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::{AsCodeOptions, 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            NetworkingInfo::Udp { .. } => {} // UDP is always lossy, which is always allowed
139        }
140        deploy_maelstrom_m2m(RuntimeData::new("__hydro_lang_maelstrom_meta"))
141    }
142
143    fn m2m_connect(
144        _c1: &Self::Cluster,
145        _c1_port: &<Self::Cluster as Node>::Port,
146        _c2: &Self::Cluster,
147        _c2_port: &<Self::Cluster as Node>::Port,
148    ) -> Box<dyn FnOnce()> {
149        // No runtime connection needed for Maelstrom - all routing is via stdin/stdout
150        Box::new(|| {})
151    }
152
153    fn e2o_many_source(
154        _extra_stmts: &mut Vec<syn::Stmt>,
155        _p2: &Self::Process,
156        _p2_port: &<Self::Process as Node>::Port,
157        _codec_type: &syn::Type,
158        _shared_handle: String,
159    ) -> syn::Expr {
160        panic!("Maelstrom deployment does not support processes, only clusters")
161    }
162
163    fn e2o_many_sink(_shared_handle: String) -> syn::Expr {
164        panic!("Maelstrom deployment does not support processes, only clusters")
165    }
166
167    fn e2o_source(
168        _extra_stmts: &mut Vec<syn::Stmt>,
169        _p1: &Self::External,
170        _p1_port: &<Self::External as Node>::Port,
171        _p2: &Self::Process,
172        _p2_port: &<Self::Process as Node>::Port,
173        _codec_type: &syn::Type,
174        _shared_handle: String,
175    ) -> syn::Expr {
176        panic!("Maelstrom deployment does not support processes, only clusters")
177    }
178
179    fn e2o_connect(
180        _p1: &Self::External,
181        _p1_port: &<Self::External as Node>::Port,
182        _p2: &Self::Process,
183        _p2_port: &<Self::Process as Node>::Port,
184        _many: bool,
185        _server_hint: NetworkHint,
186    ) -> Box<dyn FnOnce()> {
187        panic!("Maelstrom deployment does not support processes, only clusters")
188    }
189
190    fn o2e_sink(
191        _p1: &Self::Process,
192        _p1_port: &<Self::Process as Node>::Port,
193        _p2: &Self::External,
194        _p2_port: &<Self::External as Node>::Port,
195        _shared_handle: String,
196    ) -> syn::Expr {
197        panic!("Maelstrom deployment does not support processes, only clusters")
198    }
199
200    fn cluster_ids(
201        _of_cluster: LocationKey,
202    ) -> impl QuotedWithContext<'a, &'a [TaglessMemberId], ()> + Clone + 'a {
203        cluster_members(RuntimeData::new("__hydro_lang_maelstrom_meta"), _of_cluster)
204    }
205
206    fn cluster_self_id() -> impl QuotedWithContext<'a, TaglessMemberId, ()> + Clone + 'a {
207        cluster_self_id(RuntimeData::new("__hydro_lang_maelstrom_meta"))
208    }
209
210    fn cluster_membership_stream(
211        _env: &mut Self::InstantiateEnv,
212        _at_location: &LocationId,
213        location_id: &LocationId,
214    ) -> impl QuotedWithContext<'a, Box<dyn Stream<Item = (TaglessMemberId, MembershipEvent)> + Unpin>, ()>
215    {
216        cluster_membership_stream(location_id)
217    }
218}
219
220/// A dummy process type for Maelstrom (processes are not supported).
221#[derive(Clone)]
222pub struct MaelstromProcess {
223    _private: (),
224}
225
226impl Node for MaelstromProcess {
227    type Port = String;
228    type Meta = ();
229    type InstantiateEnv = MaelstromDeployment;
230
231    fn next_port(&self) -> Self::Port {
232        panic!("Maelstrom deployment does not support processes")
233    }
234
235    fn update_meta(&self, _meta: &Self::Meta) {}
236
237    fn instantiate(
238        &self,
239        _env: &mut Self::InstantiateEnv,
240        _meta: &mut Self::Meta,
241        _graph: DfirGraph,
242        _extra_stmts: &[syn::Stmt],
243        _sidecars: &[syn::Expr],
244        _as_code_options: &AsCodeOptions,
245    ) {
246        panic!("Maelstrom deployment does not support processes")
247    }
248}
249
250/// Represents a cluster in Maelstrom deployment.
251#[derive(Clone)]
252pub struct MaelstromCluster {
253    next_port: Rc<RefCell<usize>>,
254    name_hint: Option<String>,
255}
256
257impl Node for MaelstromCluster {
258    type Port = String;
259    type Meta = ();
260    type InstantiateEnv = MaelstromDeployment;
261
262    fn next_port(&self) -> Self::Port {
263        let next_port = *self.next_port.borrow();
264        *self.next_port.borrow_mut() += 1;
265        format!("port_{}", next_port)
266    }
267
268    fn update_meta(&self, _meta: &Self::Meta) {}
269
270    fn instantiate(
271        &self,
272        env: &mut Self::InstantiateEnv,
273        _meta: &mut Self::Meta,
274        graph: DfirGraph,
275        extra_stmts: &[syn::Stmt],
276        sidecars: &[syn::Expr],
277        as_code_options: &AsCodeOptions,
278    ) {
279        let (bin_name, config) = create_graph_trybuild(
280            graph,
281            extra_stmts,
282            sidecars,
283            as_code_options,
284            self.name_hint.as_deref(),
285            crate::compile::trybuild::generate::DeployMode::Maelstrom,
286            LinkingMode::Dynamic,
287        );
288
289        env.bin_name = Some(bin_name);
290        env.trybuild = Some(config);
291    }
292}
293
294/// Represents an external client in Maelstrom deployment.
295#[derive(Clone)]
296pub enum MaelstromExternal {}
297
298impl Node for MaelstromExternal {
299    type Port = String;
300    type Meta = ();
301    type InstantiateEnv = MaelstromDeployment;
302
303    fn next_port(&self) -> Self::Port {
304        unreachable!()
305    }
306
307    fn update_meta(&self, _meta: &Self::Meta) {}
308
309    fn instantiate(
310        &self,
311        _env: &mut Self::InstantiateEnv,
312        _meta: &mut Self::Meta,
313        _graph: DfirGraph,
314        _extra_stmts: &[syn::Stmt],
315        _sidecars: &[syn::Expr],
316        _as_code_options: &AsCodeOptions,
317    ) {
318        unreachable!()
319    }
320}
321
322impl<'a> RegisterPort<'a, MaelstromDeploy> for MaelstromExternal {
323    fn register(&self, _external_port_id: ExternalPortId, _port: Self::Port) {
324        unreachable!()
325    }
326
327    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
328    fn as_bytes_bidi(
329        &self,
330        _external_port_id: ExternalPortId,
331    ) -> impl Future<
332        Output = (
333            Pin<Box<dyn Stream<Item = Result<BytesMut, Error>>>>,
334            Pin<Box<dyn Sink<Bytes, Error = Error>>>,
335        ),
336    > + 'a {
337        async move { unreachable!() }
338    }
339
340    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
341    fn as_bincode_bidi<InT, OutT>(
342        &self,
343        _external_port_id: ExternalPortId,
344    ) -> impl Future<
345        Output = (
346            Pin<Box<dyn Stream<Item = OutT>>>,
347            Pin<Box<dyn Sink<InT, Error = Error>>>,
348        ),
349    > + 'a
350    where
351        InT: Serialize + 'static,
352        OutT: DeserializeOwned + 'static,
353    {
354        async move { unreachable!() }
355    }
356
357    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
358    fn as_bincode_sink<T: Serialize + 'static>(
359        &self,
360        _external_port_id: ExternalPortId,
361    ) -> impl Future<Output = Pin<Box<dyn Sink<T, Error = Error>>>> + 'a {
362        async move { unreachable!() }
363    }
364
365    #[expect(clippy::manual_async_fn, reason = "false positive, involves lifetimes")]
366    fn as_bincode_source<T: DeserializeOwned + 'static>(
367        &self,
368        _external_port_id: ExternalPortId,
369    ) -> impl Future<Output = Pin<Box<dyn Stream<Item = T>>>> + 'a {
370        async move { unreachable!() }
371    }
372}
373
374/// Specification for building a Maelstrom cluster.
375#[derive(Clone)]
376pub struct MaelstromClusterSpec;
377
378impl<'a> ClusterSpec<'a, MaelstromDeploy> for MaelstromClusterSpec {
379    fn build(self, key: LocationKey, name_hint: &str) -> MaelstromCluster {
380        assert_eq!(
381            key,
382            LocationKey::FIRST,
383            "there should only be one location for a Maelstrom deployment"
384        );
385        MaelstromCluster {
386            next_port: Rc::new(RefCell::new(0)),
387            name_hint: Some(name_hint.to_owned()),
388        }
389    }
390}
391
392/// The Maelstrom deployment environment.
393///
394/// This holds configuration for the Maelstrom run and accumulates
395/// compilation artifacts during deployment.
396pub struct MaelstromDeployment {
397    /// Number of nodes in the cluster.
398    pub node_count: usize,
399    /// Path to the maelstrom binary.
400    pub maelstrom_path: PathBuf,
401    /// Workload to run (e.g., "echo", "broadcast", "g-counter").
402    pub workload: String,
403    /// Time limit in seconds.
404    pub time_limit: Option<u64>,
405    /// Rate of requests per second.
406    pub rate: Option<u64>,
407    /// The availability of nodes.
408    pub availability: Option<String>,
409    /// Nemesis to run during tests.
410    pub nemesis: Option<String>,
411    /// Additional maelstrom arguments.
412    pub extra_args: Vec<String>,
413
414    // Populated during deployment
415    pub(crate) bin_name: Option<String>,
416    pub(crate) trybuild: Option<TrybuildConfig>,
417}
418
419impl MaelstromDeployment {
420    /// Create a new Maelstrom deployment with the given node count.
421    pub fn new(workload: impl Into<String>) -> Self {
422        Self {
423            node_count: 1,
424            maelstrom_path: PathBuf::from("maelstrom"),
425            workload: workload.into(),
426            time_limit: None,
427            rate: None,
428            availability: None,
429            nemesis: None,
430            extra_args: vec![],
431            bin_name: None,
432            trybuild: None,
433        }
434    }
435
436    /// Set the node count.
437    pub fn node_count(mut self, count: usize) -> Self {
438        self.node_count = count;
439        self
440    }
441
442    /// Set the path to the maelstrom binary.
443    pub fn maelstrom_path(mut self, path: impl Into<PathBuf>) -> Self {
444        self.maelstrom_path = path.into();
445        self
446    }
447
448    /// Set the time limit in seconds.
449    pub fn time_limit(mut self, seconds: u64) -> Self {
450        self.time_limit = Some(seconds);
451        self
452    }
453
454    /// Set the request rate per second.
455    pub fn rate(mut self, rate: u64) -> Self {
456        self.rate = Some(rate);
457        self
458    }
459
460    /// Set the availability for the test.
461    pub fn availability(mut self, availability: impl Into<String>) -> Self {
462        self.availability = Some(availability.into());
463        self
464    }
465
466    /// Set the nemesis for the test.
467    pub fn nemesis(mut self, nemesis: impl Into<String>) -> Self {
468        self.nemesis = Some(nemesis.into());
469        self
470    }
471
472    /// Add extra arguments to pass to maelstrom.
473    pub fn extra_args(mut self, args: impl IntoIterator<Item = impl Into<String>>) -> Self {
474        self.extra_args.extend(args.into_iter().map(Into::into));
475        self
476    }
477
478    /// Build the compiled binary in dev mode.
479    /// Returns the path to the compiled binary.
480    ///
481    /// This shares the same parallel-compilation machinery as the simulator: the
482    /// program is linked dynamically against a prebuilt dylib of its dependencies,
483    /// so repeated and concurrent builds only need to recompile the generated
484    /// example itself.
485    pub fn build(&self) -> Result<PathBuf, Error> {
486        let bin_name = self
487            .bin_name
488            .as_ref()
489            .expect("No binary name set - did you call deploy?");
490        let trybuild = self
491            .trybuild
492            .as_ref()
493            .expect("No trybuild config set - did you call deploy?");
494
495        let out = compile_trybuild_example(ExampleBuildConfig {
496            trybuild: trybuild.clone(),
497            bin_name: bin_name.clone(),
498            runtime_feature: "hydro___feature_maelstrom_runtime",
499            // Maelstrom builds the generated example directly as an executable.
500            example_name: bin_name.clone(),
501            crate_type: None,
502            set_trybuild_lib_name: false,
503            allow_fuzz: false,
504        })
505        .map_err(|()| Error::other("Maelstrom binary compilation failed"))?;
506
507        // Persist the built executable so it survives past the temporary build guards.
508        out.keep().map_err(|e| Error::other(e.to_string()))
509    }
510
511    /// Run Maelstrom with the compiled binary, return Ok(()) if all checks pass.
512    ///
513    /// This will block until Maelstrom completes.
514    pub fn run(self) -> Result<(), Error> {
515        let binary_path = self.build()?;
516
517        // Warm up the binary before handing it to Maelstrom. On macOS, the
518        // first execution of a freshly written binary triggers a Gatekeeper /
519        // XProtect (`syspolicyd`) scan that can take several seconds on loaded
520        // CI machines. Maelstrom only waits 10 seconds for each node to answer
521        // the `init` RPC, so a cold first exec (multiplied across concurrently
522        // launched nodes) can cause spurious node-startup timeouts. The warmup
523        // invocation sees EOF on stdin and exits immediately, priming the
524        // system's first-exec caches for the real run.
525        std::process::Command::new(&binary_path)
526            .stdin(Stdio::null())
527            .stdout(Stdio::null())
528            .stderr(Stdio::null())
529            .spawn()?
530            .wait()?;
531
532        // Use a unique working directory per run to avoid conflicts with concurrent tests.
533        let run_dir = tempfile::tempdir().map_err(Error::other)?;
534
535        let mut cmd = std::process::Command::new(&self.maelstrom_path);
536        cmd.arg("test")
537            .arg("-w")
538            .arg(&self.workload)
539            .arg("--bin")
540            .arg(&binary_path)
541            .arg("--node-count")
542            .arg(self.node_count.to_string())
543            .current_dir(run_dir.path())
544            .stdout(Stdio::piped());
545
546        if let Some(time_limit) = self.time_limit {
547            cmd.arg("--time-limit").arg(time_limit.to_string());
548        }
549
550        if let Some(rate) = self.rate {
551            cmd.arg("--rate").arg(rate.to_string());
552        }
553
554        if let Some(availability) = self.availability {
555            cmd.arg("--availability").arg(availability);
556        }
557
558        if let Some(nemesis) = self.nemesis {
559            cmd.arg("--nemesis").arg(nemesis);
560        }
561
562        for arg in &self.extra_args {
563            cmd.arg(arg);
564        }
565
566        let spawned = cmd.spawn()?;
567
568        for line in BufReader::new(spawned.stdout.unwrap()).lines() {
569            let line = line?;
570            eprintln!("{}", line);
571
572            if line.starts_with("Analysis invalid!") {
573                let path = run_dir.keep();
574                dump_node_logs(&path);
575                return Err(Error::other(format!(
576                    "Analysis was invalid. Maelstrom store at: {}",
577                    path.display()
578                )));
579            } else if line.starts_with("Errors occurred during analysis, but no anomalies found.")
580                || line.starts_with("Everything looks good!")
581            {
582                return Ok(());
583            }
584        }
585
586        let path = run_dir.keep();
587        dump_node_logs(&path);
588        Err(Error::other(format!(
589            "Maelstrom produced an unexpected result. Store at: {}",
590            path.display()
591        )))
592    }
593
594    /// Get the path to the compiled binary, building it if necessary.
595    pub fn binary_path(&self) -> Option<PathBuf> {
596        self.build().ok()
597    }
598}
599
600/// Print the per-node logs from a Maelstrom run directory to stderr.
601///
602/// Maelstrom truncates node stderr in its own error messages (keeping only the
603/// tail), so when a node crashes the actual panic message is often cut off.
604/// The full logs live in `store/<workload>/<timestamp>/node-logs/*.log` under
605/// the run directory; dump them so failures are debuggable in CI, where the
606/// preserved store directory is not otherwise accessible.
607fn dump_node_logs(run_dir: &Path) {
608    fn collect(dir: &Path, out: &mut Vec<PathBuf>) {
609        let Ok(entries) = std::fs::read_dir(dir) else {
610            return;
611        };
612        for entry in entries.flatten() {
613            let path = entry.path();
614            // Skip symlinks (e.g. `store/latest`) to avoid duplicates.
615            if path.is_symlink() || !path.is_dir() {
616                continue;
617            }
618            if path.file_name().is_some_and(|name| name == "node-logs") {
619                if let Ok(logs) = std::fs::read_dir(&path) {
620                    out.extend(logs.flatten().map(|e| e.path()));
621                }
622            } else {
623                collect(&path, out);
624            }
625        }
626    }
627
628    let mut log_files = Vec::new();
629    collect(&run_dir.join("store"), &mut log_files);
630    log_files.sort();
631
632    for log in log_files {
633        eprintln!("==== Maelstrom node log: {} ====", log.display());
634        match std::fs::read_to_string(&log) {
635            Ok(contents) => eprint!("{}", contents),
636            Err(e) => eprintln!("(failed to read log: {})", e),
637        }
638        eprintln!("==== end of node log: {} ====", log.display());
639    }
640}
hydro_lang/deploy/maelstrom/deploy_maelstrom.rs

hydro_lang/deploy/maelstrom/
deploy_maelstrom.rs
