Skip to main content

orbit_runtime/
server.rs

1//! HTTP API (RFC-0001 §11.2) and the HTTP/JSON binding of the worker
2//! protocol (§8). See `protocol/README.md` for the wire format.
3
4use crate::engine::{Engine, EngineError, RegisterWorker, TaskAssignment};
5use crate::identity::{Identities, WorkerIdentity};
6use crate::model::TaskError;
7use axum::extract::{DefaultBodyLimit, Path, Query, Request, State};
8use axum::http::{HeaderMap, HeaderValue, StatusCode};
9use axum::middleware::{self, Next};
10use axum::response::{IntoResponse, Response};
11use axum::routing::{get, post};
12use axum::{Extension, Json, Router};
13use serde::Deserialize;
14use serde_json::{json, Value};
15use std::net::SocketAddr;
16use std::sync::{Arc, Mutex};
17use std::time::Duration;
18use tokio::sync::Notify;
19
20pub const API_VERSION: &str = "orbit-api/1";
21
22pub struct AppState {
23    pub engine: Mutex<Engine>,
24    pub changed: Arc<Notify>,
25    pub auth_token: Option<String>,
26    /// Optional credential restricted to the worker protocol.
27    pub worker_token: Option<String>,
28    /// Per-worker identities (`--workers-file`).
29    pub identities: Identities,
30}
31
32pub type Shared = Arc<AppState>;
33
34pub struct ApiError(pub(crate) EngineError);
35
36impl From<EngineError> for ApiError {
37    fn from(e: EngineError) -> Self {
38        ApiError(e)
39    }
40}
41
42impl IntoResponse for ApiError {
43    fn into_response(self) -> Response {
44        let status =
45            StatusCode::from_u16(self.0.status).unwrap_or(StatusCode::INTERNAL_SERVER_ERROR);
46        (
47            status,
48            Json(json!({ "error": { "code": self.0.code, "message": self.0.message } })),
49        )
50            .into_response()
51    }
52}
53
54fn bad_request(msg: impl Into<String>) -> ApiError {
55    ApiError(EngineError::new(400, "INVALID_REQUEST", msg))
56}
57
58type ApiResult<T> = Result<T, ApiError>;
59
60pub(crate) fn engine(state: &Shared) -> std::sync::MutexGuard<'_, Engine> {
61    state.engine.lock().unwrap_or_else(|p| p.into_inner())
62}
63
64pub fn router(state: Shared, body_limit: usize) -> Router {
65    Router::new()
66        .route("/v1/health", get(health))
67        .route("/v1/metrics", get(metrics))
68        .route("/v1/workflows", post(register_workflow).get(list_workflows))
69        .route("/v1/workflows/{name}", get(get_workflow_latest))
70        .route("/v1/workflows/{name}/{version}", get(get_workflow))
71        .route("/v1/executions", post(start_execution).get(list_executions))
72        .route("/v1/executions/{id}", get(get_execution))
73        .route("/v1/executions/{id}/history", get(get_history))
74        .route("/v1/executions/{id}/cancel", post(cancel_execution))
75        .route("/v1/executions/{id}/signals/{name}", post(send_signal))
76        .route("/v1/executions/{id}/queries/{name}", get(run_query))
77        .route("/v1/executions/{id}/wait", get(wait_execution))
78        .route("/v1/workers", get(list_workers))
79        .route("/v1/workers/register", post(register_worker))
80        .route("/v1/workers/heartbeat", post(heartbeat))
81        .route("/v1/workers/poll", post(poll))
82        .route("/v1/workers/unregister", post(unregister))
83        .route("/v1/tasks/complete", post(complete_task))
84        .route("/v1/tasks/fail", post(fail_task))
85        .layer(middleware::from_fn_with_state(state.clone(), auth))
86        .layer(DefaultBodyLimit::max(body_limit))
87        .with_state(state)
88}
89
90/// Endpoints a worker credential may call.
91fn is_worker_path(path: &str) -> bool {
92    path.starts_with("/v1/workers/") || path.starts_with("/v1/tasks/")
93}
94
95/// Who is calling, as established by [`auth`].
96#[derive(Clone)]
97pub enum Caller {
98    /// No authentication is configured (loopback development).
99    Open,
100    Admin,
101    /// The shared, restricted worker token (`--worker-token`).
102    SharedWorker,
103    /// A worker from the workers file, by token or client certificate.
104    Worker(Arc<WorkerIdentity>),
105    /// An application from the workers file (`[[client]]`): the execution
106    /// API for the workflows and actions it was granted.
107    Client(Arc<crate::identity::ClientIdentity>),
108}
109
110impl Caller {
111    fn identity(&self) -> Option<&WorkerIdentity> {
112        match self {
113            Caller::Worker(w) => Some(w),
114            _ => None,
115        }
116    }
117}
118
119fn forbidden(msg: impl Into<String>) -> ApiError {
120    ApiError(EngineError::new(403, "FORBIDDEN", msg))
121}
122
123/// A client identity may only act on the workflows and actions it was granted.
124fn allow(caller: &Caller, action: &str, workflow: &str) -> ApiResult<()> {
125    if let Caller::Client(c) = caller {
126        if !c.allows(action, workflow) {
127            return Err(forbidden(format!(
128                "client `{}` may not {action} executions of `{workflow}`",
129                c.id
130            )));
131        }
132    }
133    Ok(())
134}
135
136/// A client may read a workflow definition only if it may read its executions.
137fn allow_definition(caller: &Caller, workflow: &str) -> ApiResult<()> {
138    if let Caller::Client(c) = caller {
139        if !c.can_read(workflow) {
140            return Err(forbidden(format!(
141                "client `{}` may not read the definition of `{workflow}`",
142                c.id
143            )));
144        }
145    }
146    Ok(())
147}
148
149/// `allow` for an existing execution, looked up by id.
150fn allow_execution(state: &Shared, caller: &Caller, action: &str, id: &str) -> ApiResult<()> {
151    if matches!(caller, Caller::Client(_)) {
152        let workflow = engine(state).execution(id)?.workflow;
153        allow(caller, action, &workflow)?;
154    }
155    Ok(())
156}
157
158/// Authentication.
159///
160/// When any credential is configured, every endpoint except `/v1/health`
161/// requires one. The admin token can call anything. Worker credentials (the
162/// shared `--worker-token`, a per-worker token from the workers file, or a
163/// client certificate under mutual TLS) can only call the worker protocol
164/// endpoints, so a compromised worker host cannot start, cancel or read
165/// executions. Per-worker identities are further limited by the handlers to
166/// their own id, tasks, sessions and task tokens.
167/// Establish the caller from a bearer token and/or a verified client
168/// certificate. Shared by the HTTP API and the gRPC binding. `Ok(None)` means
169/// no credential was presented while authentication is required.
170pub(crate) fn identify(
171    state: &AppState,
172    bearer: Option<&str>,
173    cert_der: Option<&[u8]>,
174) -> Result<Option<Caller>, ApiError> {
175    let secured = state.auth_token.is_some() || !state.identities.is_empty();
176    if !secured {
177        return Ok(Some(Caller::Open));
178    }
179    if let Some(t) = bearer {
180        if state
181            .auth_token
182            .as_ref()
183            .is_some_and(|a| constant_time_eq(t.as_bytes(), a.as_bytes()))
184        {
185            return Ok(Some(Caller::Admin));
186        }
187        if let Some(w) = state.identities.by_token(t) {
188            return Ok(Some(Caller::Worker(Arc::new(w.clone()))));
189        }
190        if let Some(c) = state.identities.client_by_token(t) {
191            return Ok(Some(Caller::Client(Arc::new(c.clone()))));
192        }
193        if state
194            .worker_token
195            .as_ref()
196            .is_some_and(|w| constant_time_eq(t.as_bytes(), w.as_bytes()))
197        {
198            return Ok(Some(Caller::SharedWorker));
199        }
200    }
201    if let Some(der) = cert_der {
202        if let Some(w) = state.identities.by_certificate(der) {
203            return Ok(Some(Caller::Worker(Arc::new(w.clone()))));
204        }
205        return match state.identities.client_by_certificate(der) {
206            Some(c) => Ok(Some(Caller::Client(Arc::new(c.clone())))),
207            None => Err(forbidden(
208                "the client certificate is valid but matches no worker or client in the workers file",
209            )),
210        };
211    }
212    Ok(None)
213}
214
215async fn auth(State(state): State<Shared>, mut req: Request, next: Next) -> Response {
216    let path = req.uri().path().to_string();
217    let bearer = req
218        .headers()
219        .get("authorization")
220        .and_then(|v| v.to_str().ok())
221        .and_then(|v| v.strip_prefix("Bearer "))
222        .map(str::to_string);
223    let cert = req
224        .extensions()
225        .get::<Arc<axum_server::tls::TlsInfo>>()
226        .and_then(|info| info.peer_certificates.first().cloned());
227    let caller = match identify(&state, bearer.as_deref(), cert.as_deref()) {
228        Ok(Some(c)) => c,
229        Ok(None) if path == "/v1/health" => Caller::Open,
230        Ok(None) => {
231            return ApiError(EngineError::new(
232                401,
233                "UNAUTHENTICATED",
234                "missing or invalid credentials",
235            ))
236            .into_response()
237        }
238        Err(e) => return e.into_response(),
239    };
240    if matches!(caller, Caller::SharedWorker | Caller::Worker(_))
241        && !is_worker_path(&path)
242        && path != "/v1/health"
243    {
244        return forbidden("worker credentials may only call /v1/workers/* and /v1/tasks/*")
245            .into_response();
246    }
247    if let Caller::Client(c) = &caller {
248        let admin_only = (path == "/v1/workflows" && req.method() == axum::http::Method::POST)
249            || path.starts_with("/v1/workers")
250            || path == "/v1/metrics";
251        if is_worker_path(&path) || admin_only {
252            return forbidden(format!(
253                "client `{}` may only use the execution API (start, signal, query, read, cancel)",
254                c.id
255            ))
256            .into_response();
257        }
258    }
259    req.extensions_mut().insert(caller);
260    let mut resp = next.run(req).await;
261    resp.headers_mut()
262        .insert("x-orbit-api-version", HeaderValue::from_static(API_VERSION));
263    resp
264}
265
266/// A per-worker identity may only use sessions it opened.
267pub(crate) fn authorize_session(state: &Shared, caller: &Caller, session: &str) -> ApiResult<()> {
268    if let Some(id) = caller.identity() {
269        if let Some(owner) = engine(state).session_worker(session) {
270            if owner != id.id {
271                return Err(forbidden(format!(
272                    "session belongs to worker `{owner}`, not `{}`",
273                    id.id
274                )));
275            }
276        }
277    }
278    Ok(())
279}
280
281/// A per-worker identity may only report results for tokens issued to it.
282pub(crate) fn authorize_token(state: &Shared, caller: &Caller, token: &str) -> ApiResult<()> {
283    if let Some(id) = caller.identity() {
284        if let Some(owner) = engine(state).token_worker(token) {
285            if owner != id.id {
286                return Err(forbidden(format!(
287                    "task token was issued to worker `{owner}`, not `{}`",
288                    id.id
289                )));
290            }
291        }
292    }
293    Ok(())
294}
295
296fn constant_time_eq(a: &[u8], b: &[u8]) -> bool {
297    a.len() == b.len() && a.iter().zip(b).fold(0u8, |acc, (x, y)| acc | (x ^ y)) == 0
298}
299
300async fn health(State(state): State<Shared>) -> Json<Value> {
301    let e = engine(&state);
302    Json(json!({
303        "status": "ok",
304        "api_version": API_VERSION,
305        "runtime_version": env!("CARGO_PKG_VERSION"),
306        "ir_versions": orbit_ir::SUPPORTED_IR_VERSIONS,
307        "active_executions": e.active_count(),
308        "worker_sessions": e.session_count(),
309    }))
310}
311
312async fn metrics(State(state): State<Shared>) -> Response {
313    let e = engine(&state);
314    let body = e.metrics.render(e.active_count(), e.session_count());
315    let mut headers = HeaderMap::new();
316    headers.insert(
317        "content-type",
318        HeaderValue::from_static("text/plain; version=0.0.4"),
319    );
320    (headers, body).into_response()
321}
322
323async fn register_workflow(State(state): State<Shared>, body: String) -> ApiResult<Response> {
324    let ir = orbit_ir::WorkflowIr::from_json(&body)
325        .map_err(|e| ApiError(EngineError::new(400, "INVALID_IR", e.0)))?;
326    let (name, version, hash) = (
327        ir.name.clone(),
328        ir.version.clone(),
329        ir.definition_hash.clone(),
330    );
331    let created = engine(&state).register_definition(ir)?;
332    let status = if created {
333        StatusCode::CREATED
334    } else {
335        StatusCode::OK
336    };
337    Ok((status, Json(json!({ "name": name, "version": version, "definition_hash": hash, "created": created }))).into_response())
338}
339
340fn summary(d: &orbit_ir::WorkflowIr) -> Value {
341    json!({
342        "name": d.name,
343        "version": d.version,
344        "description": d.description,
345        "definition_hash": d.definition_hash,
346        "steps": d.steps.iter().map(|s| s.id.clone()).collect::<Vec<_>>(),
347        "tasks": d.contracts.keys().collect::<Vec<_>>(),
348    })
349}
350
351async fn list_workflows(
352    State(state): State<Shared>,
353    Extension(caller): Extension<Caller>,
354) -> Json<Value> {
355    let defs = engine(&state).definitions();
356    let visible = |name: &str| match &caller {
357        Caller::Client(c) => c.can_read(name),
358        _ => true,
359    };
360    Json(
361        json!({ "workflows": defs.iter().filter(|d| visible(&d.name)).map(|d| summary(d)).collect::<Vec<_>>() }),
362    )
363}
364
365async fn get_workflow(
366    State(state): State<Shared>,
367    Extension(caller): Extension<Caller>,
368    Path((name, version)): Path<(String, String)>,
369) -> ApiResult<Json<Value>> {
370    allow_definition(&caller, &name)?;
371    let d = engine(&state)
372        .definition(&name, Some(&version))
373        .ok_or_else(|| {
374            ApiError(EngineError::new(
375                404,
376                "WORKFLOW_NOT_FOUND",
377                "workflow not found",
378            ))
379        })?;
380    Ok(Json(serde_json::to_value(&*d).unwrap_or_default()))
381}
382
383async fn get_workflow_latest(
384    State(state): State<Shared>,
385    Extension(caller): Extension<Caller>,
386    Path(name): Path<String>,
387) -> ApiResult<Json<Value>> {
388    allow_definition(&caller, &name)?;
389    let d = engine(&state).definition(&name, None).ok_or_else(|| {
390        ApiError(EngineError::new(
391            404,
392            "WORKFLOW_NOT_FOUND",
393            "workflow not found",
394        ))
395    })?;
396    Ok(Json(serde_json::to_value(&*d).unwrap_or_default()))
397}
398
399#[derive(Deserialize)]
400struct StartRequest {
401    workflow: String,
402    #[serde(default)]
403    version: Option<String>,
404    #[serde(default)]
405    input: Value,
406    #[serde(default)]
407    execution_id: Option<String>,
408}
409
410async fn start_execution(
411    State(state): State<Shared>,
412    Extension(caller): Extension<Caller>,
413    Json(req): Json<StartRequest>,
414) -> ApiResult<Response> {
415    allow(&caller, "start", &req.workflow)?;
416    let r = engine(&state).start(
417        &req.workflow,
418        req.version.as_deref(),
419        req.input,
420        req.execution_id,
421    )?;
422    let status = if r.created {
423        StatusCode::CREATED
424    } else {
425        StatusCode::OK
426    };
427    Ok((status, Json(serde_json::to_value(r).unwrap_or_default())).into_response())
428}
429
430#[derive(Deserialize)]
431struct ListQuery {
432    status: Option<String>,
433    workflow: Option<String>,
434    limit: Option<u32>,
435}
436
437async fn list_executions(
438    State(state): State<Shared>,
439    Extension(caller): Extension<Caller>,
440    Query(q): Query<ListQuery>,
441) -> ApiResult<Json<Value>> {
442    let mut list = engine(&state).list_executions(
443        q.status.as_deref().map(str::to_uppercase).as_deref(),
444        q.workflow.as_deref(),
445        q.limit.unwrap_or(50),
446    )?;
447    if let Caller::Client(c) = &caller {
448        list.retain(|e| c.can_read(&e.workflow));
449    }
450    Ok(Json(json!({ "executions": list })))
451}
452
453async fn get_execution(
454    State(state): State<Shared>,
455    Extension(caller): Extension<Caller>,
456    Path(id): Path<String>,
457) -> ApiResult<Json<Value>> {
458    allow_execution(&state, &caller, "read", &id)?;
459    let s = engine(&state).execution(&id)?;
460    Ok(Json(serde_json::to_value(s).unwrap_or_default()))
461}
462
463async fn get_history(
464    State(state): State<Shared>,
465    Extension(caller): Extension<Caller>,
466    Path(id): Path<String>,
467) -> ApiResult<Json<Value>> {
468    allow_execution(&state, &caller, "read", &id)?;
469    let h = engine(&state).history(&id)?;
470    Ok(Json(json!({ "execution_id": id, "events": h })))
471}
472
473#[derive(Deserialize, Default)]
474struct CancelRequest {
475    #[serde(default)]
476    reason: Option<String>,
477}
478
479async fn cancel_execution(
480    State(state): State<Shared>,
481    Extension(caller): Extension<Caller>,
482    Path(id): Path<String>,
483    body: Option<Json<CancelRequest>>,
484) -> ApiResult<Json<Value>> {
485    allow_execution(&state, &caller, "cancel", &id)?;
486    let reason = body
487        .and_then(|b| b.0.reason)
488        .unwrap_or_else(|| "requested via API".into());
489    let s = engine(&state).cancel(&id, &reason)?;
490    Ok(Json(serde_json::to_value(s).unwrap_or_default()))
491}
492
493/// `POST /v1/executions/{id}/signals/{name}` with the payload as the body
494/// (an empty body sends `{}`).
495async fn send_signal(
496    State(state): State<Shared>,
497    Extension(caller): Extension<Caller>,
498    Path((id, name)): Path<(String, String)>,
499    body: Option<Json<Value>>,
500) -> ApiResult<Json<Value>> {
501    allow_execution(&state, &caller, "signal", &id)?;
502    let payload = body.map(|b| b.0).unwrap_or_else(|| json!({}));
503    let seq = engine(&state).signal(&id, &name, payload)?;
504    Ok(Json(json!({
505        "execution_id": id,
506        "signal": name,
507        "accepted": true,
508        "seq": seq,
509    })))
510}
511
512/// `GET /v1/executions/{id}/queries/{name}`: evaluate a declared query.
513async fn run_query(
514    State(state): State<Shared>,
515    Extension(caller): Extension<Caller>,
516    Path((id, name)): Path<(String, String)>,
517) -> ApiResult<Json<Value>> {
518    allow_execution(&state, &caller, "query", &id)?;
519    let value = engine(&state).query(&id, &name)?;
520    Ok(Json(
521        json!({ "execution_id": id, "query": name, "value": value }),
522    ))
523}
524
525#[derive(Deserialize)]
526struct WaitQuery {
527    timeout_ms: Option<u64>,
528}
529
530/// Long-poll until the execution reaches a terminal status (or the timeout).
531async fn wait_execution(
532    State(state): State<Shared>,
533    Extension(caller): Extension<Caller>,
534    Path(id): Path<String>,
535    Query(q): Query<WaitQuery>,
536) -> ApiResult<Json<Value>> {
537    allow_execution(&state, &caller, "read", &id)?;
538    let deadline = tokio::time::Instant::now()
539        + Duration::from_millis(q.timeout_ms.unwrap_or(30_000).min(300_000));
540    loop {
541        let notified = state.changed.notified();
542        tokio::pin!(notified);
543        notified.as_mut().enable();
544        let s = engine(&state).execution(&id)?;
545        if s.status.is_terminal() || tokio::time::Instant::now() >= deadline {
546            return Ok(Json(serde_json::to_value(s).unwrap_or_default()));
547        }
548        let _ = tokio::time::timeout_at(deadline, notified).await;
549    }
550}
551
552async fn list_workers(State(state): State<Shared>) -> ApiResult<Json<Value>> {
553    Ok(Json(json!({ "workers": engine(&state).workers()? })))
554}
555
556async fn register_worker(
557    State(state): State<Shared>,
558    Extension(caller): Extension<Caller>,
559    Json(req): Json<RegisterWorker>,
560) -> ApiResult<Json<Value>> {
561    if let Some(id) = caller.identity() {
562        if req.worker_id != id.id {
563            return Err(forbidden(format!(
564                "this credential belongs to worker `{}`, not `{}`",
565                id.id, req.worker_id
566            )));
567        }
568        if let Some(rt) = &id.runtime {
569            if !rt.eq_ignore_ascii_case(&req.runtime) {
570                return Err(forbidden(format!(
571                    "worker `{}` must announce runtime `{rt}`",
572                    id.id
573                )));
574            }
575        }
576        let denied: Vec<&str> = req
577            .contracts
578            .iter()
579            .map(|c| c.task_name.as_str())
580            .filter(|t| !id.may_serve(t))
581            .collect();
582        if !denied.is_empty() {
583            return Err(forbidden(format!(
584                "worker `{}` may not serve: {} (allowed: {})",
585                id.id,
586                denied.join(", "),
587                id.tasks.join(", ")
588            )));
589        }
590    }
591    Ok(Json(
592        serde_json::to_value(engine(&state).register_worker(req)?).unwrap_or_default(),
593    ))
594}
595
596#[derive(Deserialize)]
597struct SessionRequest {
598    worker_session_id: String,
599    #[serde(default)]
600    wait_ms: Option<u64>,
601}
602
603async fn heartbeat(
604    State(state): State<Shared>,
605    Extension(caller): Extension<Caller>,
606    Json(req): Json<SessionRequest>,
607) -> ApiResult<Json<Value>> {
608    authorize_session(&state, &caller, &req.worker_session_id)?;
609    Ok(Json(
610        serde_json::to_value(engine(&state).heartbeat(&req.worker_session_id)?).unwrap_or_default(),
611    ))
612}
613
614async fn unregister(
615    State(state): State<Shared>,
616    Extension(caller): Extension<Caller>,
617    Json(req): Json<SessionRequest>,
618) -> ApiResult<Json<Value>> {
619    authorize_session(&state, &caller, &req.worker_session_id)?;
620    engine(&state).unregister_worker(&req.worker_session_id)?;
621    Ok(Json(json!({ "accepted": true })))
622}
623
624/// Long-poll for a task. Returns 204 when nothing arrived within `wait_ms`.
625async fn poll(
626    State(state): State<Shared>,
627    Extension(caller): Extension<Caller>,
628    Json(req): Json<SessionRequest>,
629) -> ApiResult<Response> {
630    authorize_session(&state, &caller, &req.worker_session_id)?;
631    let wait = Duration::from_millis(req.wait_ms.unwrap_or(20_000).min(60_000));
632    let deadline = tokio::time::Instant::now() + wait;
633    loop {
634        let notified = state.changed.notified();
635        tokio::pin!(notified);
636        notified.as_mut().enable();
637        let task: Option<TaskAssignment> = engine(&state).poll(&req.worker_session_id)?;
638        if let Some(t) = task {
639            return Ok(Json(serde_json::to_value(t).unwrap_or_default()).into_response());
640        }
641        if tokio::time::Instant::now() >= deadline {
642            return Ok(StatusCode::NO_CONTENT.into_response());
643        }
644        let _ = tokio::time::timeout_at(deadline, notified).await;
645    }
646}
647
648#[derive(Deserialize)]
649struct CompleteRequest {
650    token: String,
651    #[serde(default)]
652    output: Value,
653}
654
655async fn complete_task(
656    State(state): State<Shared>,
657    Extension(caller): Extension<Caller>,
658    Json(req): Json<CompleteRequest>,
659) -> ApiResult<Json<Value>> {
660    if req.token.is_empty() {
661        return Err(bad_request("token is required"));
662    }
663    authorize_token(&state, &caller, &req.token)?;
664    Ok(Json(
665        serde_json::to_value(engine(&state).complete(&req.token, req.output)?).unwrap_or_default(),
666    ))
667}
668
669#[derive(Deserialize)]
670struct FailRequest {
671    token: String,
672    error_code: String,
673    #[serde(default)]
674    error_message: String,
675    #[serde(default = "default_true")]
676    retryable: bool,
677    #[serde(default)]
678    details: Option<Value>,
679}
680
681fn default_true() -> bool {
682    true
683}
684
685async fn fail_task(
686    State(state): State<Shared>,
687    Extension(caller): Extension<Caller>,
688    Json(req): Json<FailRequest>,
689) -> ApiResult<Json<Value>> {
690    if req.error_code.is_empty() {
691        return Err(bad_request("error_code is required"));
692    }
693    authorize_token(&state, &caller, &req.token)?;
694    let error = TaskError {
695        code: req.error_code,
696        message: req.error_message,
697        retryable: req.retryable,
698        details: req.details,
699    };
700    Ok(Json(
701        serde_json::to_value(engine(&state).fail(&req.token, error)?).unwrap_or_default(),
702    ))
703}
704
705/// Background loop: timers, retries, deadlines, liveness and built-in tasks.
706pub async fn run_ticker(state: Shared, interval: Duration) {
707    let mut ticker = tokio::time::interval(interval);
708    ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
709    loop {
710        ticker.tick().await;
711        let jobs = match engine(&state).tick() {
712            Ok(j) => j,
713            Err(e) => {
714                tracing::error!(error = %e, "tick failed");
715                continue;
716            }
717        };
718        for job in jobs {
719            let st = state.clone();
720            tokio::spawn(async move {
721                let input = job.input.clone();
722                let timeout = job.timeout_ms;
723                let result = tokio::task::spawn_blocking(move || {
724                    crate::http_adapter::execute(&input, timeout)
725                })
726                .await
727                .unwrap_or_else(|e| Err(TaskError::new("INTERNAL", e.to_string(), true)));
728                let mut e = engine(&st);
729                let r = match result {
730                    Ok(out) => e.complete(&job.token, out),
731                    Err(err) => e.fail(&job.token, err),
732                };
733                if let Err(err) = r {
734                    tracing::error!(error = %err, "recording built-in task result failed");
735                }
736            });
737        }
738    }
739}
740
741/// PEM files for serving HTTPS directly.
742pub struct TlsConfig {
743    pub cert: std::path::PathBuf,
744    pub key: std::path::PathBuf,
745    /// Enables mutual TLS: client certificates signed by this CA are verified
746    /// and identify workers. Clients without a certificate can still connect
747    /// and authenticate with a token.
748    pub client_ca: Option<std::path::PathBuf>,
749}
750
751fn io_err(msg: String) -> std::io::Error {
752    std::io::Error::new(std::io::ErrorKind::InvalidInput, msg)
753}
754
755fn read_pem_certs(
756    path: &std::path::Path,
757) -> std::io::Result<Vec<rustls::pki_types::CertificateDer<'static>>> {
758    let pem =
759        std::fs::read(path).map_err(|e| io_err(format!("cannot read {}: {e}", path.display())))?;
760    let certs = rustls_pemfile::certs(&mut pem.as_slice())
761        .collect::<Result<Vec<_>, _>>()
762        .map_err(|e| io_err(format!("{}: {e}", path.display())))?;
763    if certs.is_empty() {
764        return Err(io_err(format!(
765            "{} contains no certificates",
766            path.display()
767        )));
768    }
769    Ok(certs)
770}
771
772/// Build the rustls server configuration (with client verification for mTLS).
773fn rustls_config(tls: &TlsConfig) -> std::io::Result<rustls::ServerConfig> {
774    let chain = read_pem_certs(&tls.cert)?;
775    let key_pem = std::fs::read(&tls.key)
776        .map_err(|e| io_err(format!("cannot read {}: {e}", tls.key.display())))?;
777    let key = rustls_pemfile::private_key(&mut key_pem.as_slice())
778        .map_err(|e| io_err(format!("{}: {e}", tls.key.display())))?
779        .ok_or_else(|| io_err(format!("{} contains no private key", tls.key.display())))?;
780    let provider = Arc::new(rustls::crypto::aws_lc_rs::default_provider());
781    let builder = rustls::ServerConfig::builder_with_provider(provider.clone())
782        .with_safe_default_protocol_versions()
783        .map_err(|e| io_err(e.to_string()))?;
784    let builder = match &tls.client_ca {
785        Some(ca) => {
786            let mut roots = rustls::RootCertStore::empty();
787            for c in read_pem_certs(ca)? {
788                roots
789                    .add(c)
790                    .map_err(|e| io_err(format!("{}: {e}", ca.display())))?;
791            }
792            let verifier = rustls::server::WebPkiClientVerifier::builder_with_provider(
793                Arc::new(roots),
794                provider,
795            )
796            .allow_unauthenticated()
797            .build()
798            .map_err(|e| io_err(format!("client CA: {e}")))?;
799            builder.with_client_cert_verifier(verifier)
800        }
801        None => builder.with_no_client_auth(),
802    };
803    let mut config = builder
804        .with_single_cert(chain, key)
805        .map_err(|e| io_err(format!("TLS certificate/key: {e}")))?;
806    config.alpn_protocols = vec![b"h2".to_vec(), b"http/1.1".to_vec()];
807    Ok(config)
808}
809
810pub struct ServerConfig {
811    pub addr: SocketAddr,
812    pub auth_token: Option<String>,
813    pub worker_token: Option<String>,
814    pub identities: Identities,
815    pub tls: Option<TlsConfig>,
816    /// Also serve the gRPC binding of the worker protocol on this address.
817    pub grpc_addr: Option<SocketAddr>,
818    /// Pre-bound listeners (socket activation, tests): used instead of
819    /// binding `addr` / `grpc_addr`.
820    pub listener: Option<std::net::TcpListener>,
821    pub grpc_listener: Option<std::net::TcpListener>,
822    pub body_limit: usize,
823    pub tick_interval: Duration,
824}
825
826pub async fn serve(
827    engine: Engine,
828    config: ServerConfig,
829    shutdown: impl std::future::Future<Output = ()> + Send + 'static,
830) -> std::io::Result<()> {
831    serve_node(engine, config, None, shutdown).await
832}
833
834/// [`serve`] as a member of a cluster (`orbit server --cluster`): `engine`
835/// is the node's standby engine ([`crate::cluster::Node::standby_engine`]);
836/// the node's election thread swaps in a leader engine once elected.
837pub async fn serve_node(
838    engine: Engine,
839    config: ServerConfig,
840    cluster: Option<Arc<crate::cluster::Node>>,
841    shutdown: impl std::future::Future<Output = ()> + Send + 'static,
842) -> std::io::Result<()> {
843    // Several rustls crypto backends are compiled in (via different
844    // dependencies); choose one before any TLS configuration is built.
845    let _ = rustls::crypto::aws_lc_rs::default_provider().install_default();
846    let changed = engine.changed.clone();
847    let state = Arc::new(AppState {
848        engine: Mutex::new(engine),
849        changed,
850        auth_token: config.auth_token,
851        worker_token: config.worker_token,
852        identities: config.identities,
853    });
854    let app = router(state.clone(), config.body_limit);
855    let app = crate::cluster::layer(app, cluster.clone(), state.clone());
856    if let Some(node) = &cluster {
857        crate::cluster::start(node.clone(), state.clone())?;
858    }
859    let ticker = tokio::spawn(run_ticker(state.clone(), config.tick_interval));
860    // One shutdown signal fanned out to every listener.
861    let (stop_tx, stop_rx) = tokio::sync::watch::channel(false);
862    tokio::spawn(async move {
863        shutdown.await;
864        let _ = stop_tx.send(true);
865    });
866    let stopped = |mut rx: tokio::sync::watch::Receiver<bool>| async move {
867        let _ = rx.wait_for(|v| *v).await;
868    };
869    let bind = |addr: SocketAddr| {
870        std::net::TcpListener::bind(addr)
871            .map_err(|e| std::io::Error::new(e.kind(), format!("cannot listen on {addr}: {e}")))
872    };
873    let grpc_listener = match (config.grpc_listener, config.grpc_addr) {
874        (Some(l), _) => Some(l),
875        (None, Some(addr)) => Some(bind(addr)?),
876        (None, None) => None,
877    };
878    let http_listener = match config.listener {
879        Some(l) => l,
880        None => bind(config.addr)?,
881    };
882    http_listener.set_nonblocking(true)?;
883    let http_addr = http_listener.local_addr()?;
884    let grpc = match grpc_listener {
885        Some(std_listener) => {
886            std_listener.set_nonblocking(true)?;
887            let addr = std_listener.local_addr()?;
888            let listener = tokio::net::TcpListener::from_std(std_listener)?;
889            let incoming =
890                tonic::transport::server::TcpIncoming::from_listener(listener, true, None)
891                    .map_err(|e| io_err(format!("gRPC listener: {e}")))?;
892            let mut builder = tonic::transport::Server::builder();
893            if let Some(tls) = &config.tls {
894                let _ = rustls::crypto::aws_lc_rs::default_provider().install_default();
895                let read = |p: &std::path::Path| {
896                    std::fs::read(p)
897                        .map_err(|e| io_err(format!("cannot read {}: {e}", p.display())))
898                };
899                let mut t = tonic::transport::ServerTlsConfig::new().identity(
900                    tonic::transport::Identity::from_pem(read(&tls.cert)?, read(&tls.key)?),
901                );
902                if let Some(ca) = &tls.client_ca {
903                    t = t
904                        .client_ca_root(tonic::transport::Certificate::from_pem(read(ca)?))
905                        .client_auth_optional(true);
906                }
907                builder = builder
908                    .tls_config(t)
909                    .map_err(|e| io_err(format!("gRPC TLS: {e}")))?;
910            }
911            let service = crate::grpc::GrpcWorkers::new(state.clone());
912            let rx = stop_rx.clone();
913            tracing::info!(addr = %addr, tls = config.tls.is_some(), "gRPC worker protocol listening");
914            let router = match &cluster {
915                Some(node) => {
916                    builder.add_service(tonic::service::interceptor::InterceptedService::new(
917                        service,
918                        crate::cluster::grpc_gate(node.clone()),
919                    ))
920                }
921                None => builder.add_service(service),
922            };
923            Some(tokio::spawn(async move {
924                if let Err(e) = router
925                    .serve_with_incoming_shutdown(incoming, stopped(rx))
926                    .await
927                {
928                    tracing::error!(error = %e, "gRPC server failed");
929                }
930            }))
931        }
932        None => None,
933    };
934    let shutdown = stopped(stop_rx);
935    let result = match config.tls {
936        None => {
937            let listener = tokio::net::TcpListener::from_std(http_listener)?;
938            tracing::info!(addr = %http_addr, tls = false, "ORBIT runtime listening");
939            axum::serve(listener, app)
940                .with_graceful_shutdown(shutdown)
941                .await
942        }
943        Some(tls) => {
944            // Several rustls crypto backends are compiled in (via different
945            // dependencies); pick one explicitly so the choice is not ambiguous.
946            let _ = rustls::crypto::aws_lc_rs::default_provider().install_default();
947            let rustls =
948                axum_server::tls_rustls::RustlsConfig::from_config(Arc::new(rustls_config(&tls)?));
949            let handle = axum_server::Handle::new();
950            let h = handle.clone();
951            tokio::spawn(async move {
952                shutdown.await;
953                h.graceful_shutdown(Some(Duration::from_secs(10)));
954            });
955            tracing::info!(addr = %http_addr, tls = true, mtls = tls.client_ca.is_some(), "ORBIT runtime listening");
956            axum_server::from_tcp_rustls(http_listener, rustls)?
957                .handle(handle)
958                .serve(app.into_make_service())
959                .await
960        }
961    };
962    ticker.abort();
963    if let Some(g) = grpc {
964        let _ = g.await;
965    }
966    result
967}