>>(
922 self,
923 to: &Cluster<'a, L2>,
924 via: N,
925 nondet_membership: NonDet,
926 ) -> KeyedStream<MemberId<L>, T, Cluster<'a, L2>, Unbounded, N::OrderingGuarantee, ExactlyOnce>
927 {
928 // TODO(#1875): the membership snapshot below is over a `KeyedSingleton`, and keyed
929 // sim hooks do not exist yet. Once they do, expose a composite hook payload here
930 // (`NonDet<(Option<KeyedSnapshotHook<..>>, Option<BatchHook<T, O, R>>)>`) so tests
931 // can script the membership snapshot and the element batching independently.
932 let ids = track_membership(self.location.source_cluster_membership_stream(
933 to,
934 nondet!(/** dropped prefixes don't affect broadcast */),
935 ));
936 sliced! {
937 let members_snapshot = use::snapshot(ids, nondet!(
938 /// membership timing is captured by the caller's guard
939 nondet_membership
940 ));
941 let elements = use::batch(self.enumerate(), nondet!(
942 /// batching timing is captured by the caller's guard
943 nondet_membership
944 ));
945
946 let current_members = members_snapshot
947 .filter(q!(|b| *b))
948 .keys()
949 .assume_ordering::<TotalOrder>(nondet!(/** membership timing is captured by the caller guard */ nondet_membership))
950 .collect_vec();
951
952 elements
953 .cross_singleton(current_members)
954 .filter_map(q!(|(data, members)| {
955 if members.is_empty() {
956 None
957 } else {
958 Some((members[data.0 % members.len()].clone(), data.1))
959 }
960 }))
961 }
962 .demux(to, via)
963 }
964}
965
966impl<'a, T, L, B: Boundedness, C: Consistency, O: Ordering, R: Retries>
967 Stream<T, Cluster<'a, L, C>, B, O, R>
968{
969 #[deprecated = "use Stream::send(..., TCP.fail_stop().bincode()) instead"]
970 /// "Moves" elements of this stream from a cluster to a process by sending them over the network,
971 /// using [`bincode`] to serialize/deserialize messages.
972 ///
973 /// Each cluster member sends its local stream elements, and they are collected at the destination
974 /// as a [`KeyedStream`] where keys identify the source cluster member.
975 ///
976 /// # Example
977 /// ```rust
978 /// # #[cfg(feature = "deploy")] {
979 /// # use hydro_lang::prelude::*;
980 /// # use futures::StreamExt;
981 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, process| {
982 /// let workers: Cluster<()> = flow.cluster::<()>();
983 /// let numbers: Stream<_, Cluster<_>, _> = workers.source_iter(q!(vec![1]));
984 /// let all_received = numbers.send_bincode(&process); // KeyedStream<MemberId<()>, i32, ...>
985 /// # all_received.entries()
986 /// # }, |mut stream| async move {
987 /// // if there are 4 members in the cluster, we should receive 4 elements
988 /// // { MemberId::<()>(0): [1], MemberId::<()>(1): [1], MemberId::<()>(2): [1], MemberId::<()>(3): [1] }
989 /// # let mut results = Vec::new();
990 /// # for w in 0..4 {
991 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
992 /// # }
993 /// # results.sort();
994 /// # assert_eq!(results, vec!["(MemberId::<()>(0), 1)", "(MemberId::<()>(1), 1)", "(MemberId::<()>(2), 1)", "(MemberId::<()>(3), 1)"]);
995 /// # }));
996 /// # }
997 /// ```
998 ///
999 /// If you don't need to know the source for each element, you can use `.values()`
1000 /// to get just the data:
1001 /// ```rust
1002 /// # #[cfg(feature = "deploy")] {
1003 /// # use hydro_lang::prelude::*;
1004 /// # use hydro_lang::live_collections::stream::NoOrder;
1005 /// # use futures::StreamExt;
1006 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, process| {
1007 /// # let workers: Cluster<()> = flow.cluster::<()>();
1008 /// # let numbers: Stream<_, Cluster<_>, _> = workers.source_iter(q!(vec![1]));
1009 /// let values: Stream<i32, _, _, NoOrder> = numbers.send_bincode(&process).values();
1010 /// # values
1011 /// # }, |mut stream| async move {
1012 /// # let mut results = Vec::new();
1013 /// # for w in 0..4 {
1014 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
1015 /// # }
1016 /// # results.sort();
1017 /// // if there are 4 members in the cluster, we should receive 4 elements
1018 /// // 1, 1, 1, 1
1019 /// # assert_eq!(results, vec!["1", "1", "1", "1"]);
1020 /// # }));
1021 /// # }
1022 /// ```
1023 pub fn send_bincode<L2>(
1024 self,
1025 other: &Process<'a, L2>,
1026 ) -> KeyedStream<MemberId<L>, T, Process<'a, L2>, Unbounded, O, R>
1027 where
1028 T: Serialize + DeserializeOwned,
1029 {
1030 self.send(other, TCP.fail_stop().bincode())
1031 }
1032
1033 /// "Moves" elements of this stream from a cluster to a process by sending them over the network,
1034 /// using the configuration in `via` to set up the message transport.
1035 ///
1036 /// Each cluster member sends its local stream elements, and they are collected at the destination
1037 /// as a [`KeyedStream`] where keys identify the source cluster member.
1038 ///
1039 /// # Example
1040 /// ```rust
1041 /// # #[cfg(feature = "deploy")] {
1042 /// # use hydro_lang::prelude::*;
1043 /// # use futures::StreamExt;
1044 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, process| {
1045 /// let workers: Cluster<()> = flow.cluster::<()>();
1046 /// let numbers: Stream<_, Cluster<_>, _> = workers.source_iter(q!(vec![1]));
1047 /// let all_received = numbers.send(&process, TCP.fail_stop().bincode()); // KeyedStream<MemberId<()>, i32, ...>
1048 /// # all_received.entries()
1049 /// # }, |mut stream| async move {
1050 /// // if there are 4 members in the cluster, we should receive 4 elements
1051 /// // { MemberId::<()>(0): [1], MemberId::<()>(1): [1], MemberId::<()>(2): [1], MemberId::<()>(3): [1] }
1052 /// # let mut results = Vec::new();
1053 /// # for w in 0..4 {
1054 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
1055 /// # }
1056 /// # results.sort();
1057 /// # assert_eq!(results, vec!["(MemberId::<()>(0), 1)", "(MemberId::<()>(1), 1)", "(MemberId::<()>(2), 1)", "(MemberId::<()>(3), 1)"]);
1058 /// # }));
1059 /// # }
1060 /// ```
1061 ///
1062 /// If you don't need to know the source for each element, you can use `.values()`
1063 /// to get just the data:
1064 /// ```rust
1065 /// # #[cfg(feature = "deploy")] {
1066 /// # use hydro_lang::prelude::*;
1067 /// # use hydro_lang::live_collections::stream::NoOrder;
1068 /// # use futures::StreamExt;
1069 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, process| {
1070 /// # let workers: Cluster<()> = flow.cluster::<()>();
1071 /// # let numbers: Stream<_, Cluster<_>, _> = workers.source_iter(q!(vec![1]));
1072 /// let values: Stream<i32, _, _, NoOrder> =
1073 /// numbers.send(&process, TCP.fail_stop().bincode()).values();
1074 /// # values
1075 /// # }, |mut stream| async move {
1076 /// # let mut results = Vec::new();
1077 /// # for w in 0..4 {
1078 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
1079 /// # }
1080 /// # results.sort();
1081 /// // if there are 4 members in the cluster, we should receive 4 elements
1082 /// // 1, 1, 1, 1
1083 /// # assert_eq!(results, vec!["1", "1", "1", "1"]);
1084 /// # }));
1085 /// # }
1086 /// ```
1087 pub fn send<L2, N: NetworkFor<T>>(
1088 self,
1089 to: &Process<'a, L2>,
1090 via: N,
1091 ) -> KeyedStream<
1092 MemberId<L>,
1093 T,
1094 Process<'a, L2>,
1095 Unbounded,
1096 <O as MinOrder<N::OrderingGuarantee>>::Min,
1097 R,
1098 >
1099 where
1100 O: MinOrder<N::OrderingGuarantee>,
1101 {
1102 let name = via.name();
1103 if to.multiversioned() && name.is_none() {
1104 panic!(
1105 "Cannot send to a multiversioned location without a channel name. Please provide a name for the network."
1106 );
1107 }
1108
1109 let (serialize, deserialize) = if N::is_embedded() {
1110 (
1111 NetworkSend::Embedded {
1112 tag: None,
1113 element_type: quote_type::<T>().into(),
1114 },
1115 NetworkRecv::Embedded {
1116 tag: Some(quote_type::<L>().into()),
1117 element_type: quote_type::<T>().into(),
1118 },
1119 )
1120 } else {
1121 (
1122 NetworkSend::Custom {
1123 serialize_fn: Some(N::serialize_thunk(false).into()),
1124 },
1125 NetworkRecv::Custom {
1126 deserialize_fn: Some(N::deserialize_thunk(Some("e_type::<L>())).into()),
1127 },
1128 )
1129 };
1130
1131 let raw_stream: Stream<
1132 (MemberId<L>, T),
1133 Process<'a, L2>,
1134 Unbounded,
1135 <O as MinOrder<N::OrderingGuarantee>>::Min,
1136 R,
1137 > = Stream::new(
1138 to.clone(),
1139 HydroNode::Network {
1140 name: name.map(ToOwned::to_owned),
1141 networking_info: N::networking_info(),
1142 serialize,
1143 deserialize,
1144 instantiate_fn: DebugInstantiate::Building,
1145 input: Box::new(self.ir_node.replace(HydroNode::Placeholder)),
1146 metadata: to.new_node_metadata(Stream::<
1147 (MemberId<L>, T),
1148 Process<'a, L2>,
1149 Unbounded,
1150 <O as MinOrder<N::OrderingGuarantee>>::Min,
1151 R,
1152 >::collection_kind()),
1153 },
1154 );
1155
1156 raw_stream.into_keyed()
1157 }
1158
1159 #[deprecated = "use Stream::broadcast(..., TCP.fail_stop().bincode()) instead"]
1160 /// Broadcasts elements of this stream at each source member to all members of a destination
1161 /// cluster, using [`bincode`] to serialize/deserialize messages.
1162 ///
1163 /// Each source member sends each of its stream elements to **every** member of the cluster
1164 /// based on its latest membership information. Unlike [`Stream::demux_bincode`], which requires
1165 /// `(MemberId, T)` tuples to target specific members, `broadcast_bincode` takes a stream of
1166 /// **only data elements** and sends each element to all cluster members.
1167 ///
1168 /// # Non-Determinism
1169 /// The set of cluster members may asynchronously change over time. Each element is only broadcast
1170 /// to the current cluster members known _at that point in time_ at the source member. Depending
1171 /// on when each source member is notified of membership changes, it will broadcast each element
1172 /// to different members.
1173 ///
1174 /// # Example
1175 /// ```rust
1176 /// # #[cfg(feature = "deploy")] {
1177 /// # use hydro_lang::prelude::*;
1178 /// # use hydro_lang::location::MemberId;
1179 /// # use futures::StreamExt;
1180 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
1181 /// # type Source = ();
1182 /// # type Destination = ();
1183 /// let source: Cluster<Source> = flow.cluster::<Source>();
1184 /// let numbers: Stream<_, Cluster<Source>, _> = source.source_iter(q!(vec![123]));
1185 /// let destination: Cluster<Destination> = flow.cluster::<Destination>();
1186 /// let on_destination: KeyedStream<MemberId<Source>, _, Cluster<Destination>, _> = numbers.broadcast_bincode(&destination, nondet!(/** assuming stable membership */));
1187 /// # on_destination.entries().send_bincode(&p2).entries()
1188 /// // if there are 4 members in the desination, each receives one element from each source member
1189 /// // - Destination(0): { Source(0): [123], Source(1): [123], ... }
1190 /// // - Destination(1): { Source(0): [123], Source(1): [123], ... }
1191 /// // - ...
1192 /// # }, |mut stream| async move {
1193 /// # let mut results = Vec::new();
1194 /// # for w in 0..16 {
1195 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
1196 /// # }
1197 /// # results.sort();
1198 /// # assert_eq!(results, vec![
1199 /// # "(MemberId::<()>(0), (MemberId::<()>(0), 123))", "(MemberId::<()>(0), (MemberId::<()>(1), 123))", "(MemberId::<()>(0), (MemberId::<()>(2), 123))", "(MemberId::<()>(0), (MemberId::<()>(3), 123))",
1200 /// # "(MemberId::<()>(1), (MemberId::<()>(0), 123))", "(MemberId::<()>(1), (MemberId::<()>(1), 123))", "(MemberId::<()>(1), (MemberId::<()>(2), 123))", "(MemberId::<()>(1), (MemberId::<()>(3), 123))",
1201 /// # "(MemberId::<()>(2), (MemberId::<()>(0), 123))", "(MemberId::<()>(2), (MemberId::<()>(1), 123))", "(MemberId::<()>(2), (MemberId::<()>(2), 123))", "(MemberId::<()>(2), (MemberId::<()>(3), 123))",
1202 /// # "(MemberId::<()>(3), (MemberId::<()>(0), 123))", "(MemberId::<()>(3), (MemberId::<()>(1), 123))", "(MemberId::<()>(3), (MemberId::<()>(2), 123))", "(MemberId::<()>(3), (MemberId::<()>(3), 123))"
1203 /// # ]);
1204 /// # }));
1205 /// # }
1206 /// ```
1207 pub fn broadcast_bincode<L2: 'a>(
1208 self,
1209 other: &Cluster<'a, L2>,
1210 nondet_membership: NonDet,
1211 ) -> KeyedStream<MemberId<L>, T, Cluster<'a, L2>, Unbounded, O, R>
1212 where
1213 T: Clone + Serialize + DeserializeOwned,
1214 {
1215 self.broadcast(other, TCP.fail_stop().bincode(), nondet_membership)
1216 }
1217
1218 /// Broadcasts elements of this stream at each source member to all members of a destination
1219 /// cluster, using the configuration in `via` to set up the message transport.
1220 ///
1221 /// Each source member sends each of its stream elements to **every** member of the cluster
1222 /// based on its latest membership information. Unlike [`Stream::demux`], which requires
1223 /// `(MemberId, T)` tuples to target specific members, `broadcast` takes a stream of
1224 /// **only data elements** and sends each element to all cluster members.
1225 ///
1226 /// # Non-Determinism
1227 /// The set of cluster members may asynchronously change over time. Each element is only broadcast
1228 /// to the current cluster members known _at that point in time_ at the source member. Depending
1229 /// on when each source member is notified of membership changes, it will broadcast each element
1230 /// to different members.
1231 ///
1232 /// # Example
1233 /// ```rust
1234 /// # #[cfg(feature = "deploy")] {
1235 /// # use hydro_lang::prelude::*;
1236 /// # use hydro_lang::location::MemberId;
1237 /// # use futures::StreamExt;
1238 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
1239 /// # type Source = ();
1240 /// # type Destination = ();
1241 /// let source: Cluster<Source> = flow.cluster::<Source>();
1242 /// let numbers: Stream<_, Cluster<Source>, _> = source.source_iter(q!(vec![123]));
1243 /// let destination: Cluster<Destination> = flow.cluster::<Destination>();
1244 /// let on_destination: KeyedStream<MemberId<Source>, _, Cluster<Destination>, _> = numbers.broadcast(&destination, TCP.fail_stop().bincode(), nondet!(/** assuming stable membership */));
1245 /// # on_destination.entries().send(&p2, TCP.fail_stop().bincode()).entries()
1246 /// // if there are 4 members in the desination, each receives one element from each source member
1247 /// // - Destination(0): { Source(0): [123], Source(1): [123], ... }
1248 /// // - Destination(1): { Source(0): [123], Source(1): [123], ... }
1249 /// // - ...
1250 /// # }, |mut stream| async move {
1251 /// # let mut results = Vec::new();
1252 /// # for w in 0..16 {
1253 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
1254 /// # }
1255 /// # results.sort();
1256 /// # assert_eq!(results, vec![
1257 /// # "(MemberId::<()>(0), (MemberId::<()>(0), 123))", "(MemberId::<()>(0), (MemberId::<()>(1), 123))", "(MemberId::<()>(0), (MemberId::<()>(2), 123))", "(MemberId::<()>(0), (MemberId::<()>(3), 123))",
1258 /// # "(MemberId::<()>(1), (MemberId::<()>(0), 123))", "(MemberId::<()>(1), (MemberId::<()>(1), 123))", "(MemberId::<()>(1), (MemberId::<()>(2), 123))", "(MemberId::<()>(1), (MemberId::<()>(3), 123))",
1259 /// # "(MemberId::<()>(2), (MemberId::<()>(0), 123))", "(MemberId::<()>(2), (MemberId::<()>(1), 123))", "(MemberId::<()>(2), (MemberId::<()>(2), 123))", "(MemberId::<()>(2), (MemberId::<()>(3), 123))",
1260 /// # "(MemberId::<()>(3), (MemberId::<()>(0), 123))", "(MemberId::<()>(3), (MemberId::<()>(1), 123))", "(MemberId::<()>(3), (MemberId::<()>(2), 123))", "(MemberId::<()>(3), (MemberId::<()>(3), 123))"
1261 /// # ]);
1262 /// # }));
1263 /// # }
1264 /// ```
1265 pub fn broadcast<L2: 'a, N: NetworkFor<T>>(
1266 self,
1267 to: &Cluster<'a, L2>,
1268 via: N,
1269 nondet_membership: NonDet,
1270 ) -> KeyedStream<
1271 MemberId<L>,
1272 T,
1273 Cluster<'a, L2>,
1274 Unbounded,
1275 <O as MinOrder<N::OrderingGuarantee>>::Min,
1276 R,
1277 >
1278 where
1279 T: Clone,
1280 O: MinOrder<N::OrderingGuarantee>,
1281 {
1282 // TODO(#1875): the membership snapshot below is over a `KeyedSingleton`, and keyed
1283 // sim hooks do not exist yet. Once they do, expose a composite hook payload here
1284 // (`NonDet<(Option<KeyedSnapshotHook<..>>, Option<BatchHook<T, O, R>>)>`) so tests
1285 // can script the membership snapshot and the element batching independently.
1286 let ids = track_membership(self.location.source_cluster_membership_stream(
1287 to,
1288 nondet!(/** dropped prefixes don't affect broadcast */),
1289 ));
1290 sliced! {
1291 let members_snapshot = use::snapshot(ids, nondet!(
1292 /// membership timing is captured by the caller's guard
1293 nondet_membership
1294 ));
1295 let elements = use::batch(self, nondet!(
1296 /// batching timing is captured by the caller's guard
1297 nondet_membership
1298 ));
1299
1300 let current_members = members_snapshot.filter(q!(|b| *b));
1301 elements.repeat_with_keys(current_members)
1302 }
1303 .demux(to, via)
1304 }
1305
1306 /// Broadcasts elements of this stream at each source member to all members of a destination
1307 /// cluster, assuming membership is closed (fixed at deploy time).
1308 ///
1309 /// Unlike [`Stream::broadcast`], this does not require a [`NonDet`] guard.
1310 /// The membership set is obtained from deploy metadata via [`ClusterIds`], making the
1311 /// broadcast fully deterministic.
1312 ///
1313 /// The consistency guarantee of the output depends on the network's failure policy
1314 /// ([`NetworkFor::ConsistencyGuarantee`]). Policies like `fail_stop` and
1315 /// `lossy_delayed_forever` guarantee that every live destination member eventually
1316 /// materializes the same elements from each source, so the output is
1317 /// [`EventualConsistency`](crate::location::cluster::EventualConsistency). A plain `lossy`
1318 /// policy can drop individual messages for some
1319 /// members while delivering them to others, so replicas may permanently diverge and the
1320 /// output only has [`NoConsistency`].
1321 ///
1322 /// This is only available in deployment targets with static cluster membership
1323 /// (legacy Hydro Deploy and simulation). On dynamic targets, use [`Stream::broadcast`].
1324 pub fn broadcast_closed<L2: 'a, N: NetworkFor<T>>(
1325 self,
1326 to: &Cluster<'a, L2>,
1327 via: N,
1328 ) -> KeyedStream<
1329 MemberId<L>,
1330 T,
1331 Cluster<'a, L2, N::ConsistencyGuarantee>,
1332 Unbounded,
1333 <O as MinOrder<N::OrderingGuarantee>>::Min,
1334 R,
1335 >
1336 where
1337 T: Clone,
1338 O: MinOrder<N::OrderingGuarantee>,
1339 {
1340 let cluster_ids = ClusterIds {
1341 key: to.key,
1342 _phantom: PhantomData,
1343 };
1344 let member_ids = self
1345 .location
1346 .source_iter(q!(cluster_ids
1347 .iter()
1348 .map(|id| MemberId::from_tagless(id.clone()))))
1349 .assert_has_consistency_of_trusted::<Cluster<'a, L, C>>(manual_proof!(
1350 /// ClusterIds is deploy-time metadata, identical on every cluster member.
1351 ));
1352
1353 self.cross_product(member_ids)
1354 .map(q!(|(data, member_id)| (member_id, data)))
1355 .into_keyed()
1356 .demux(to, via)
1357 .assert_has_consistency_of_trusted(manual_proof!(
1358 /// Closed broadcast with fixed membership: every source member sends to every
1359 /// destination member, and the network's failure policy (tracked by
1360 /// `NetworkFor::ConsistencyGuarantee`) delivers the same messages to every live
1361 /// member, so all destinations materialize the same elements.
1362 ))
1363 }
1364
1365 #[cfg(feature = "sim")]
1366 /// Sends elements of this cluster stream to an external location using bincode serialization.
1367 fn send_bincode_external<L2>(self, other: &External<'_, L2>) -> ExternalBincodeStream<T, O, R>
1368 where
1369 T: Serialize + DeserializeOwned,
1370 {
1371 let serialize_pipeline = Some(serialize_bincode::<T>(false));
1372
1373 let mut flow_state_borrow = self.location.flow_state().borrow_mut();
1374
1375 let external_port_id = flow_state_borrow.next_external_port();
1376
1377 flow_state_borrow.push_root(HydroRoot::SendExternal {
1378 to_external_key: other.key,
1379 to_port_id: external_port_id,
1380 to_many: false,
1381 unpaired: true,
1382 serialize_fn: serialize_pipeline.map(|e| e.into()),
1383 instantiate_fn: DebugInstantiate::Building,
1384 input: Box::new(self.ir_node.replace(HydroNode::Placeholder)),
1385 op_metadata: HydroIrOpMetadata::new(),
1386 });
1387
1388 ExternalBincodeStream {
1389 process_key: other.key,
1390 port_id: external_port_id,
1391 _phantom: PhantomData,
1392 }
1393 }
1394
1395 #[cfg(feature = "sim")]
1396 /// Sets up a simulation output port for this cluster stream, allowing test code
1397 /// to receive `(member_id, T)` pairs during simulation.
1398 pub fn sim_cluster_output(self) -> crate::sim::SimClusterReceiver<T, O, R>
1399 where
1400 T: Serialize + DeserializeOwned,
1401 {
1402 let external_location: External<'a, ()> = External {
1403 key: LocationKey::FIRST,
1404 flow_state: self.location.flow_state().clone(),
1405 _phantom: PhantomData,
1406 };
1407
1408 let external = self.send_bincode_external(&external_location);
1409
1410 crate::sim::SimClusterReceiver(external.port_id, PhantomData)
1411 }
1412}
1413
1414impl<'a, T, L, L2, B: Boundedness, C: Consistency, O: Ordering, R: Retries>
1415 Stream<(MemberId<L2>, T), Cluster<'a, L, C>, B, O, R>
1416{
1417 #[deprecated = "use Stream::demux(..., TCP.fail_stop().bincode()) instead"]
1418 /// Sends elements of this stream at each source member to specific members of a destination
1419 /// cluster, identified by a [`MemberId`], using [`bincode`] to serialize/deserialize messages.
1420 ///
1421 /// Each element in the stream must be a tuple `(MemberId<L2>, T)` where the first element
1422 /// specifies which cluster member should receive the data. Unlike [`Stream::broadcast_bincode`],
1423 /// this API allows precise targeting of specific cluster members rather than broadcasting to
1424 /// all members.
1425 ///
1426 /// Each cluster member sends its local stream elements, and they are collected at each
1427 /// destination member as a [`KeyedStream`] where keys identify the source cluster member.
1428 ///
1429 /// # Example
1430 /// ```rust
1431 /// # #[cfg(feature = "deploy")] {
1432 /// # use hydro_lang::prelude::*;
1433 /// # use futures::StreamExt;
1434 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
1435 /// # type Source = ();
1436 /// # type Destination = ();
1437 /// let source: Cluster<Source> = flow.cluster::<Source>();
1438 /// let to_send: Stream<_, Cluster<_>, _> = source
1439 /// .source_iter(q!(vec![0, 1, 2, 3]))
1440 /// .map(q!(|x| (hydro_lang::location::MemberId::from_raw_id(x), x)));
1441 /// let destination: Cluster<Destination> = flow.cluster::<Destination>();
1442 /// let all_received = to_send.demux_bincode(&destination); // KeyedStream<MemberId<Source>, i32, ...>
1443 /// # all_received.entries().send_bincode(&p2).entries()
1444 /// # }, |mut stream| async move {
1445 /// // if there are 4 members in the destination cluster, each receives one message from each source member
1446 /// // - Destination(0): { Source(0): [0], Source(1): [0], ... }
1447 /// // - Destination(1): { Source(0): [1], Source(1): [1], ... }
1448 /// // - ...
1449 /// # let mut results = Vec::new();
1450 /// # for w in 0..16 {
1451 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
1452 /// # }
1453 /// # results.sort();
1454 /// # assert_eq!(results, vec![
1455 /// # "(MemberId::<()>(0), (MemberId::<()>(0), 0))", "(MemberId::<()>(0), (MemberId::<()>(1), 0))", "(MemberId::<()>(0), (MemberId::<()>(2), 0))", "(MemberId::<()>(0), (MemberId::<()>(3), 0))",
1456 /// # "(MemberId::<()>(1), (MemberId::<()>(0), 1))", "(MemberId::<()>(1), (MemberId::<()>(1), 1))", "(MemberId::<()>(1), (MemberId::<()>(2), 1))", "(MemberId::<()>(1), (MemberId::<()>(3), 1))",
1457 /// # "(MemberId::<()>(2), (MemberId::<()>(0), 2))", "(MemberId::<()>(2), (MemberId::<()>(1), 2))", "(MemberId::<()>(2), (MemberId::<()>(2), 2))", "(MemberId::<()>(2), (MemberId::<()>(3), 2))",
1458 /// # "(MemberId::<()>(3), (MemberId::<()>(0), 3))", "(MemberId::<()>(3), (MemberId::<()>(1), 3))", "(MemberId::<()>(3), (MemberId::<()>(2), 3))", "(MemberId::<()>(3), (MemberId::<()>(3), 3))"
1459 /// # ]);
1460 /// # }));
1461 /// # }
1462 /// ```
1463 pub fn demux_bincode(
1464 self,
1465 other: &Cluster<'a, L2>,
1466 ) -> KeyedStream<MemberId<L>, T, Cluster<'a, L2>, Unbounded, O, R>
1467 where
1468 T: Serialize + DeserializeOwned,
1469 {
1470 self.demux(other, TCP.fail_stop().bincode())
1471 }
1472
1473 /// Sends elements of this stream at each source member to specific members of a destination
1474 /// cluster, identified by a [`MemberId`], using the configuration in `via` to set up the
1475 /// message transport.
1476 ///
1477 /// Each element in the stream must be a tuple `(MemberId<L2>, T)` where the first element
1478 /// specifies which cluster member should receive the data. Unlike [`Stream::broadcast`],
1479 /// this API allows precise targeting of specific cluster members rather than broadcasting to
1480 /// all members.
1481 ///
1482 /// Each cluster member sends its local stream elements, and they are collected at each
1483 /// destination member as a [`KeyedStream`] where keys identify the source cluster member.
1484 ///
1485 /// # Example
1486 /// ```rust
1487 /// # #[cfg(feature = "deploy")] {
1488 /// # use hydro_lang::prelude::*;
1489 /// # use futures::StreamExt;
1490 /// # tokio_test::block_on(hydro_lang::test_util::multi_location_test(|flow, p2| {
1491 /// # type Source = ();
1492 /// # type Destination = ();
1493 /// let source: Cluster<Source> = flow.cluster::<Source>();
1494 /// let to_send: Stream<_, Cluster<_>, _> = source
1495 /// .source_iter(q!(vec![0, 1, 2, 3]))
1496 /// .map(q!(|x| (hydro_lang::location::MemberId::from_raw_id(x), x)));
1497 /// let destination: Cluster<Destination> = flow.cluster::<Destination>();
1498 /// let all_received = to_send.demux(&destination, TCP.fail_stop().bincode()); // KeyedStream<MemberId<Source>, i32, ...>
1499 /// # all_received.entries().send(&p2, TCP.fail_stop().bincode()).entries()
1500 /// # }, |mut stream| async move {
1501 /// // if there are 4 members in the destination cluster, each receives one message from each source member
1502 /// // - Destination(0): { Source(0): [0], Source(1): [0], ... }
1503 /// // - Destination(1): { Source(0): [1], Source(1): [1], ... }
1504 /// // - ...
1505 /// # let mut results = Vec::new();
1506 /// # for w in 0..16 {
1507 /// # results.push(format!("{:?}", stream.next().await.unwrap()));
1508 /// # }
1509 /// # results.sort();
1510 /// # assert_eq!(results, vec![
1511 /// # "(MemberId::<()>(0), (MemberId::<()>(0), 0))", "(MemberId::<()>(0), (MemberId::<()>(1), 0))", "(MemberId::<()>(0), (MemberId::<()>(2), 0))", "(MemberId::<()>(0), (MemberId::<()>(3), 0))",
1512 /// # "(MemberId::<()>(1), (MemberId::<()>(0), 1))", "(MemberId::<()>(1), (MemberId::<()>(1), 1))", "(MemberId::<()>(1), (MemberId::<()>(2), 1))", "(MemberId::<()>(1), (MemberId::<()>(3), 1))",
1513 /// # "(MemberId::<()>(2), (MemberId::<()>(0), 2))", "(MemberId::<()>(2), (MemberId::<()>(1), 2))", "(MemberId::<()>(2), (MemberId::<()>(2), 2))", "(MemberId::<()>(2), (MemberId::<()>(3), 2))",
1514 /// # "(MemberId::<()>(3), (MemberId::<()>(0), 3))", "(MemberId::<()>(3), (MemberId::<()>(1), 3))", "(MemberId::<()>(3), (MemberId::<()>(2), 3))", "(MemberId::<()>(3), (MemberId::<()>(3), 3))"
1515 /// # ]);
1516 /// # }));
1517 /// # }
1518 /// ```
1519 pub fn demux<N: NetworkFor<T>>(
1520 self,
1521 to: &Cluster<'a, L2>,
1522 via: N,
1523 ) -> KeyedStream<
1524 MemberId<L>,
1525 T,
1526 Cluster<'a, L2, NoConsistency>,
1527 Unbounded,
1528 <O as MinOrder<N::OrderingGuarantee>>::Min,
1529 R,
1530 >
1531 where
1532 O: MinOrder<N::OrderingGuarantee>,
1533 {
1534 self.into_keyed().demux(to, via)
1535 }
1536}
1537
1538#[cfg(test)]
1539mod tests {
1540 #[cfg(feature = "sim")]
1541 use stageleft::q;
1542
1543 #[cfg(feature = "sim")]
1544 use crate::live_collections::sliced::sliced;
1545 #[cfg(feature = "sim")]
1546 use crate::location::{Location, MemberId};
1547 #[cfg(feature = "sim")]
1548 use crate::networking::TCP;
1549 #[cfg(feature = "sim")]
1550 use crate::nondet::nondet;
1551 #[cfg(feature = "sim")]
1552 use crate::prelude::FlowBuilder;
1553
1554 #[cfg(feature = "sim")]
1555 #[test]
1556 fn sim_send_bincode_o2o() {
1557 use crate::networking::TCP;
1558
1559 let mut flow = FlowBuilder::new();
1560 let node = flow.process::<()>();
1561 let node2 = flow.process::<()>();
1562
1563 let (in_send, input) = node.sim_input();
1564
1565 let out_recv = input
1566 .send(&node2, TCP.fail_stop().bincode())
1567 .batch(&node2.tick(), nondet!(/** test */))
1568 .count()
1569 .all_ticks()
1570 .sim_output();
1571
1572 let instances = flow.sim().exhaustive(async || {
1573 in_send.send(());
1574 in_send.send(());
1575 in_send.send(());
1576
1577 let received = out_recv.collect::<Vec<_>>().await;
1578 assert!(received.into_iter().sum::<usize>() == 3);
1579 });
1580
1581 assert_eq!(instances, 4); // 2^{3 - 1}
1582 }
1583
1584 #[cfg(feature = "sim")]
1585 #[test]
1586 fn sim_send_bincode_m2o() {
1587 let mut flow = FlowBuilder::new();
1588 let cluster = flow.cluster::<()>();
1589 let node = flow.process::<()>();
1590
1591 let input = cluster.source_iter(q!(vec![1]));
1592
1593 let out_recv = input
1594 .send(&node, TCP.fail_stop().bincode())
1595 .entries()
1596 .batch(&node.tick(), nondet!(/** test */))
1597 .all_ticks()
1598 .sim_output();
1599
1600 let instances = flow
1601 .sim()
1602 .with_cluster_size(&cluster, 4)
1603 .exhaustive(async || {
1604 out_recv
1605 .assert_yields_only_unordered(vec![
1606 (MemberId::from_raw_id(0), 1),
1607 (MemberId::from_raw_id(1), 1),
1608 (MemberId::from_raw_id(2), 1),
1609 (MemberId::from_raw_id(3), 1),
1610 ])
1611 .await
1612 });
1613
1614 assert_eq!(instances, 75); // ∑ (k=1 to 4) S(4,k) × k! = 75
1615 }
1616
1617 #[cfg(feature = "sim")]
1618 #[test]
1619 fn sim_send_bincode_multiple_m2o() {
1620 let mut flow = FlowBuilder::new();
1621 let cluster1 = flow.cluster::<()>();
1622 let cluster2 = flow.cluster::<()>();
1623 let node = flow.process::<()>();
1624
1625 let out_recv_1 = cluster1
1626 .source_iter(q!(vec![1]))
1627 .send(&node, TCP.fail_stop().bincode())
1628 .entries()
1629 .sim_output();
1630
1631 let out_recv_2 = cluster2
1632 .source_iter(q!(vec![2]))
1633 .send(&node, TCP.fail_stop().bincode())
1634 .entries()
1635 .sim_output();
1636
1637 let instances = flow
1638 .sim()
1639 .with_cluster_size(&cluster1, 3)
1640 .with_cluster_size(&cluster2, 4)
1641 .exhaustive(async || {
1642 out_recv_1
1643 .assert_yields_only_unordered(vec![
1644 (MemberId::from_raw_id(0), 1),
1645 (MemberId::from_raw_id(1), 1),
1646 (MemberId::from_raw_id(2), 1),
1647 ])
1648 .await;
1649
1650 out_recv_2
1651 .assert_yields_only_unordered(vec![
1652 (MemberId::from_raw_id(0), 2),
1653 (MemberId::from_raw_id(1), 2),
1654 (MemberId::from_raw_id(2), 2),
1655 (MemberId::from_raw_id(3), 2),
1656 ])
1657 .await;
1658 });
1659
1660 assert_eq!(instances, 1);
1661 }
1662
1663 #[cfg(feature = "sim")]
1664 #[test]
1665 fn sim_send_bincode_o2m() {
1666 let mut flow = FlowBuilder::new();
1667 let cluster = flow.cluster::<()>();
1668 let node = flow.process::<()>();
1669
1670 let input = node.source_iter(q!(vec![
1671 (MemberId::from_raw_id(0), 123),
1672 (MemberId::from_raw_id(1), 456),
1673 ]));
1674
1675 let out_recv = input
1676 .demux(&cluster, TCP.fail_stop().bincode())
1677 .map(q!(|x| x + 1))
1678 .send(&node, TCP.fail_stop().bincode())
1679 .entries()
1680 .sim_output();
1681
1682 flow.sim()
1683 .with_cluster_size(&cluster, 4)
1684 .exhaustive(async || {
1685 out_recv
1686 .assert_yields_only_unordered(vec![
1687 (MemberId::from_raw_id(0), 124),
1688 (MemberId::from_raw_id(1), 457),
1689 ])
1690 .await
1691 });
1692 }
1693
1694 #[cfg(feature = "sim")]
1695 #[test]
1696 fn sim_broadcast_bincode_o2m() {
1697 let mut flow = FlowBuilder::new();
1698 let cluster = flow.cluster::<()>();
1699 let node = flow.process::<()>();
1700
1701 let input = node.source_iter(q!(vec![123, 456]));
1702
1703 let out_recv = input
1704 .broadcast(&cluster, TCP.fail_stop().bincode(), nondet!(/** test */))
1705 .map(q!(|x| x + 1))
1706 .send(&node, TCP.fail_stop().bincode())
1707 .entries()
1708 .sim_output();
1709
1710 let mut c_1_produced = false;
1711 let mut c_2_produced = false;
1712 let mut c_1_saw_457_but_not_124 = false;
1713
1714 flow.sim()
1715 .with_cluster_size(&cluster, 2)
1716 .exhaustive(async || {
1717 let all_out = out_recv.collect_sorted::<Vec<_>>().await;
1718
1719 // check that order is preserved
1720 if all_out.contains(&(MemberId::from_raw_id(0), 124)) {
1721 assert!(all_out.contains(&(MemberId::from_raw_id(0), 457)));
1722 c_1_produced = true;
1723 }
1724
1725 if all_out.contains(&(MemberId::from_raw_id(1), 124)) {
1726 assert!(all_out.contains(&(MemberId::from_raw_id(1), 457)));
1727 c_2_produced = true;
1728 }
1729
1730 if all_out.contains(&(MemberId::from_raw_id(0), 457))
1731 && !all_out.contains(&(MemberId::from_raw_id(0), 124))
1732 {
1733 c_1_saw_457_but_not_124 = true;
1734 }
1735 });
1736
1737 assert!(c_1_produced && c_2_produced); // in at least one execution each, the cluster member received both messages
1738
1739 // in at least one execution, the cluster member received 457 but not 124, this tests
1740 // that the simulator properly explores dynamic membership additions (a member that joins after 123 is broadcast)
1741 assert!(c_1_saw_457_but_not_124);
1742 }
1743
1744 #[cfg(feature = "sim")]
1745 #[test]
1746 fn sim_send_bincode_m2m() {
1747 let mut flow = FlowBuilder::new();
1748 let cluster = flow.cluster::<()>();
1749 let node = flow.process::<()>();
1750
1751 let input = node.source_iter(q!(vec![
1752 (MemberId::from_raw_id(0), 123),
1753 (MemberId::from_raw_id(1), 456),
1754 ]));
1755
1756 let out_recv = input
1757 .demux(&cluster, TCP.fail_stop().bincode())
1758 .map(q!(|x| x + 1))
1759 .flat_map_ordered(q!(|x| vec![
1760 (MemberId::from_raw_id(0), x),
1761 (MemberId::from_raw_id(1), x),
1762 ]))
1763 .demux(&cluster, TCP.fail_stop().bincode())
1764 .entries()
1765 .send(&node, TCP.fail_stop().bincode())
1766 .entries()
1767 .sim_output();
1768
1769 flow.sim()
1770 .with_cluster_size(&cluster, 4)
1771 .exhaustive(async || {
1772 out_recv
1773 .assert_yields_only_unordered(vec![
1774 (MemberId::from_raw_id(0), (MemberId::from_raw_id(0), 124)),
1775 (MemberId::from_raw_id(0), (MemberId::from_raw_id(1), 457)),
1776 (MemberId::from_raw_id(1), (MemberId::from_raw_id(0), 124)),
1777 (MemberId::from_raw_id(1), (MemberId::from_raw_id(1), 457)),
1778 ])
1779 .await
1780 });
1781 }
1782
1783 #[cfg(feature = "sim")]
1784 #[test]
1785 fn sim_lossy_delayed_forever_o2o() {
1786 use std::collections::HashSet;
1787
1788 use crate::properties::manual_proof;
1789
1790 let mut flow = FlowBuilder::new();
1791 let node = flow.process::<()>();
1792 let node2 = flow.process::<()>();
1793
1794 let received = node
1795 .source_iter(q!(0..3_u32))
1796 .send(&node2, TCP.lossy_delayed_forever().bincode())
1797 .fold(
1798 q!(|| std::collections::HashSet::<u32>::new()),
1799 q!(
1800 |set, v| {
1801 set.insert(v);
1802 },
1803 commutative = manual_proof!(/** set insert is commutative */)
1804 ),
1805 );
1806
1807 let out_recv = sliced! {
1808 let snapshot = use::snapshot(received, nondet!(/** test */));
1809 snapshot.into_stream()
1810 }
1811 .sim_output();
1812
1813 let mut saw_non_contiguous = false;
1814
1815 flow.sim().test_safety_only().exhaustive(async || {
1816 let snapshots = out_recv.collect::<Vec<HashSet<u32>>>().await;
1817
1818 // Check each individual snapshot for a non-contiguous subset.
1819 for set in &snapshots {
1820 #[expect(clippy::disallowed_methods, reason = "min / max are deterministic")]
1821 if set.len() >= 2 && set.len() < 3 {
1822 let min = *set.iter().min().unwrap();
1823 let max = *set.iter().max().unwrap();
1824 if set.len() < (max - min + 1) as usize {
1825 saw_non_contiguous = true;
1826 }
1827 }
1828 }
1829 });
1830
1831 assert!(
1832 saw_non_contiguous,
1833 "Expected at least one execution with a non-contiguous subset of inputs"
1834 );
1835 }
1836
1837 #[cfg(feature = "sim")]
1838 #[test]
1839 fn sim_udp_lossy_delayed_forever_o2o() {
1840 use std::collections::HashSet;
1841
1842 use crate::networking::UDP;
1843 use crate::properties::manual_proof;
1844
1845 let mut flow = FlowBuilder::new();
1846 let node = flow.process::<()>();
1847 let node2 = flow.process::<()>();
1848
1849 let received = node
1850 .source_iter(q!(0..3_u32))
1851 .send(&node2, UDP.lossy_delayed_forever().bincode())
1852 .fold(
1853 q!(|| std::collections::HashSet::<u32>::new()),
1854 q!(
1855 |set, v| {
1856 set.insert(v);
1857 },
1858 commutative = manual_proof!(/** set insert is commutative */)
1859 ),
1860 );
1861
1862 let out_recv = sliced! {
1863 let snapshot = use::snapshot(received, nondet!(/** test */));
1864 snapshot.into_stream()
1865 }
1866 .sim_output();
1867
1868 let mut saw_non_contiguous = false;
1869
1870 flow.sim().test_safety_only().exhaustive(async || {
1871 let snapshots = out_recv.collect::<Vec<HashSet<u32>>>().await;
1872
1873 // Check each individual snapshot for a non-contiguous subset.
1874 for set in &snapshots {
1875 #[expect(clippy::disallowed_methods, reason = "min / max are deterministic")]
1876 if set.len() >= 2 && set.len() < 3 {
1877 let min = *set.iter().min().unwrap();
1878 let max = *set.iter().max().unwrap();
1879 if set.len() < (max - min + 1) as usize {
1880 saw_non_contiguous = true;
1881 }
1882 }
1883 }
1884 });
1885
1886 assert!(
1887 saw_non_contiguous,
1888 "Expected at least one execution with a non-contiguous subset of inputs"
1889 );
1890 }
1891
1892 #[cfg(feature = "sim")]
1893 #[test]
1894 fn sim_broadcast_closed_o2m() {
1895 let mut flow = FlowBuilder::new();
1896 let cluster = flow.cluster::<()>();
1897 let node = flow.process::<()>();
1898
1899 let input = node.source_iter(q!(vec![123, 456]));
1900
1901 let out_recv = input
1902 .broadcast_closed(&cluster, TCP.fail_stop().bincode())
1903 .send(&node, TCP.fail_stop().bincode())
1904 .entries()
1905 .sim_output();
1906
1907 flow.sim()
1908 .with_cluster_size(&cluster, 2)
1909 .exhaustive(async || {
1910 out_recv
1911 .assert_yields_only_unordered(vec![
1912 (MemberId::from_raw_id(0), 123),
1913 (MemberId::from_raw_id(0), 456),
1914 (MemberId::from_raw_id(1), 123),
1915 (MemberId::from_raw_id(1), 456),
1916 ])
1917 .await
1918 });
1919 }
1920
1921 #[cfg(feature = "sim")]
1922 #[test]
1923 fn sim_broadcast_closed_m2m() {
1924 let mut flow = FlowBuilder::new();
1925 let source = flow.cluster::<()>();
1926 let dest: crate::location::Cluster<'_, ()> = flow.cluster::<()>();
1927 let node = flow.process::<()>();
1928
1929 let input = source.source_iter(q!(vec![123]));
1930
1931 // Broadcast from source cluster to dest cluster, then collect at a process.
1932 let out_recv = input
1933 .broadcast_closed(&dest, TCP.fail_stop().bincode())
1934 .entries()
1935 .send(&node, TCP.fail_stop().bincode())
1936 .entries()
1937 .sim_output();
1938
1939 flow.sim()
1940 .with_cluster_size(&source, 2)
1941 .with_cluster_size(&dest, 2)
1942 .exhaustive(async || {
1943 // Each source member (0, 1) broadcasts 123 to each dest member (0, 1).
1944 // The dest members then send to the process keyed by dest member id.
1945 // Each dest member receives (source_0, 123) and (source_1, 123).
1946 out_recv
1947 .assert_yields_only_unordered(vec![
1948 (MemberId::from_raw_id(0), (MemberId::from_raw_id(0), 123)),
1949 (MemberId::from_raw_id(0), (MemberId::from_raw_id(1), 123)),
1950 (MemberId::from_raw_id(1), (MemberId::from_raw_id(0), 123)),
1951 (MemberId::from_raw_id(1), (MemberId::from_raw_id(1), 123)),
1952 ])
1953 .await
1954 });
1955 }
1956
1957 /// Compile-time check that the consistency guarantee of `broadcast_closed` output tracks
1958 /// the network's failure policy: `fail_stop` and `lossy_delayed_forever` preserve
1959 /// [`EventualConsistency`], while plain `lossy` only provides [`NoConsistency`].
1960 #[cfg(feature = "sim")]
1961 #[test]
1962 fn broadcast_closed_consistency_tracks_failure_policy() {
1963 use crate::live_collections::keyed_stream::KeyedStream;
1964 use crate::live_collections::stream::Stream;
1965 use crate::location::Cluster;
1966 use crate::location::cluster::{EventualConsistency, NoConsistency};
1967
1968 let mut flow = FlowBuilder::new();
1969 let cluster = flow.cluster::<()>();
1970 let source = flow.cluster::<()>();
1971 let node = flow.process::<()>();
1972
1973 // `fail_stop` models a failed connection as the recipient having failed, preserving
1974 // eventual consistency across live members.
1975 let _: Stream<u32, Cluster<'_, (), EventualConsistency>, _, _, _> = node
1976 .source_iter(q!(vec![1u32]))
1977 .broadcast_closed(&cluster, TCP.fail_stop().bincode());
1978
1979 // `lossy_delayed_forever` models drops as indefinite delays, preserving eventual
1980 // consistency.
1981 let _: Stream<u32, Cluster<'_, (), EventualConsistency>, _, _, _> = node
1982 .source_iter(q!(vec![1u32]))
1983 .broadcast_closed(&cluster, TCP.lossy_delayed_forever().bincode());
1984
1985 // Plain `lossy` can drop messages for some members while delivering them to others,
1986 // so replicas may permanently diverge.
1987 let _: Stream<u32, Cluster<'_, (), NoConsistency>, _, _, _> = node
1988 .source_iter(q!(vec![1u32]))
1989 .broadcast_closed(&cluster, TCP.lossy(nondet!(/** test */)).bincode());
1990
1991 // The same applies to cluster-to-cluster closed broadcasts.
1992 let _: KeyedStream<MemberId<()>, u32, Cluster<'_, (), EventualConsistency>, _, _, _> =
1993 source
1994 .source_iter(q!(vec![1u32]))
1995 .broadcast_closed(&cluster, TCP.fail_stop().bincode());
1996
1997 let _: KeyedStream<MemberId<()>, u32, Cluster<'_, (), NoConsistency>, _, _, _> = source
1998 .source_iter(q!(vec![1u32]))
1999 .broadcast_closed(&cluster, TCP.lossy(nondet!(/** test */)).bincode());
2000
2001 let _ = flow.finalize();
2002 }
2003}