hydro_lang/deploy/maelstrom/
deploy_maelstrom.rs

1//! Deployment backend for Hydro that targets Maelstrom for distributed systems testing.
2//!
3//! Maelstrom is a workbench for learning distributed systems by writing your own.
4//! This backend compiles Hydro programs to binaries that communicate via Maelstrom's
5//! stdin/stdout JSON protocol.
6
7use std::cell::RefCell;
8use std::future::Future;
9use std::io::{BufRead, BufReader, Error};
10use std::path::{Path, PathBuf};
11use std::pin::Pin;
12use std::process::Stdio;
13use std::rc::Rc;
14
15use bytes::{Bytes, BytesMut};
16use dfir_lang::graph::DfirGraph;
17use futures::{Sink, Stream};
18use serde::Serialize;
19use serde::de::DeserializeOwned;
20use stageleft::{QuotedWithContext, RuntimeData};
21
22use super::deploy_runtime_maelstrom::*;
23use crate::compile::builder::ExternalPortId;
24use crate::compile::deploy_provider::{ClusterSpec, Deploy, Node, RegisterPort};
25use crate::compile::trybuild::generate::{
26    ExampleBuildConfig, LinkingMode, TrybuildConfig, compile_trybuild_example,
27    create_graph_trybuild,
28};
29use crate::location::dynamic::LocationId;
30use crate::location::member_id::TaglessMemberId;
31use crate::location::{LocationKey, MembershipEvent, NetworkHint};
32
33/// Deployment backend that targets Maelstrom for distributed systems testing.
34///
35/// This backend compiles Hydro programs to binaries that communicate via Maelstrom's
36/// stdin/stdout JSON protocol. It is restricted to programs with:
37/// - Exactly one cluster (no processes)
38/// - A single external input channel for client communication
39pub enum MaelstromDeploy {}
40
41impl<'a> Deploy<'a> for MaelstromDeploy {
42    type Meta = ();
43    type InstantiateEnv = MaelstromDeployment;
44
45    type Process = MaelstromProcess;
46    type Cluster = MaelstromCluster;
47    type External = MaelstromExternal;
48
49    fn o2o_sink_source(
50        _env: &mut Self::InstantiateEnv,
51        _p1: &Self::Process,
52        _p1_port: &<Self::Process as Node>::Port,
53        _p2: &Self::Process,
54        _p2_port: &<Self::Process as Node>::Port,
55        _name: Option<&str>,
56        _networking_info: &crate::networking::NetworkingInfo,
57        _external_types: Option<(&syn::Type, &syn::Type)>,
58    ) -> (syn::Expr, syn::Expr) {
59        panic!("Maelstrom deployment does not support processes, only clusters")
60    }
61
62    fn o2o_connect(
63        _p1: &Self::Process,
64        _p1_port: &<Self::Process as Node>::Port,
65        _p2: &Self::Process,
66        _p2_port: &<Self::Process as Node>::Port,
67    ) -> Box<dyn FnOnce()> {
68        panic!("Maelstrom deployment does not support processes, only clusters")
69    }
70
71    fn o2m_sink_source(
72        _env: &mut Self::InstantiateEnv,
73        _p1: &Self::Process,
74        _p1_port: &<Self::Process as Node>::Port,
75        _c2: &Self::Cluster,
76        _c2_port: &<Self::Cluster as Node>::Port,
77        _name: Option<&str>,
78        _networking_info: &crate::networking::NetworkingInfo,
79        _external_types: Option<(&syn::Type, &syn::Type)>,
80    ) -> (syn::Expr, syn::Expr) {
81        panic!("Maelstrom deployment does not support processes, only clusters")
82    }
83
84    fn o2m_connect(
85        _p1: &Self::Process,
86        _p1_port: &<Self::Process as Node>::Port,
87        _c2: &Self::Cluster,
88        _c2_port: &<Self::Cluster as Node>::Port,
89    ) -> Box<dyn FnOnce()> {
90        panic!("Maelstrom deployment does not support processes, only clusters")
91    }
92
93    fn m2o_sink_source(
94        _env: &mut Self::InstantiateEnv,
95        _c1: &Self::Cluster,
96        _c1_port: &<Self::Cluster as Node>::Port,
97        _p2: &Self::Process,
98        _p2_port: &<Self::Process as Node>::Port,
99        _name: Option<&str>,
100        _networking_info: &crate::networking::NetworkingInfo,
101        _external_types: Option<(&syn::Type, &syn::Type)>,
102    ) -> (syn::Expr, syn::Expr) {
103        panic!("Maelstrom deployment does not support processes, only clusters")
104    }
105
106    fn m2o_connect(
107        _c1: &Self::Cluster,
108        _c1_port: &<Self::Cluster as Node>::Port,
109        _p2: &Self::Process,
110        _p2_port: &<Self::Process as Node>::Port,
111    ) -> Box<dyn FnOnce()> {
112        panic!("Maelstrom deployment does not support processes, only clusters")
113    }
114
115    fn m2m_sink_source(
116        env: &mut Self::InstantiateEnv,
117        _c1: &Self::Cluster,
118        _c1_port: &<Self::Cluster as Node>::Port,
119        _c2: &Self::Cluster,
120        _c2_port: &<Self::Cluster as Node>::Port,
121        _name: Option<&str>,
122        networking_info: &crate::networking::NetworkingInfo,
123        _external_types: Option<(&syn::Type, &syn::Type)>,
124    ) -> (syn::Expr, syn::Expr) {
125        use crate::networking::{NetworkingInfo, TcpFault};
126        match networking_info {
127            NetworkingInfo::Tcp { fault } => match (fault, env.nemesis.as_deref()) {
128                (TcpFault::Lossy | TcpFault::LossyDelayedForever, _) => {} /* lossy/delayed are always allowed */
129                (_, None) => {} // no nemesis means any fault model is fine
130                (TcpFault::FailStop, Some("partition")) => {
131                    panic!(
132                        "Maelstrom partition nemesis requires lossy networking, but fail_stop was used. \
133                         Use `TCP.lossy().bincode()` or `TCP.lossy_delayed_forever().bincode()` instead of `TCP.fail_stop().bincode()`."
134                    );
135                }
136                (TcpFault::FailStop, Some(_)) => {} // other nemeses are fine with fail_stop
137            },
138        }
139        deploy_maelstrom_m2m(RuntimeData::new("__hydro_lang_maelstrom_meta"))
140    }
141
142    fn m2m_connect(
143        _c1: &Self::Cluster,
144        _c1_port: &<Self::Cluster as Node>::Port,
145        _c2: &Self::Cluster,
146        _c2_port: &<Self::Cluster as Node>::Port,
147    ) -> Box<dyn FnOnce()> {
148        // No runtime connection needed for Maelstrom - all routing is via stdin/stdout
149        Box::new(|| {})
150    }
151
152    fn e2o_many_source(
153        _extra_stmts: &mut Vec<syn::Stmt>,
154        _p2: &Self::Process,
155        _p2_port: &<Self::Process as Node>::Port,
156        _codec_type: &syn::Type,
157        _shared_handle: String,
158    ) -> syn::Expr {
159        panic!("Maelstrom deployment does not support processes, only clusters")
160    }
161
162    fn e2o_many_sink(_shared_handle: String) -> syn::Expr {
163        panic!("Maelstrom deployment does not support processes, only clusters")
164    }
165
166    fn e2o_source(
167        _extra_stmts: &mut Vec<syn::Stmt>,
168        _p1: &Self::External,
169        _p1_port: &<Self::External as Node>::Port,
170        _p2: &Self::Process,
171        _p2_port: &<Self::Process as Node>::Port,
172        _codec_type: &syn::Type,
173        _shared_handle: String,
174    ) -> syn::Expr {
175        panic!("Maelstrom deployment does not support processes, only clusters")
176    }
177
178    fn e2o_connect(
179        _p1: &Self::External,
180        _p1_port: &<Self::External as Node>::Port,
181        _p2: &Self::Process,
182        _p2_port: &<Self::Process as Node>::Port,
183        _many: bool,
184        _server_hint: NetworkHint,
185    ) -> Box<dyn FnOnce()> {
186        panic!("Maelstrom deployment does not support processes, only clusters")
187    }
188
189    fn o2e_sink(
190        _p1: &Self::Process,
191        _p1_port: &<Self::Process as Node>::Port,
192        _p2: &Self::External,
193        _p2_port: &<Self::External as Node>::Port,
194        _shared_handle: String,
195    ) -> syn::Expr {
196        panic!("Maelstrom deployment does not support processes, only clusters")
197    }
198
199    fn cluster_ids(
200        _of_cluster: LocationKey,
201    ) -> impl QuotedWithContext<'a, &'a [TaglessMemberId], ()> + Clone + 'a {
202        cluster_members(RuntimeData::new("__hydro_lang_maelstrom_meta"), _of_cluster)
203    }
204
205    fn cluster_self_id() -> impl QuotedWithContext<'a, TaglessMemberId, ()> + Clone + 'a {
206        cluster_self_id(RuntimeData::new("__hydro_lang_maelstrom_meta"))
207    }
208
209    fn cluster_membership_stream(
210        _env: &mut Self::InstantiateEnv,
211        _at_location: &LocationId,
212        location_id: &LocationId,
213    ) -> impl QuotedWithContext<'a, Box<dyn Stream<Item = (TaglessMemberId, MembershipEvent)> + Unpin>, ()>
214    {
215        cluster_membership_stream(location_id)
216    }
217}
218
219/// A dummy process type for Maelstrom (processes are not supported).
220#[derive(Clone)]
221pub struct MaelstromProcess {
222    _private: (),
223}
224
225impl Node for MaelstromProcess {
226    type Port = String;
227    type Meta = ();
228    type InstantiateEnv = MaelstromDeployment;
229
230    fn next_port(&self) -> Self::Port {
231        panic!("Maelstrom deployment does not support processes")
232    }
233
234    fn update_meta(&self, _meta: &Self::Meta) {}
235
236    fn instantiate(
237        &self,
238        _env: &mut Self::InstantiateEnv,
239        _meta: &mut Self::Meta,
240        _graph: DfirGraph,
241        _extra_stmts: &[syn::Stmt],
242        _sidecars: &[syn::Expr],
243    ) {
244        panic!("Maelstrom deployment does not support processes")
245    }
246}
247
248/// Represents a cluster in Maelstrom deployment.
249#[derive(Clone)]
250pub struct MaelstromCluster {
251    next_port: Rc<RefCell<usize>>,
252    name_hint: Option<String>,
253}
254
255impl Node for MaelstromCluster {
256    type Port = String;
257    type Meta = ();
258    type InstantiateEnv = MaelstromDeployment;
259
260    fn next_port(&self) -> Self::Port {
261        let next_port = *self.next_port.borrow();
262        *self.next_port.borrow_mut() += 1;
263        format!("port_{}", next_port)
264    }
265
266    fn update_meta(&self, _meta: &Self::Meta) {}
267
268    fn instantiate(
269        &self,
270        env: &mut Self::InstantiateEnv,
271        _meta: &mut Self::Meta,
272        graph: DfirGraph,
273        extra_stmts: &[syn::Stmt],
274        sidecars: &[syn::Expr],
275    ) {
276        let (bin_name, config) = create_graph_trybuild(
277            graph,
278            extra_stmts,
279            sidecars,
280            self.name_hint.as_deref(),
281            crate::compile::trybuild::generate::DeployMode::Maelstrom,
282            LinkingMode::Dynamic,
283        );
284
285        env.bin_name = Some(bin_name);
286        env.trybuild = Some(config);
287    }
288}
289
290/// Represents an external client in Maelstrom deployment.
291#[derive(Clone)]
292pub enum MaelstromExternal {}
293
294impl Node for MaelstromExternal {
295    type Port = String;
296    type Meta = ();
297    type InstantiateEnv = MaelstromDeployment;
298
299    fn next_port(&self) -> Self::Port {
300        unreachable!()
301    }
302
303    fn update_meta(&self, _meta: &Self::Meta) {}
304
305    fn instantiate(
306        &self,
307        _env: &mut Self::InstantiateEnv,
308        _meta: &mut Self::Meta,
309        _graph: DfirGraph,
310        _extra_stmts: &[syn::Stmt],
311        _sidecars: &[syn::Expr],
312    ) {
313        unreachable!()
314    }
315}
316
317impl<'a> RegisterPort<'a, MaelstromDeploy> for MaelstromExternal {
318    fn register(&self, _external_port_id: ExternalPortId, _port: Self::Port) {
319        unreachable!()
320    }
321
322    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
323    fn as_bytes_bidi(
324        &self,
325        _external_port_id: ExternalPortId,
326    ) -> impl Future<
327        Output = (
328            Pin<Box<dyn Stream<Item = Result<BytesMut, Error>>>>,
329            Pin<Box<dyn Sink<Bytes, Error = Error>>>,
330        ),
331    > + 'a {
332        async move { unreachable!() }
333    }
334
335    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
336    fn as_bincode_bidi<InT, OutT>(
337        &self,
338        _external_port_id: ExternalPortId,
339    ) -> impl Future<
340        Output = (
341            Pin<Box<dyn Stream<Item = OutT>>>,
342            Pin<Box<dyn Sink<InT, Error = Error>>>,
343        ),
344    > + 'a
345    where
346        InT: Serialize + 'static,
347        OutT: DeserializeOwned + 'static,
348    {
349        async move { unreachable!() }
350    }
351
352    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
353    fn as_bincode_sink<T: Serialize + 'static>(
354        &self,
355        _external_port_id: ExternalPortId,
356    ) -> impl Future<Output = Pin<Box<dyn Sink<T, Error = Error>>>> + 'a {
357        async move { unreachable!() }
358    }
359
360    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
361    fn as_bincode_source<T: DeserializeOwned + 'static>(
362        &self,
363        _external_port_id: ExternalPortId,
364    ) -> impl Future<Output = Pin<Box<dyn Stream<Item = T>>>> + 'a {
365        async move { unreachable!() }
366    }
367}
368
369/// Specification for building a Maelstrom cluster.
370#[derive(Clone)]
371pub struct MaelstromClusterSpec;
372
373impl<'a> ClusterSpec<'a, MaelstromDeploy> for MaelstromClusterSpec {
374    fn build(self, key: LocationKey, name_hint: &str) -> MaelstromCluster {
375        assert_eq!(
376            key,
377            LocationKey::FIRST,
378            "there should only be one location for a Maelstrom deployment"
379        );
380        MaelstromCluster {
381            next_port: Rc::new(RefCell::new(0)),
382            name_hint: Some(name_hint.to_owned()),
383        }
384    }
385}
386
387/// The Maelstrom deployment environment.
388///
389/// This holds configuration for the Maelstrom run and accumulates
390/// compilation artifacts during deployment.
391pub struct MaelstromDeployment {
392    /// Number of nodes in the cluster.
393    pub node_count: usize,
394    /// Path to the maelstrom binary.
395    pub maelstrom_path: PathBuf,
396    /// Workload to run (e.g., "echo", "broadcast", "g-counter").
397    pub workload: String,
398    /// Time limit in seconds.
399    pub time_limit: Option<u64>,
400    /// Rate of requests per second.
401    pub rate: Option<u64>,
402    /// The availability of nodes.
403    pub availability: Option<String>,
404    /// Nemesis to run during tests.
405    pub nemesis: Option<String>,
406    /// Additional maelstrom arguments.
407    pub extra_args: Vec<String>,
408
409    // Populated during deployment
410    pub(crate) bin_name: Option<String>,
411    pub(crate) trybuild: Option<TrybuildConfig>,
412}
413
414impl MaelstromDeployment {
415    /// Create a new Maelstrom deployment with the given node count.
416    pub fn new(workload: impl Into<String>) -> Self {
417        Self {
418            node_count: 1,
419            maelstrom_path: PathBuf::from("maelstrom"),
420            workload: workload.into(),
421            time_limit: None,
422            rate: None,
423            availability: None,
424            nemesis: None,
425            extra_args: vec![],
426            bin_name: None,
427            trybuild: None,
428        }
429    }
430
431    /// Set the node count.
432    pub fn node_count(mut self, count: usize) -> Self {
433        self.node_count = count;
434        self
435    }
436
437    /// Set the path to the maelstrom binary.
438    pub fn maelstrom_path(mut self, path: impl Into<PathBuf>) -> Self {
439        self.maelstrom_path = path.into();
440        self
441    }
442
443    /// Set the time limit in seconds.
444    pub fn time_limit(mut self, seconds: u64) -> Self {
445        self.time_limit = Some(seconds);
446        self
447    }
448
449    /// Set the request rate per second.
450    pub fn rate(mut self, rate: u64) -> Self {
451        self.rate = Some(rate);
452        self
453    }
454
455    /// Set the availability for the test.
456    pub fn availability(mut self, availability: impl Into<String>) -> Self {
457        self.availability = Some(availability.into());
458        self
459    }
460
461    /// Set the nemesis for the test.
462    pub fn nemesis(mut self, nemesis: impl Into<String>) -> Self {
463        self.nemesis = Some(nemesis.into());
464        self
465    }
466
467    /// Add extra arguments to pass to maelstrom.
468    pub fn extra_args(mut self, args: impl IntoIterator<Item = impl Into<String>>) -> Self {
469        self.extra_args.extend(args.into_iter().map(Into::into));
470        self
471    }
472
473    /// Build the compiled binary in dev mode.
474    /// Returns the path to the compiled binary.
475    ///
476    /// This shares the same parallel-compilation machinery as the simulator: the
477    /// program is linked dynamically against a prebuilt dylib of its dependencies,
478    /// so repeated and concurrent builds only need to recompile the generated
479    /// example itself.
480    pub fn build(&self) -> Result<PathBuf, Error> {
481        let bin_name = self
482            .bin_name
483            .as_ref()
484            .expect("No binary name set - did you call deploy?");
485        let trybuild = self
486            .trybuild
487            .as_ref()
488            .expect("No trybuild config set - did you call deploy?");
489
490        let out = compile_trybuild_example(ExampleBuildConfig {
491            trybuild: trybuild.clone(),
492            bin_name: bin_name.clone(),
493            runtime_feature: "hydro___feature_maelstrom_runtime",
494            // Maelstrom builds the generated example directly as an executable.
495            example_name: bin_name.clone(),
496            crate_type: None,
497            set_trybuild_lib_name: false,
498            allow_fuzz: false,
499        })
500        .map_err(|()| Error::other("Maelstrom binary compilation failed"))?;
501
502        // Persist the built executable so it survives past the temporary build guards.
503        out.keep().map_err(|e| Error::other(e.to_string()))
504    }
505
506    /// Run Maelstrom with the compiled binary, return Ok(()) if all checks pass.
507    ///
508    /// This will block until Maelstrom completes.
509    pub fn run(self) -> Result<(), Error> {
510        let binary_path = self.build()?;
511
512        // Warm up the binary before handing it to Maelstrom. On macOS, the
513        // first execution of a freshly written binary triggers a Gatekeeper /
514        // XProtect (`syspolicyd`) scan that can take several seconds on loaded
515        // CI machines. Maelstrom only waits 10 seconds for each node to answer
516        // the `init` RPC, so a cold first exec (multiplied across concurrently
517        // launched nodes) can cause spurious node-startup timeouts. The warmup
518        // invocation sees EOF on stdin and exits immediately, priming the
519        // system's first-exec caches for the real run.
520        std::process::Command::new(&binary_path)
521            .stdin(Stdio::null())
522            .stdout(Stdio::null())
523            .stderr(Stdio::null())
524            .spawn()?
525            .wait()?;
526
527        // Use a unique working directory per run to avoid conflicts with concurrent tests.
528        let run_dir = tempfile::tempdir().map_err(Error::other)?;
529
530        let mut cmd = std::process::Command::new(&self.maelstrom_path);
531        cmd.arg("test")
532            .arg("-w")
533            .arg(&self.workload)
534            .arg("--bin")
535            .arg(&binary_path)
536            .arg("--node-count")
537            .arg(self.node_count.to_string())
538            .current_dir(run_dir.path())
539            .stdout(Stdio::piped());
540
541        if let Some(time_limit) = self.time_limit {
542            cmd.arg("--time-limit").arg(time_limit.to_string());
543        }
544
545        if let Some(rate) = self.rate {
546            cmd.arg("--rate").arg(rate.to_string());
547        }
548
549        if let Some(availability) = self.availability {
550            cmd.arg("--availability").arg(availability);
551        }
552
553        if let Some(nemesis) = self.nemesis {
554            cmd.arg("--nemesis").arg(nemesis);
555        }
556
557        for arg in &self.extra_args {
558            cmd.arg(arg);
559        }
560
561        let spawned = cmd.spawn()?;
562
563        for line in BufReader::new(spawned.stdout.unwrap()).lines() {
564            let line = line?;
565            eprintln!("{}", &line);
566
567            if line.starts_with("Analysis invalid!") {
568                let path = run_dir.keep();
569                dump_node_logs(&path);
570                return Err(Error::other(format!(
571                    "Analysis was invalid. Maelstrom store at: {}",
572                    path.display()
573                )));
574            } else if line.starts_with("Errors occurred during analysis, but no anomalies found.")
575                || line.starts_with("Everything looks good!")
576            {
577                return Ok(());
578            }
579        }
580
581        let path = run_dir.keep();
582        dump_node_logs(&path);
583        Err(Error::other(format!(
584            "Maelstrom produced an unexpected result. Store at: {}",
585            path.display()
586        )))
587    }
588
589    /// Get the path to the compiled binary, building it if necessary.
590    pub fn binary_path(&self) -> Option<PathBuf> {
591        self.build().ok()
592    }
593}
594
595/// Print the per-node logs from a Maelstrom run directory to stderr.
596///
597/// Maelstrom truncates node stderr in its own error messages (keeping only the
598/// tail), so when a node crashes the actual panic message is often cut off.
599/// The full logs live in `store/<workload>/<timestamp>/node-logs/*.log` under
600/// the run directory; dump them so failures are debuggable in CI, where the
601/// preserved store directory is not otherwise accessible.
602fn dump_node_logs(run_dir: &Path) {
603    fn collect(dir: &Path, out: &mut Vec<PathBuf>) {
604        let Ok(entries) = std::fs::read_dir(dir) else {
605            return;
606        };
607        for entry in entries.flatten() {
608            let path = entry.path();
609            // Skip symlinks (e.g. `store/latest`) to avoid duplicates.
610            if path.is_symlink() || !path.is_dir() {
611                continue;
612            }
613            if path.file_name().is_some_and(|name| name == "node-logs") {
614                if let Ok(logs) = std::fs::read_dir(&path) {
615                    out.extend(logs.flatten().map(|e| e.path()));
616                }
617            } else {
618                collect(&path, out);
619            }
620        }
621    }
622
623    let mut log_files = Vec::new();
624    collect(&run_dir.join("store"), &mut log_files);
625    log_files.sort();
626
627    for log in log_files {
628        eprintln!("==== Maelstrom node log: {} ====", log.display());
629        match std::fs::read_to_string(&log) {
630            Ok(contents) => eprint!("{}", contents),
631            Err(e) => eprintln!("(failed to read log: {})", e),
632        }
633        eprintln!("==== end of node log: {} ====", log.display());
634    }
635}
hydro_lang/deploy/maelstrom/deploy_maelstrom.rs

hydro_lang/deploy/maelstrom/
deploy_maelstrom.rs
