assay_workflow/
ctx.rs

1use std::sync::Arc;
2
3use tokio::task::JoinHandle;
4use tracing::info;
5
6use crate::dispatch_recovery;
7use crate::events::{WorkflowBusEvent, WorkflowEventBus};
8use crate::health;
9use crate::scheduler;
10use crate::store::WorkflowStore;
11use crate::timers;
12
13/// Holds the background-task JoinHandles. When the last Arc<BackgroundTasks>
14/// is dropped the tasks are abandoned (tokio will cancel them on shutdown).
15pub struct BackgroundTasks {
16    _scheduler: JoinHandle<()>,
17    _timer_poller: JoinHandle<()>,
18    _health_monitor: JoinHandle<()>,
19    _dispatch_recovery: JoinHandle<()>,
20    #[cfg(feature = "s3-archival")]
21    _archival: Option<JoinHandle<()>>,
22}
23
24/// The workflow context. Owns the store, background-task handles, and
25/// per-request config. Serves as both the orchestrator (all engine methods
26/// live as `impl WorkflowCtx<S>`) and the axum state (`Arc<WorkflowCtx<S>>`).
27///
28/// `S` is the concrete store backend (`SqliteStore` or `PostgresStore`).
29/// `WorkflowStore` uses RPIT futures and is not dyn-compatible, so the
30/// generic parameter is retained here to avoid boxing every async call.
31pub struct WorkflowCtx<S: WorkflowStore> {
32    pub(crate) store: Arc<S>,
33    /// Engine-wide CDC outbox. When wired, state-mutating methods
34    /// publish typed `WorkflowBusEvent` variants via `emit(...)`.
35    /// `None` for tests / embedders without a dashboard — emit becomes
36    /// a no-op.
37    pub(crate) bus: Option<WorkflowEventBus>,
38    pub(crate) _bg: Arc<BackgroundTasks>,
39    /// Version of the containing binary (e.g. the `assay-lua` CLI) — set
40    /// by embedders so `/api/v1/engine/workflow/version` reflects the user-facing
41    /// binary, not this internal engine-crate version.
42    pub binary_version: Option<&'static str>,
43}
44
45impl<S: WorkflowStore> WorkflowCtx<S> {
46    /// Start the context with all background tasks.
47    pub fn start(store: Arc<S>) -> Self {
48        let _scheduler = tokio::spawn(scheduler::run_scheduler(Arc::clone(&store)));
49        let _timer_poller = tokio::spawn(timers::run_timer_poller(Arc::clone(&store)));
50        let _health_monitor = tokio::spawn(health::run_health_monitor(Arc::clone(&store)));
51        let _dispatch_recovery =
52            tokio::spawn(dispatch_recovery::run_dispatch_recovery(Arc::clone(&store)));
53
54        #[cfg(feature = "s3-archival")]
55        let _archival = crate::archival::ArchivalConfig::from_env()
56            .map(|cfg| tokio::spawn(crate::archival::run_archival(Arc::clone(&store), cfg)));
57
58        info!("Workflow engine started");
59
60        Self {
61            store,
62            bus: None,
63            _bg: Arc::new(BackgroundTasks {
64                _scheduler,
65                _timer_poller,
66                _health_monitor,
67                _dispatch_recovery,
68                #[cfg(feature = "s3-archival")]
69                _archival,
70            }),
71            binary_version: None,
72        }
73    }
74
75    /// Attach the engine-wide event bus. The API layer sets this up so
76    /// the SSE stream (`/events/stream`) and the dispatch-wakeup loop
77    /// see state transitions as they happen. Returns the context by
78    /// value so callers can chain.
79    pub fn with_event_bus(mut self, bus: WorkflowEventBus) -> Self {
80        self.bus = Some(bus);
81        self
82    }
83
84    /// Set the binary version string surfaced by `/api/v1/engine/workflow/version`.
85    pub fn with_binary_version(mut self, version: &'static str) -> Self {
86        self.binary_version = Some(version);
87        self
88    }
89
90    /// Access the underlying store (for the API layer).
91    pub fn store(&self) -> &S {
92        &self.store
93    }
94
95    /// Access the event bus (for SSE + scheduler wake-up).
96    pub fn bus(&self) -> Option<&WorkflowEventBus> {
97        self.bus.as_ref()
98    }
99
100    /// Emit a typed workflow event. No-op when no bus is wired (tests,
101    /// embedders without a dashboard). Errors are logged, not returned —
102    /// an emission failure must not fail the state-mutating method that
103    /// triggered it (atomicity for the state change is the DB tx's job;
104    /// this is a notification fired *after* the row write).
105    pub(crate) async fn emit(&self, namespace: &str, ev: WorkflowBusEvent) {
106        if let Some(bus) = &self.bus
107            && let Err(e) = bus.publish(namespace, ev).await
108        {
109            tracing::warn!(?e, "engine event emit failed");
110        }
111    }
112
113    pub(crate) async fn emit_retry_requested(
114        &self,
115        namespace: &str,
116        workflow_id: &str,
117        activity_id: i64,
118        activity_seq: i32,
119    ) {
120        if let Some(bus) = &self.bus
121            && let Err(e) = bus
122                .publish_retry_requested(namespace, workflow_id, activity_id, activity_seq)
123                .await
124        {
125            tracing::warn!(?e, "engine retry event emit failed");
126        }
127    }
128
129    /// Mark a workflow dispatchable AND emit a `WorkflowNeedsDispatch`
130    /// on the bus so the dispatch-wakeup loop wakes workers
131    /// on this node / across the cluster. The extra SELECT is skipped
132    /// when no bus is wired.
133    pub(crate) async fn mark_and_emit_needs_dispatch(
134        &self,
135        workflow_id: &str,
136    ) -> anyhow::Result<()> {
137        self.store.mark_workflow_dispatchable(workflow_id).await?;
138        if self.bus.is_some()
139            && let Some(wf) = self.store.get_workflow(workflow_id).await?
140        {
141            self.emit(
142                &wf.namespace,
143                WorkflowBusEvent::WorkflowNeedsDispatch {
144                    workflow_id: workflow_id.to_string(),
145                    task_queue: wf.task_queue,
146                },
147            )
148            .await;
149        }
150        Ok(())
151    }
152}
153
154/// Strip a trailing `-continued-<digits>` from a workflow id so
155/// sequential continue-as-new calls don't pile up suffixes. Matches
156/// the pattern emitted by the default id-derivation path; returns the
157/// input unchanged if there's no such suffix.
158pub(crate) fn strip_continued_suffix(id: &str) -> &str {
159    if let Some(idx) = id.rfind("-continued-") {
160        let (head, tail) = id.split_at(idx);
161        let rest = &tail["-continued-".len()..];
162        if !rest.is_empty() && rest.chars().all(|c| c.is_ascii_digit()) {
163            return head;
164        }
165    }
166    id
167}
168
169pub(crate) fn timestamp_now() -> f64 {
170    std::time::SystemTime::now()
171        .duration_since(std::time::UNIX_EPOCH)
172        .unwrap()
173        .as_secs_f64()
174}
175
176/// WorkflowCtx version (the binary version pulled from Cargo at build time).
177/// Stamped into every workflow's search_attributes at start so operators
178/// can correlate runs to the engine release that executed them.
179pub(crate) const ENGINE_VERSION: &str = env!("CARGO_PKG_VERSION");
180
181/// Auto-stamp `assay_engine_version` into a workflow's search attributes.
182/// Returns `Some` JSON string for the caller to store in the record.
183///
184/// If the caller already supplied `assay_engine_version` in their patch,
185/// we leave their value alone (explicit override wins). Otherwise we
186/// backfill the running engine's version. Callers who supply no
187/// attributes at all get a single-key object with just the version.
188pub(crate) fn inject_engine_version(caller_attrs: Option<&str>) -> Option<String> {
189    let mut obj: serde_json::Map<String, serde_json::Value> = match caller_attrs {
190        Some(raw) => match serde_json::from_str::<serde_json::Value>(raw) {
191            Ok(serde_json::Value::Object(m)) => m,
192            Ok(other) => return Some(other.to_string()),
193            Err(_) => return Some(raw.to_string()),
194        },
195        None => serde_json::Map::new(),
196    };
197    obj.entry("assay_engine_version".to_string())
198        .or_insert_with(|| serde_json::Value::String(ENGINE_VERSION.to_string()));
199    Some(serde_json::Value::Object(obj).to_string())
200}
201
202#[cfg(test)]
203mod engine_version_stamp_tests {
204    use super::*;
205
206    #[test]
207    fn no_attrs_produces_single_key_object() {
208        let out = inject_engine_version(None).unwrap();
209        let v: serde_json::Value = serde_json::from_str(&out).unwrap();
210        assert_eq!(v["assay_engine_version"], ENGINE_VERSION);
211        assert_eq!(v.as_object().unwrap().len(), 1);
212    }
213
214    #[test]
215    fn existing_attrs_gain_the_version_field() {
216        let out = inject_engine_version(Some(r#"{"env":"prod","tenant":"acme"}"#)).unwrap();
217        let v: serde_json::Value = serde_json::from_str(&out).unwrap();
218        assert_eq!(v["env"], "prod");
219        assert_eq!(v["tenant"], "acme");
220        assert_eq!(v["assay_engine_version"], ENGINE_VERSION);
221    }
222
223    #[test]
224    fn caller_supplied_version_wins_on_conflict() {
225        let out = inject_engine_version(Some(r#"{"assay_engine_version":"0.0.1-test"}"#)).unwrap();
226        let v: serde_json::Value = serde_json::from_str(&out).unwrap();
227        assert_eq!(v["assay_engine_version"], "0.0.1-test");
228    }
229
230    #[test]
231    fn non_object_json_is_preserved_unchanged() {
232        let out = inject_engine_version(Some("[1, 2, 3]")).unwrap();
233        assert_eq!(out, "[1,2,3]");
234    }
235
236    #[test]
237    fn unparsable_json_is_preserved_unchanged() {
238        let out = inject_engine_version(Some("not json")).unwrap();
239        assert_eq!(out, "not json");
240    }
241}
assay_workflow/ctx.rs

assay_workflow/
ctx.rs
