Merge remote-tracking branch 'origin/work/T-674-backend-job-runner' into develop

This commit is contained in:
2026-10-01 12:33:46 +09:00
33 changed files with 4158 additions and 160 deletions
+7
View File
@@ -94,6 +94,13 @@ pub const BUILTIN_PROFILE_RESOURCES: &[BuiltinProfileResource] = &[
description: "Workspace orchestration and Worker control.",
imports: BASE_IMPORT,
},
BuiltinProfileResource {
selector: Some("builtin:backend-job"),
path: "profiles/backend-job.dcdl",
source: include_str!("../../../resources/profiles/backend-job.dcdl"),
description: "Backend-owned bounded Job Worker.",
imports: BASE_IMPORT,
},
BuiltinProfileResource {
selector: Some("builtin:memory-consolidation"),
path: "profiles/memory-consolidation.dcdl",
+38
View File
@@ -1005,6 +1005,18 @@ pub enum Event {
submission_request_id: String,
message: String,
},
/// Durable notification acceptance. The request identity is persisted with
/// the notification receipt before this event is emitted. Repeating the
/// same request id and exact payload returns the same acknowledgement
/// without appending the notification twice.
NotificationAccepted {
notification_request_id: String,
},
/// Correlated notification rejection before durable acceptance.
NotificationRejected {
notification_request_id: String,
message: String,
},
/// Revisioned FIFO replacement following enqueue, activation, cancel, or clear.
PendingSubmissionsChanged {
pending: PendingSubmissionsSnapshot,
@@ -2809,6 +2821,32 @@ mod tests {
}
}
#[test]
fn notification_receipt_events_roundtrip_with_request_identity() {
for event in [
Event::NotificationAccepted {
notification_request_id: "notification-1".into(),
},
Event::NotificationRejected {
notification_request_id: "notification-1".into(),
message: "invalid notification".into(),
},
] {
let json = serde_json::to_string(&event).unwrap();
let decoded: Event = serde_json::from_str(&json).unwrap();
match decoded {
Event::NotificationAccepted {
notification_request_id,
}
| Event::NotificationRejected {
notification_request_id,
..
} => assert_eq!(notification_request_id, "notification-1"),
other => panic!("expected notification receipt, got {other:?}"),
}
}
}
#[test]
fn event_error_format() {
let event = Event::Error {
+26
View File
@@ -631,6 +631,24 @@ impl SubscriptionWorkerWorkdirAttachment {
}
}
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[cfg_attr(feature = "json-schema", derive(schemars::JsonSchema))]
#[serde(deny_unknown_fields)]
pub struct SubscriptionWorkerJob {
pub job_id: String,
pub attempt_id: String,
pub purpose: String,
}
impl SubscriptionWorkerJob {
pub fn validate(&self) -> Result<(), SubscriptionValidationError> {
validate_identifier("job_id", &self.job_id, MAX_RESOURCE_ID_BYTES)?;
validate_identifier("attempt_id", &self.attempt_id, MAX_RESOURCE_ID_BYTES)?;
validate_identifier("job_purpose", &self.purpose, MAX_RESOURCE_ID_BYTES)
}
}
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[cfg_attr(feature = "json-schema", derive(schemars::JsonSchema))]
@@ -664,6 +682,10 @@ pub struct SubscriptionWorker {
pub display_name: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub profile: Option<String>,
/// Backend-owned Job binding. Runtime producers leave this unset; the
/// Workspace projection enriches both snapshots and live updates.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub job: Option<SubscriptionWorkerJob>,
#[serde(default, skip_serializing_if = "Vec::is_empty")]
pub workdir_attachments: Vec<SubscriptionWorkerWorkdirAttachment>,
}
@@ -677,6 +699,9 @@ impl SubscriptionWorker {
if let Some(resource_key) = &self.resource_key {
validate_identifier("resource_key", resource_key, MAX_RESOURCE_ID_BYTES)?;
}
if let Some(job) = &self.job {
job.validate()?;
}
let mut aliases = HashSet::new();
let mut workdir_ids = HashSet::new();
for attachment in &self.workdir_attachments {
@@ -952,6 +977,7 @@ mod tests {
workspace_id: Some("workspace-1".to_string()),
display_name: Some(format!("Worker {value}")),
profile: Some("builtin:coder".to_string()),
job: None,
workdir_attachments: Vec::new(),
}
}
+3 -1
View File
@@ -28,7 +28,7 @@ use crate::{
SubscriptionRequest, SubscriptionRequestId, SubscriptionResponse, SubscriptionSnapshot,
SubscriptionTerminationCode, SubscriptionWorkdirId, SubscriptionWorker,
SubscriptionWorkerAvailability, SubscriptionWorkerId, SubscriptionWorkerIds,
SubscriptionWorkerProtocolMethod, SubscriptionWorkerState,
SubscriptionWorkerJob, SubscriptionWorkerProtocolMethod, SubscriptionWorkerState,
SubscriptionWorkerWorkdirAttachment, WorkspaceSubscriptionWorkdir,
},
};
@@ -132,6 +132,7 @@ pub fn generated_protocol_types() -> String {
push_decl::<SubscriptionWorkerAvailability>(&cfg, &mut output);
push_decl::<EventSubscriptionSelector>(&cfg, &mut output);
push_decl::<SubscriptionWorkerWorkdirAttachment>(&cfg, &mut output);
push_decl::<SubscriptionWorkerJob>(&cfg, &mut output);
push_decl::<SubscriptionWorker>(&cfg, &mut output);
push_decl::<WorkspaceSubscriptionWorkdir>(&cfg, &mut output);
push_decl::<SubscriptionSnapshot>(&cfg, &mut output);
@@ -232,6 +233,7 @@ fn rust_serialized_subscription_frame_fixtures() -> Vec<SubscriptionFrame> {
workspace_id: Some("fixture-workspace".to_string()),
display_name: None,
profile: None,
job: None,
workdir_attachments: vec![SubscriptionWorkerWorkdirAttachment {
alias: "external".to_string(),
repository_id: None,
+3 -2
View File
@@ -1143,8 +1143,9 @@ impl App {
}
match event {
Event::SubmissionAccepted { .. } => {}
Event::SubmissionRejected { message, .. } => self.push_error(message),
Event::SubmissionAccepted { .. } | Event::NotificationAccepted { .. } => {}
Event::SubmissionRejected { message, .. }
| Event::NotificationRejected { message, .. } => self.push_error(message),
Event::PendingSubmissionsChanged { pending } => {
self.pending_submissions = pending;
}
+30
View File
@@ -66,6 +66,12 @@ pub struct WorkerSubmissionAck {
pub disposition: protocol::SubmissionDisposition,
}
/// Evidence that a notification request reached the durable Worker session boundary.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct WorkerNotificationAck {
pub notification_request_id: String,
}
/// Typed execution result class. Results are transient operation outcomes and
/// are not persisted as Worker lifecycle authority.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
@@ -78,6 +84,8 @@ pub struct WorkerExecutionResult {
pub message: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub submission: Option<WorkerSubmissionAck>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub notification: Option<WorkerNotificationAck>,
}
/// Backend result class for a Worker execution operation.
@@ -99,6 +107,7 @@ impl WorkerExecutionResult {
worker_state: None,
message: None,
submission: None,
notification: None,
}
}
@@ -123,6 +132,23 @@ impl WorkerExecutionResult {
submission_id: submission_id.into(),
disposition,
}),
notification: None,
}
}
pub fn accepted_notification(
operation: WorkerExecutionOperation,
notification_request_id: impl Into<String>,
) -> Self {
Self {
operation,
outcome: WorkerExecutionOutcome::Accepted,
worker_state: None,
message: None,
submission: None,
notification: Some(WorkerNotificationAck {
notification_request_id: notification_request_id.into(),
}),
}
}
@@ -133,6 +159,7 @@ impl WorkerExecutionResult {
worker_state: None,
message: Some(message.into()),
submission: None,
notification: None,
}
}
@@ -143,6 +170,7 @@ impl WorkerExecutionResult {
worker_state: None,
message: Some(message.into()),
submission: None,
notification: None,
}
}
@@ -153,6 +181,7 @@ impl WorkerExecutionResult {
worker_state: None,
message: Some(message.into()),
submission: None,
notification: None,
}
}
@@ -163,6 +192,7 @@ impl WorkerExecutionResult {
worker_state: None,
message: Some(message.into()),
submission: None,
notification: None,
}
}
+4 -1
View File
@@ -94,6 +94,9 @@ pub struct WorkerInteractionAck {
pub worker_ref: WorkerRef,
pub status: WorkerStatus,
/// Present for User Submit and absent for non-Submit interactions.
#[serde(skip_serializing_if = "Option::is_none")]
#[serde(default, skip_serializing_if = "Option::is_none")]
pub submission: Option<crate::execution::WorkerSubmissionAck>,
/// Present only after a Notify request reaches the Worker's durable receipt boundary.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub notification: Option<crate::execution::WorkerNotificationAck>,
}
+71 -10
View File
@@ -2104,6 +2104,12 @@ impl Runtime {
.clone()
.expect("User input request id must be assigned before dispatch")
});
let expected_notification_id = (input.kind == WorkerInputKind::Notify).then(|| {
input
.submission_request_id
.clone()
.expect("Notify request id must be assigned before dispatch")
});
self.ensure_worker_execution(worker_ref)?;
let (backend, handle) = {
let state = self.lock()?;
@@ -2160,7 +2166,28 @@ impl Runtime {
});
}
if let Some(expected_notification_id) = expected_notification_id
&& dispatch_result
.notification
.as_ref()
.is_none_or(|ack| ack.notification_request_id != expected_notification_id)
{
let result = WorkerExecutionResult::rejected(
WorkerExecutionOperation::Input,
"execution backend did not acknowledge the durable Runtime notification request id",
);
self.record_execution_result(worker_ref, result.clone())?;
return Err(RuntimeError::WorkerExecutionRejected {
worker_id: worker_ref.worker_id.clone(),
operation: result.operation,
outcome: result.outcome,
message: result.message_or_default(),
result,
});
}
let submission = dispatch_result.submission.clone();
let notification = dispatch_result.notification.clone();
let mut state = self.lock()?;
state.ensure_running()?;
let worker = state.worker_mut(worker_ref)?;
@@ -2180,6 +2207,7 @@ impl Runtime {
worker_ref: worker_ref.clone(),
status,
submission,
notification,
})
}
@@ -4180,6 +4208,7 @@ impl RuntimeState {
workspace_id: worker.workspace_id.clone(),
display_name: worker.display_name.clone(),
profile,
job: None,
workdir_attachments,
})
}
@@ -6450,7 +6479,8 @@ mod tests {
_handle: &WorkerExecutionHandle,
input: WorkerInput,
) -> WorkerExecutionResult {
let submission_id = input.submission_request_id.clone();
let request_id = input.submission_request_id.clone();
let input_kind = input.kind.clone();
self.dispatched_inputs.lock().unwrap().push(input);
let mut result = self
.dispatch_result
@@ -6458,20 +6488,28 @@ mod tests {
.unwrap()
.clone()
.unwrap_or_else(|| {
WorkerExecutionResult::accepted_submission(
WorkerExecutionOperation::Input,
"request-test",
"test-submission",
protocol::SubmissionDisposition::Started,
)
if input_kind == WorkerInputKind::Notify {
WorkerExecutionResult::accepted_notification(
WorkerExecutionOperation::Input,
request_id
.clone()
.expect("Runtime assigns Notify request identity before dispatch"),
)
} else {
WorkerExecutionResult::accepted_submission(
WorkerExecutionOperation::Input,
"request-test",
"test-submission",
protocol::SubmissionDisposition::Started,
)
}
});
if !self
.preserve_submission_acknowledgement_id
.load(Ordering::SeqCst)
&& let (Some(ack), Some(submission_id)) =
(result.submission.as_mut(), submission_id)
&& let (Some(ack), Some(request_id)) = (result.submission.as_mut(), request_id)
{
ack.submission_request_id = submission_id;
ack.submission_request_id = request_id;
}
result
}
@@ -7910,6 +7948,29 @@ mod tests {
assert!(runtime.list_workers().unwrap().is_empty());
}
#[test]
fn notify_rejects_backend_enqueue_without_durable_worker_receipt() {
let (runtime, backend) = runtime_and_backend();
backend.set_dispatch_result(WorkerExecutionResult::accepted(
WorkerExecutionOperation::Input,
));
let detail = runtime
.create_worker(task_request("notify receipt"))
.unwrap();
let mut input = WorkerInput::notify("do not acknowledge enqueue alone");
input.submission_request_id = Some("notification-request".to_string());
let error = runtime.send_input(&detail.worker_ref, input).unwrap_err();
assert!(matches!(
error,
RuntimeError::WorkerExecutionRejected {
outcome: crate::execution::WorkerExecutionOutcome::Rejected,
..
}
));
}
#[test]
fn connected_backend_busy_dispatch_is_typed_and_not_transcribed() {
let (runtime, backend) = runtime_and_backend();
+85 -3
View File
@@ -45,6 +45,11 @@ use protocol::{Event, Method, Segment, WorkerCommandEnvelope};
static NEXT_INTERNAL_COMMAND_ID: AtomicU64 = AtomicU64::new(1);
enum NotificationAcceptance {
Accepted,
Rejected(String),
}
fn rollback_materialized_spawn_workdirs(
materializer: &dyn WorkingDirectoryMaterializer,
bindings: &BTreeMap<WorkdirAttachmentAlias, WorkingDirectoryBinding>,
@@ -1883,6 +1888,75 @@ where
.unwrap_or_else(|message| WorkerExecutionResult::errored(operation, message))
}
fn send_notification_and_wait_for_acceptance(
&self,
operation: WorkerExecutionOperation,
worker: WorkerHandle,
method: Method,
notification_request_id: String,
) -> WorkerExecutionResult {
let request_id = notification_request_id.clone();
let result = self.run_cancellable_on_adapter_runtime(USER_INPUT_TASK_TIMEOUT, async move {
// Subscribe before enqueueing so a fast durable acceptance cannot
// race the Runtime acknowledgement.
let mut events = worker.subscribe();
worker
.send(method)
.await
.map_err(|err| format!("failed to send Worker method: {err}"))?;
tokio::time::timeout(USER_INPUT_COMMIT_TIMEOUT, async move {
loop {
match events.recv().await {
Ok(Event::NotificationAccepted {
notification_request_id,
}) if notification_request_id == request_id => {
return Ok(NotificationAcceptance::Accepted);
}
Ok(Event::NotificationRejected {
notification_request_id,
message,
}) if notification_request_id == request_id => {
return Ok(NotificationAcceptance::Rejected(message));
}
Ok(Event::Shutdown) => {
return Err(
"worker shut down before notification was durably accepted"
.to_string(),
);
}
Ok(_) => {}
Err(tokio::sync::broadcast::error::RecvError::Lagged(skipped)) => {
return Err(format!(
"worker notification acknowledgement lagged by {skipped} protocol event(s)"
));
}
Err(tokio::sync::broadcast::error::RecvError::Closed) => {
return Err(
"worker event stream closed before notification was durably accepted"
.to_string(),
);
}
}
}
})
.await
.map_err(|_| {
"timed out waiting for durable Worker notification acceptance".to_string()
})?
});
match result {
Ok(NotificationAcceptance::Accepted) => {
WorkerExecutionResult::accepted_notification(operation, notification_request_id)
}
Ok(NotificationAcceptance::Rejected(message)) => {
WorkerExecutionResult::rejected(operation, message)
}
Err(message) => WorkerExecutionResult::errored(operation, message),
}
}
fn retain_uncertain_unconnected_controller(
&self,
worker_ref: &crate::identity::WorkerRef,
@@ -2879,16 +2953,17 @@ where
let notification_request_id = input
.submission_request_id
.unwrap_or_else(protocol::new_submission_request_id);
return self.send_method(
return self.send_notification_and_wait_for_acceptance(
WorkerExecutionOperation::Input,
worker,
Method::NotifyTracked {
notification_request_id: notification_request_id.clone(),
message: input.content,
source: protocol::AuthenticatedInputSource::Backend {
operation_id: notification_request_id,
operation_id: notification_request_id.clone(),
},
},
notification_request_id,
);
}
@@ -5108,7 +5183,7 @@ mod tests {
}
#[test]
fn running_worker_accepts_notify_without_a_submit_receipt() {
fn running_worker_waits_for_durable_notify_receipt() {
let client = MockClient::sequential(vec![MockResponse::Hang(vec![])]);
let runtime_base = tempfile::tempdir().unwrap();
let cwd = tempfile::tempdir().unwrap();
@@ -5143,6 +5218,13 @@ mod tests {
.expect("Running Worker must accept Notify without a Submit receipt");
assert!(acknowledgement.submission.is_none());
assert_eq!(
acknowledgement
.notification
.as_ref()
.map(|ack| ack.notification_request_id.as_str()),
Some("notification-request")
);
wait_for_adapter_state(&backend, &detail.worker_ref, WorkerStatus::Running);
}
+156 -113
View File
@@ -1178,39 +1178,42 @@ where
),
None => durable_parent_notifications,
};
let backend_job_profile = worker.manifest().engine.instruction == "internal.backend_job_system";
let prompts = worker.prompts().clone();
let paste_store = worker.store().clone();
let paste_session_id = worker.session_id();
worker
.engine_mut()
.register_tool(crate::paste_artifact_tool::search_input_artifact_tool(
paste_store.clone(),
paste_session_id,
));
worker
.engine_mut()
.register_tool(crate::paste_artifact_tool::read_input_artifact_tool(
paste_store,
paste_session_id,
));
// Keep the alias-routed schemas installed for the Worker lifetime. The
// attachment set may be empty or mutate after registration, so target
// resolution and provider capability checks happen per invocation.
let tracker = tools::Tracker::new();
let workdir_tools = if restrict_workdir_tools_to_current_capabilities {
tools::routed_builtin_tools_for_current_capabilities(
workdir_sessions.clone(),
tracker.clone(),
bash_output_dir.clone(),
)
} else {
tools::routed_builtin_tools(
workdir_sessions.clone(),
tracker.clone(),
bash_output_dir.clone(),
)
};
worker.engine_mut().register_tools(workdir_tools);
if !backend_job_profile {
let paste_store = worker.store().clone();
let paste_session_id = worker.session_id();
worker
.engine_mut()
.register_tool(crate::paste_artifact_tool::search_input_artifact_tool(
paste_store.clone(),
paste_session_id,
));
worker
.engine_mut()
.register_tool(crate::paste_artifact_tool::read_input_artifact_tool(
paste_store,
paste_session_id,
));
// Keep the alias-routed schemas installed for the Worker lifetime. The
// attachment set may be empty or mutate after registration, so target
// resolution and provider capability checks happen per invocation.
let workdir_tools = if restrict_workdir_tools_to_current_capabilities {
tools::routed_builtin_tools_for_current_capabilities(
workdir_sessions.clone(),
tracker.clone(),
bash_output_dir.clone(),
)
} else {
tools::routed_builtin_tools(
workdir_sessions.clone(),
tracker.clone(),
bash_output_dir.clone(),
)
};
worker.engine_mut().register_tools(workdir_tools);
}
if feature_config.image.enabled && model_supports_image_attachments(&spawner_manifest.model) {
worker
.engine_mut()
@@ -1225,6 +1228,20 @@ where
let worker_enabled = feature_config.worker.enabled;
let sub_worker_enabled = feature_config.sub_worker.enabled;
let mut feature_registry = FeatureRegistryBuilder::new();
if backend_job_profile {
let workspace_client = worker.workspace_client_handle();
if !workspace_client.is_available() || workspace_client.workspace_id().is_none() {
return Err(std::io::Error::new(
std::io::ErrorKind::InvalidInput,
"Backend Job result capability requires Backend Workspace API authority",
));
}
feature_registry.add_module(
crate::feature::builtin::backend_job_result::BackendJobResultFeature::new(
workspace_client,
),
);
}
let memory_install_plan = crate::feature::builtin::memory::MemoryFeatureInstallPlan::prepare(
worker.manifest(),
worker.workspace_client_handle(),
@@ -1472,6 +1489,40 @@ where
Ok(workdir_for_view)
}
fn durably_accept_notification<St>(
notification_request_id: String,
message: String,
source_namespace: String,
provenance: session_store::LoggedSessionHistoryOrigin,
pending_submissions: &PendingSubmissionHandle<St>,
working_event_tx: &broadcast::Sender<Event>,
) -> bool
where
St: session_store::Store + Clone,
{
let request_id = notification_request_id.clone();
match pending_submissions.accept_notification_from_source(
notification_request_id,
message,
source_namespace,
provenance,
) {
Ok(_) => {
let _ = working_event_tx.send(Event::NotificationAccepted {
notification_request_id: request_id,
});
true
}
Err(error) => {
let _ = working_event_tx.send(Event::NotificationRejected {
notification_request_id: request_id,
message: error.to_string(),
});
false
}
}
}
fn durably_accept_method_while_busy<St>(
method: Method,
pending_submissions: &PendingSubmissionHandle<St>,
@@ -1544,17 +1595,14 @@ where
notification_request_id,
message,
} => {
if let Err(error) = pending_submissions.accept_notification_from_source(
durably_accept_notification(
notification_request_id,
message,
pending_submissions.direct_client_namespace(),
session_store::LoggedSessionHistoryOrigin::LegacyUnknown,
) {
let _ = working_event_tx.send(Event::Error {
code: ErrorCode::InvalidRequest,
message: error.to_string(),
});
}
pending_submissions,
working_event_tx,
);
None
}
Method::NotifyTracked {
@@ -1564,17 +1612,14 @@ where
} => {
let (source_namespace, provenance) =
resolved_input_source(pending_submissions, &source);
if let Err(error) = pending_submissions.accept_notification_from_source(
durably_accept_notification(
notification_request_id,
message,
source_namespace,
provenance,
) {
let _ = working_event_tx.send(Event::Error {
code: ErrorCode::InvalidRequest,
message: error.to_string(),
});
}
pending_submissions,
working_event_tx,
);
None
}
method => Some(method),
@@ -1602,9 +1647,16 @@ fn reject_method_while_attention_locked(
});
true
}
Method::Notify { .. } | Method::NotifyTracked { .. } => {
let _ = working_event_tx.send(Event::Error {
code: ErrorCode::InvalidRequest,
Method::Notify {
notification_request_id,
..
}
| Method::NotifyTracked {
notification_request_id,
..
} => {
let _ = working_event_tx.send(Event::NotificationRejected {
notification_request_id: notification_request_id.clone(),
message: message.to_owned(),
});
true
@@ -2014,21 +2066,14 @@ async fn controller_loop<C, St>(
notification_request_id,
message,
} => {
let source_namespace = pending_submissions.direct_client_namespace();
match pending_submissions.accept_notification_from_source(
durably_accept_notification(
notification_request_id,
message,
source_namespace.clone(),
pending_submissions.direct_client_namespace(),
session_store::LoggedSessionHistoryOrigin::LegacyUnknown,
) {
Ok(_) => {}
Err(error) => {
let _ = working_event_tx.send(Event::Error {
code: ErrorCode::InvalidRequest,
message: error.to_string(),
});
}
}
&pending_submissions,
&working_event_tx,
);
}
Method::NotifyTracked {
@@ -2038,20 +2083,14 @@ async fn controller_loop<C, St>(
} => {
let (source_namespace, provenance) =
resolved_input_source(&pending_submissions, &source);
match pending_submissions.accept_notification_from_source(
durably_accept_notification(
notification_request_id,
message,
source_namespace.clone(),
source_namespace,
provenance,
) {
Ok(_) => {}
Err(error) => {
let _ = working_event_tx.send(Event::Error {
code: ErrorCode::InvalidRequest,
message: error.to_string(),
});
}
}
&pending_submissions,
&working_event_tx,
);
}
Method::ListPendingSubmissions => {
@@ -3096,22 +3135,15 @@ where
notification_request_id,
message,
}) => {
let source_namespace = pending_submissions.direct_client_namespace();
match pending_submissions.accept_notification_from_source(
if durably_accept_notification(
notification_request_id,
message,
source_namespace.clone(),
pending_submissions.direct_client_namespace(),
session_store::LoggedSessionHistoryOrigin::LegacyUnknown,
pending_submissions,
working_event_tx,
) {
Ok(_) => {
stage_pending_notifications(&pending_submissions, notify_buffer);
}
Err(error) => {
let _ = working_event_tx.send(Event::Error {
code: ErrorCode::InvalidRequest,
message: error.to_string(),
});
}
stage_pending_notifications(&pending_submissions, notify_buffer);
}
let _ = working_event_tx.send(Event::PendingSubmissionsChanged {
pending: pending_submissions.snapshot(),
@@ -3124,21 +3156,15 @@ where
}) => {
let (source_namespace, provenance) =
resolved_input_source(pending_submissions, &source);
match pending_submissions.accept_notification_from_source(
if durably_accept_notification(
notification_request_id,
message,
source_namespace.clone(),
source_namespace,
provenance,
pending_submissions,
working_event_tx,
) {
Ok(_) => {
stage_pending_notifications(&pending_submissions, notify_buffer);
}
Err(error) => {
let _ = working_event_tx.send(Event::Error {
code: ErrorCode::InvalidRequest,
message: error.to_string(),
});
}
stage_pending_notifications(&pending_submissions, notify_buffer);
}
let _ = working_event_tx.send(Event::PendingSubmissionsChanged {
pending: pending_submissions.snapshot(),
@@ -3605,29 +3631,46 @@ mod tests {
}
#[test]
fn busy_notification_is_durable_before_shutdown_can_observe_it() {
fn busy_notification_receipt_follows_durable_idempotent_acceptance() {
let dir = tempfile::tempdir().unwrap();
let pending = PendingSubmissionHandle::for_test(dir.path());
let (event_tx, _event_rx) = broadcast::channel(4);
let (event_tx, mut event_rx) = broadcast::channel(4);
let method = Method::NotifyTracked {
notification_request_id: "completion-1".into(),
message: "subworker completed".into(),
source: protocol::AuthenticatedInputSource::UntrustedWire,
};
assert!(
durably_accept_method_while_busy(
Method::NotifyTracked {
notification_request_id: "completion-1".into(),
message: "subworker completed".into(),
source: protocol::AuthenticatedInputSource::UntrustedWire,
},
&pending,
&event_tx,
)
.is_none()
for _ in 0..2 {
assert!(
durably_accept_method_while_busy(method.clone(), &pending, &event_tx).is_none()
);
let entries = pending.persisted_entries_for_test();
assert!(entries.iter().any(|entry| matches!(
entry,
LogEntry::Extension { domain, .. }
if domain == "worker.pending_activations.v1"
)));
assert!(matches!(
event_rx.try_recv().unwrap(),
Event::NotificationAccepted {
notification_request_id,
} if notification_request_id == "completion-1"
));
}
assert_eq!(pending.snapshot().notification_count, 1);
assert_eq!(
pending
.persisted_entries_for_test()
.iter()
.filter(|entry| matches!(
entry,
LogEntry::Extension { domain, .. }
if domain == "worker.pending_activations.v1"
))
.count(),
1
);
let entries = pending.persisted_entries_for_test();
assert!(entries.iter().any(|entry| matches!(
entry,
LogEntry::Extension { domain, .. }
if domain == "worker.pending_activations.v1"
)));
}
#[test]
+1
View File
@@ -4,6 +4,7 @@
//! same descriptor-approved registry path used by feature modules. They are not
//! an external plugin-loading surface.
pub mod backend_job_result;
pub mod flow_transition;
pub mod manage_workdir;
pub mod manage_worker;
@@ -0,0 +1,175 @@
//! Structured completion capability for Backend-owned Job Workers.
//!
//! The tool intentionally accepts only the result value. Workspace and Worker
//! identity come from the authenticated Workspace client, while Job, attempt,
//! and input revision are resolved from the Backend-owned binding.
use std::sync::Arc;
use agen::tool::{Tool, ToolError, ToolExecutionContext, ToolMeta, ToolOutput};
use async_trait::async_trait;
use schemars::JsonSchema;
use serde::Deserialize;
use serde_json::Value;
use crate::feature::{
FeatureDescriptor, FeatureInstallContext, FeatureInstallError, FeatureModule, ToolContribution,
ToolDeclaration, ToolDefinition,
};
use crate::worker::{WorkspaceClient, WorkspaceRequest, WorkspaceRequestMethod};
const TOOL_NAME: &str = "SubmitBackendJobResult";
const TOOL_DESCRIPTION: &str = "Submit the bounded structured result for this Backend-owned Job attempt. Workspace, Job, attempt, input revision, and source Worker identity are resolved and fenced by Backend authority; normal final prose is not success.";
#[derive(Debug, Deserialize, JsonSchema)]
#[serde(deny_unknown_fields)]
struct SubmitResultInput {
result: Value,
}
#[derive(Clone)]
struct SubmitResultTool {
client: Arc<dyn WorkspaceClient>,
}
#[async_trait]
impl Tool for SubmitResultTool {
async fn execute(
&self,
input_json: &str,
_context: ToolExecutionContext,
) -> Result<ToolOutput, ToolError> {
let input: SubmitResultInput = serde_json::from_str(input_json)
.map_err(|error| ToolError::InvalidArgument(error.to_string()))?;
let workspace_id = self.client.workspace_id().ok_or_else(|| {
ToolError::ExecutionFailed(
"Backend Job result submission requires Workspace identity".to_string(),
)
})?;
let response = self
.client
.execute(WorkspaceRequest::json(
WorkspaceRequestMethod::Post,
format!("/api/w/{workspace_id}/workers/self/backend-job-result"),
serde_json::to_string(&serde_json::json!({ "result": input.result }))
.map_err(|error| ToolError::Internal(error.to_string()))?,
))
.map_err(|error| ToolError::ExecutionFailed(error.to_string()))?;
if !response.is_success() {
return Err(ToolError::ExecutionFailed(format!(
"Backend Job result submission was rejected with HTTP {}: {}",
response.status, response.body
)));
}
Ok(ToolOutput {
summary: "Backend Job result accepted".to_string(),
content: Some(response.body),
attachments: Vec::new(),
})
}
}
fn definition(client: Arc<dyn WorkspaceClient>) -> ToolDefinition {
let schema = serde_json::to_value(schemars::schema_for!(SubmitResultInput))
.expect("Backend Job result schema must serialize");
Arc::new(move || {
(
ToolMeta::new(TOOL_NAME)
.description(TOOL_DESCRIPTION)
.input_schema(schema.clone()),
Arc::new(SubmitResultTool {
client: client.clone(),
}) as Arc<dyn Tool>,
)
})
}
#[derive(Clone)]
pub struct BackendJobResultFeature {
client: Arc<dyn WorkspaceClient>,
}
impl BackendJobResultFeature {
pub fn new(client: Arc<dyn WorkspaceClient>) -> Self {
Self { client }
}
}
impl FeatureModule for BackendJobResultFeature {
fn descriptor(&self) -> FeatureDescriptor {
FeatureDescriptor::builtin("backend-job-result", "Backend Job result")
.with_description("Single-purpose structured result capability for Backend-owned Jobs.")
.with_tool(ToolDeclaration::new(TOOL_NAME, TOOL_DESCRIPTION))
}
fn install(&self, context: &mut FeatureInstallContext<'_>) -> Result<(), FeatureInstallError> {
context.tools().register(ToolContribution::new(
TOOL_NAME,
definition(self.client.clone()),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::feature::FeatureRegistryBuilder;
use crate::hook::HookRegistryBuilder;
use crate::worker::{WorkspaceClientError, WorkspaceResponse};
use std::sync::Mutex;
#[derive(Debug, Default)]
struct RecordingClient {
requests: Mutex<Vec<WorkspaceRequest>>,
}
impl WorkspaceClient for RecordingClient {
fn workspace_id(&self) -> Option<&str> {
Some("workspace-test")
}
fn kind(&self) -> &str {
"test"
}
fn is_available(&self) -> bool {
true
}
fn execute(
&self,
request: WorkspaceRequest,
) -> Result<WorkspaceResponse, WorkspaceClientError> {
self.requests.lock().unwrap().push(request);
Ok(WorkspaceResponse {
status: 200,
body: "{\"replayed\":false}".to_string(),
})
}
}
#[tokio::test]
async fn feature_exposes_only_identity_free_structured_result_input() {
let client = Arc::new(RecordingClient::default());
let mut pending = Vec::new();
let mut hooks = HookRegistryBuilder::default();
let report = FeatureRegistryBuilder::new()
.with_module(BackendJobResultFeature::new(client.clone()))
.install_into_pending(&mut pending, &mut hooks);
assert_eq!(report.installed_tool_names(), [TOOL_NAME]);
let (meta, tool) = pending.remove(0)();
assert_eq!(meta.input_schema["required"], serde_json::json!(["result"]));
assert!(meta.input_schema["properties"].get("job_id").is_none());
tool.execute(
r#"{"result":{"valid":true}}"#,
ToolExecutionContext::direct(),
)
.await
.unwrap();
let requests = client.requests.lock().unwrap();
assert_eq!(requests.len(), 1);
assert_eq!(requests[0].method, WorkspaceRequestMethod::Post);
assert!(
requests[0]
.path
.ends_with("/workers/self/backend-job-result")
);
}
}
+3
View File
@@ -95,6 +95,7 @@ pub enum WorkerPrompt {
MemoryExtractSystem,
MemoryConsolidationSystem,
FlowVerifierSystem,
BackendJobSystem,
NotifyWrapper,
InterruptToolResultSummary,
InterruptSystemNote,
@@ -112,6 +113,7 @@ impl WorkerPrompt {
Self::MemoryExtractSystem => "internal.memory_extract_system",
Self::MemoryConsolidationSystem => "internal.memory_consolidation_system",
Self::FlowVerifierSystem => "internal.flow_verifier_system",
Self::BackendJobSystem => "internal.backend_job_system",
Self::NotifyWrapper => "internal.notify_wrapper",
Self::InterruptToolResultSummary => "internal.interrupt_tool_result_summary",
Self::InterruptSystemNote => "internal.interrupt_system_note",
@@ -130,6 +132,7 @@ impl WorkerPrompt {
WorkerPrompt::MemoryExtractSystem,
WorkerPrompt::MemoryConsolidationSystem,
WorkerPrompt::FlowVerifierSystem,
WorkerPrompt::BackendJobSystem,
WorkerPrompt::NotifyWrapper,
WorkerPrompt::InterruptToolResultSummary,
WorkerPrompt::InterruptSystemNote,
+33
View File
@@ -837,6 +837,39 @@ async fn feature_flags_default_to_core_tool_surface_only() {
assert!(!names.iter().any(|name| name == "SubWorkerSpawn"));
}
#[tokio::test]
async fn backend_job_profile_exposes_only_structured_result_capability() {
let workspace = tempfile::tempdir().unwrap();
let resolved = ProfileResolver::new()
.with_workspace_base(workspace.path())
.resolve(
&ProfileSelector::source_named(ProfileRegistrySource::Builtin, "backend-job"),
ProfileResolveOptions::with_worker_name("backend-job-worker"),
)
.unwrap();
let client = MockClient::new(simple_text_events());
let client_for_assert = client.clone();
let (worker, _pwd) = make_worker_with_manifest_and_workspace_context(
client,
resolved.manifest,
WorkerWorkspaceContext::with_client(None, Arc::new(AvailableWorkspaceClient)),
)
.await;
let handle = spawn_controller(worker).await;
handle
.send(Method::submit_text(
protocol::new_submission_request_id(),
"Run the bounded Backend Job.",
))
.await
.unwrap();
wait_for_status(&handle, WorkerStatus::Idle).await;
let request = wait_for_captured_request(&client_for_assert).await;
assert_eq!(request_tool_names(&request), vec!["SubmitBackendJobResult"]);
}
#[tokio::test]
async fn enabled_task_and_web_features_register_their_tools() {
let manifest = r#"
+538
View File
@@ -0,0 +1,538 @@
//! Durable Backend-owned jobs executed by ordinary Runtime-managed Workers.
//!
//! A job is durable intent. An attempt binds that intent to one dedicated
//! Worker, and a result is accepted only through the fenced structured-result
//! API below. Runtime lifecycle and final prose are deliberately not success
//! authority.
use serde::{Deserialize, Serialize};
use serde_json::Value;
use sha2::{Digest, Sha256};
use worker_runtime::identity::RuntimeWorkerRef;
use crate::{Error, Result};
pub const BACKEND_JOB_RUNTIME_ID: &str = crate::hosts::EMBEDDED_RUNTIME_ID;
pub const DEFAULT_MAX_CONCURRENT_JOBS: u16 = 8;
pub const ABSOLUTE_MAX_CONCURRENT_JOBS: u16 = 64;
pub const DEFAULT_JOB_TIMEOUT_SECONDS: u32 = 120;
pub const MAX_JOB_TIMEOUT_SECONDS: u32 = 600;
pub const DEFAULT_MAX_RESULT_BYTES: u32 = 16 * 1024;
pub const ABSOLUTE_MAX_RESULT_BYTES: u32 = 64 * 1024;
pub const DEFAULT_MAX_ATTEMPTS: u8 = 2;
pub const ABSOLUTE_MAX_ATTEMPTS: u8 = 3;
pub const MAX_JOB_INPUT_BYTES: usize = 128 * 1024;
pub const MAX_JOB_INSTRUCTION_BYTES: usize = 16 * 1024;
pub const MAX_JOB_PURPOSE_BYTES: usize = 120;
pub const MAX_JOB_REFERENCE_BYTES: usize = 512;
pub const MAX_JOB_FAILURE_BYTES: usize = 1024;
pub const MAX_JOB_DELIVERY_BYTES: usize = 16 * 1024;
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum BackendJobState {
Pending,
Completed,
Failed,
Unknown,
}
impl BackendJobState {
pub(crate) fn as_str(self) -> &'static str {
match self {
Self::Pending => "pending",
Self::Completed => "completed",
Self::Failed => "failed",
Self::Unknown => "unknown",
}
}
pub(crate) fn parse(value: &str) -> Result<Self> {
match value {
"pending" => Ok(Self::Pending),
"completed" => Ok(Self::Completed),
"failed" => Ok(Self::Failed),
"unknown" => Ok(Self::Unknown),
other => Err(Error::Store(format!("unknown Backend Job state `{other}`"))),
}
}
}
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum BackendJobAttemptState {
Reserved,
Dispatched,
Completed,
Failed,
Unknown,
}
impl BackendJobAttemptState {
pub(crate) fn as_str(self) -> &'static str {
match self {
Self::Reserved => "reserved",
Self::Dispatched => "dispatched",
Self::Completed => "completed",
Self::Failed => "failed",
Self::Unknown => "unknown",
}
}
pub(crate) fn parse(value: &str) -> Result<Self> {
match value {
"reserved" => Ok(Self::Reserved),
"dispatched" => Ok(Self::Dispatched),
"completed" => Ok(Self::Completed),
"failed" => Ok(Self::Failed),
"unknown" => Ok(Self::Unknown),
other => Err(Error::Store(format!(
"unknown Backend Job attempt state `{other}`"
))),
}
}
pub fn terminal(self) -> bool {
matches!(self, Self::Completed | Self::Failed | Self::Unknown)
}
}
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum BackendJobDeliveryState {
Pending,
Sending,
Completed,
Failed,
Unknown,
}
impl BackendJobDeliveryState {
pub(crate) fn as_str(self) -> &'static str {
match self {
Self::Pending => "pending",
Self::Sending => "sending",
Self::Completed => "completed",
Self::Failed => "failed",
Self::Unknown => "unknown",
}
}
pub(crate) fn parse(value: &str) -> Result<Self> {
match value {
"pending" => Ok(Self::Pending),
"sending" => Ok(Self::Sending),
"completed" => Ok(Self::Completed),
"failed" => Ok(Self::Failed),
"unknown" => Ok(Self::Unknown),
other => Err(Error::Store(format!(
"unknown Backend Job delivery state `{other}`"
))),
}
}
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[serde(deny_unknown_fields)]
pub struct BackendJobLimits {
#[serde(default = "default_max_concurrent_jobs")]
pub max_concurrent_jobs: u16,
#[serde(default = "default_job_timeout_seconds")]
pub timeout_seconds: u32,
#[serde(default = "default_max_result_bytes")]
pub max_result_bytes: u32,
#[serde(default = "default_max_attempts")]
pub max_attempts: u8,
}
impl Default for BackendJobLimits {
fn default() -> Self {
Self {
max_concurrent_jobs: DEFAULT_MAX_CONCURRENT_JOBS,
timeout_seconds: DEFAULT_JOB_TIMEOUT_SECONDS,
max_result_bytes: DEFAULT_MAX_RESULT_BYTES,
max_attempts: DEFAULT_MAX_ATTEMPTS,
}
}
}
impl BackendJobLimits {
pub fn validate(&self) -> Result<()> {
if self.max_concurrent_jobs == 0 || self.max_concurrent_jobs > ABSOLUTE_MAX_CONCURRENT_JOBS
{
return Err(Error::InvalidInput(format!(
"Backend Job max_concurrent_jobs must be in 1..={ABSOLUTE_MAX_CONCURRENT_JOBS}"
)));
}
if self.timeout_seconds == 0 || self.timeout_seconds > MAX_JOB_TIMEOUT_SECONDS {
return Err(Error::InvalidInput(format!(
"Backend Job timeout_seconds must be in 1..={MAX_JOB_TIMEOUT_SECONDS}"
)));
}
if self.max_result_bytes == 0 || self.max_result_bytes > ABSOLUTE_MAX_RESULT_BYTES {
return Err(Error::InvalidInput(format!(
"Backend Job max_result_bytes must be in 1..={ABSOLUTE_MAX_RESULT_BYTES}"
)));
}
if self.max_attempts == 0 || self.max_attempts > ABSOLUTE_MAX_ATTEMPTS {
return Err(Error::InvalidInput(format!(
"Backend Job max_attempts must be in 1..={ABSOLUTE_MAX_ATTEMPTS}"
)));
}
Ok(())
}
}
const fn default_max_concurrent_jobs() -> u16 {
DEFAULT_MAX_CONCURRENT_JOBS
}
const fn default_job_timeout_seconds() -> u32 {
DEFAULT_JOB_TIMEOUT_SECONDS
}
const fn default_max_result_bytes() -> u32 {
DEFAULT_MAX_RESULT_BYTES
}
const fn default_max_attempts() -> u8 {
DEFAULT_MAX_ATTEMPTS
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(deny_unknown_fields)]
pub struct BackendJobRequest {
/// Stable caller-selected idempotency key within one Workspace.
pub job_id: String,
/// Stable machine-readable use-site name, for example `ticket_item_check`.
pub purpose: String,
/// Revision fenced by the structured result.
pub input_revision: String,
/// Durable domain reference used to reload or explain the input.
pub input_ref: String,
/// Bounded immutable input snapshot. This must not contain credentials.
pub input: Value,
/// Bounded task instruction; the profile supplies the system instruction.
pub instruction: String,
/// Existing configured profile selector. Job dispatch never accepts raw profile source.
pub profile: String,
/// Writer provenance and optional advisory destination, not parent ownership.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub source_worker: Option<RuntimeWorkerRef>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub notification_target: Option<RuntimeWorkerRef>,
#[serde(default)]
pub limits: BackendJobLimits,
}
impl BackendJobRequest {
pub fn validate(&self) -> Result<()> {
validate_identifier("job_id", &self.job_id, 256)?;
validate_identifier("purpose", &self.purpose, MAX_JOB_PURPOSE_BYTES)?;
validate_identifier("input_revision", &self.input_revision, 256)?;
validate_bounded_text("input_ref", &self.input_ref, MAX_JOB_REFERENCE_BYTES)?;
validate_bounded_text("instruction", &self.instruction, MAX_JOB_INSTRUCTION_BYTES)?;
validate_identifier("profile", &self.profile, 256)?;
if self.profile != "builtin:backend-job" {
return Err(Error::InvalidInput(
"Backend Jobs must use the dedicated builtin:backend-job profile".to_string(),
));
}
let input = serde_json::to_vec(&self.input).map_err(|error| {
Error::InvalidInput(format!("serialize Backend Job input: {error}"))
})?;
if input.len() > MAX_JOB_INPUT_BYTES {
return Err(Error::InvalidInput(format!(
"Backend Job input exceeds {MAX_JOB_INPUT_BYTES} bytes"
)));
}
self.limits.validate()
}
pub fn fingerprint(&self) -> Result<String> {
self.validate()?;
let encoded = serde_json::to_vec(self).map_err(|error| {
Error::InvalidInput(format!("serialize Backend Job intent: {error}"))
})?;
Ok(sha256(&encoded))
}
pub fn worker_input(&self, attempt_id: &str) -> Result<String> {
let input = serde_json::to_string(&self.input).map_err(|error| {
Error::InvalidInput(format!("serialize Backend Job input: {error}"))
})?;
Ok(format!(
"{}\n\nBackend Job envelope (immutable):\njob_id: {}\nattempt_id: {}\ninput_revision: {}\ninput_ref: {}\ninput_json: {}\n\nReturn success only through the structured Backend Job result capability. Final prose and Worker Idle/Stopped state are not result authority.",
self.instruction, self.job_id, attempt_id, self.input_revision, self.input_ref, input
))
}
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct BackendJobRecord {
pub workspace_id: String,
pub request: BackendJobRequest,
pub intent_fingerprint: String,
pub state: BackendJobState,
pub current_attempt: u8,
pub result: Option<Value>,
pub result_digest: Option<String>,
pub failure_category: Option<String>,
pub failure_detail: Option<String>,
pub created_at: String,
pub updated_at: String,
pub completed_at: Option<String>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct BackendJobAttemptRecord {
pub workspace_id: String,
pub job_id: String,
pub attempt_id: String,
pub attempt: u8,
pub input_revision: String,
pub state: BackendJobAttemptState,
pub worker: Option<RuntimeWorkerRef>,
pub runtime_run_id: Option<String>,
pub dispatched_at: Option<String>,
pub deadline_at: String,
pub result: Option<Value>,
pub result_digest: Option<String>,
pub failure_category: Option<String>,
pub failure_detail: Option<String>,
pub created_at: String,
pub updated_at: String,
pub completed_at: Option<String>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct BackendJobReservation {
pub job: BackendJobRecord,
pub attempt: BackendJobAttemptRecord,
pub replayed: bool,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(deny_unknown_fields)]
pub struct BackendJobResultSubmission {
pub job_id: String,
pub attempt_id: String,
pub input_revision: String,
pub result: Value,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct BackendJobResultAcceptance {
pub job: BackendJobRecord,
pub attempt: BackendJobAttemptRecord,
pub replayed: bool,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct BackendJobWorkerBinding {
pub job_id: String,
pub attempt_id: String,
pub purpose: String,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct BackendJobDeliveryRecord {
pub workspace_id: String,
pub delivery_id: String,
pub job_id: String,
pub attempt_id: String,
pub target: RuntimeWorkerRef,
pub state: BackendJobDeliveryState,
pub failure_category: Option<String>,
pub failure_detail: Option<String>,
pub created_at: String,
pub updated_at: String,
pub delivered_at: Option<String>,
}
pub fn attempt_id(job_id: &str, attempt: u8) -> String {
format!("{job_id}:attempt:{attempt}")
}
pub fn allocation_key(attempt_id: &str) -> String {
format!("backend-job:{attempt_id}")
}
/// Fixed-size Runtime request identity for the exact Job attempt. Caller-selected
/// Job ids may be longer than the Worker protocol request-id bound, so never
/// embed them directly in tracked Runtime identities.
pub fn runtime_run_id(attempt_id: &str) -> String {
tracked_request_id("run", attempt_id.as_bytes())
}
/// Fixed-size durable delivery identity, also reused as the Worker's tracked
/// notification request id so restart replay converges on one receipt.
pub fn delivery_id(job_id: &str, attempt_id: &str) -> String {
let mut identity = Vec::with_capacity(job_id.len() + attempt_id.len() + 1);
identity.extend_from_slice(job_id.as_bytes());
identity.push(0);
identity.extend_from_slice(attempt_id.as_bytes());
tracked_request_id("notification", &identity)
}
fn tracked_request_id(kind: &str, identity: &[u8]) -> String {
format!("backend-job-{kind}:{}", sha256(identity))
}
pub fn result_digest(result: &Value, max_bytes: u32) -> Result<(String, String)> {
let encoded = serde_json::to_string(result)
.map_err(|error| Error::InvalidInput(format!("serialize Backend Job result: {error}")))?;
if encoded.len() > max_bytes as usize {
return Err(Error::InvalidInput(format!(
"Backend Job result exceeds {max_bytes} bytes"
)));
}
Ok((sha256(encoded.as_bytes()), encoded))
}
pub(crate) fn bounded_failure_detail(detail: &str) -> String {
if detail.len() <= MAX_JOB_FAILURE_BYTES {
return detail.to_string();
}
let mut end = MAX_JOB_FAILURE_BYTES;
while !detail.is_char_boundary(end) {
end -= 1;
}
detail[..end].to_string()
}
fn validate_identifier(field: &str, value: &str, max_bytes: usize) -> Result<()> {
validate_bounded_text(field, value, max_bytes)?;
if value.chars().any(char::is_control) {
return Err(Error::InvalidInput(format!(
"Backend Job {field} must not contain control characters"
)));
}
Ok(())
}
fn validate_bounded_text(field: &str, value: &str, max_bytes: usize) -> Result<()> {
if value.is_empty() || value.trim() != value {
return Err(Error::InvalidInput(format!(
"Backend Job {field} must be non-empty and trimmed"
)));
}
if value.len() > max_bytes {
return Err(Error::InvalidInput(format!(
"Backend Job {field} exceeds {max_bytes} bytes"
)));
}
Ok(())
}
fn sha256(bytes: &[u8]) -> String {
let digest = Sha256::digest(bytes);
let mut output = String::with_capacity(7 + digest.len() * 2);
output.push_str("sha256:");
for byte in digest {
use std::fmt::Write as _;
let _ = write!(output, "{byte:02x}");
}
output
}
#[cfg(test)]
mod tests {
use super::*;
fn request() -> BackendJobRequest {
BackendJobRequest {
job_id: "ticket-check:T-1:r7".to_string(),
purpose: "ticket_item_check".to_string(),
input_revision: "7".to_string(),
input_ref: "ticket://T-1/revisions/7".to_string(),
input: serde_json::json!({"title": "Check me"}),
instruction: "Check this immutable Ticket snapshot.".to_string(),
profile: "builtin:backend-job".to_string(),
source_worker: Some(RuntimeWorkerRef::new("runtime-a", "worker-a")),
notification_target: Some(RuntimeWorkerRef::new("runtime-a", "worker-a")),
limits: BackendJobLimits::default(),
}
}
#[test]
fn fingerprint_covers_revision_profile_limits_and_notification_provenance() {
let request = request();
let first = request.fingerprint().unwrap();
assert_eq!(first, request.fingerprint().unwrap());
for mutate in [
|request: &mut BackendJobRequest| request.input_revision = "8".to_string(),
|request: &mut BackendJobRequest| request.purpose = "alternate_check".to_string(),
|request: &mut BackendJobRequest| request.limits.max_attempts = 3,
|request: &mut BackendJobRequest| request.notification_target = None,
] {
let mut changed = request.clone();
mutate(&mut changed);
assert_ne!(first, changed.fingerprint().unwrap());
}
}
#[test]
fn tracked_runtime_ids_are_fixed_size_for_maximum_job_ids() {
let job_id = "j".repeat(256);
let attempt_id = attempt_id(&job_id, ABSOLUTE_MAX_ATTEMPTS);
let run_id = runtime_run_id(&attempt_id);
let notification_id = delivery_id(&job_id, &attempt_id);
assert!(run_id.len() <= 128, "run id was {} bytes", run_id.len());
assert!(
notification_id.len() <= 128,
"notification id was {} bytes",
notification_id.len()
);
assert_eq!(run_id, runtime_run_id(&attempt_id));
assert_eq!(notification_id, delivery_id(&job_id, &attempt_id));
assert_ne!(run_id, runtime_run_id(&format!("{attempt_id}-other")));
assert_ne!(
notification_id,
delivery_id(&job_id, &format!("{attempt_id}-other"))
);
}
#[test]
fn result_is_bounded_before_digesting() {
let value = serde_json::json!({"ok": true});
let (digest, encoded) = result_digest(&value, 128).unwrap();
assert!(digest.starts_with("sha256:"));
assert_eq!(encoded, r#"{"ok":true}"#);
assert!(result_digest(&serde_json::json!({"body": "too long"}), 4).is_err());
}
#[test]
fn failure_detail_is_bounded_by_utf8_bytes_without_splitting_code_points() {
let detail = "界".repeat(MAX_JOB_FAILURE_BYTES);
let bounded = bounded_failure_detail(&detail);
assert!(bounded.len() <= MAX_JOB_FAILURE_BYTES);
assert!(detail.starts_with(&bounded));
}
#[test]
fn job_limits_reject_unbounded_execution_and_retry() {
for limits in [
BackendJobLimits {
timeout_seconds: 0,
..Default::default()
},
BackendJobLimits {
timeout_seconds: MAX_JOB_TIMEOUT_SECONDS + 1,
..Default::default()
},
BackendJobLimits {
max_attempts: 0,
..Default::default()
},
BackendJobLimits {
max_attempts: ABSOLUTE_MAX_ATTEMPTS + 1,
..Default::default()
},
BackendJobLimits {
max_concurrent_jobs: 0,
..Default::default()
},
] {
assert!(limits.validate().is_err());
}
}
}
+63 -9
View File
@@ -544,6 +544,11 @@ pub enum WorkerSpawnIntent {
WorkspaceCompanion,
WorkspaceOrchestrator,
WorkspaceCoding,
BackendJob {
job_id: String,
attempt_id: String,
purpose: String,
},
TicketRole {
ticket_id: String,
role: TicketWorkerRole,
@@ -705,6 +710,8 @@ pub struct WorkerInputRequest {
pub kind: WorkerInputKind,
pub content: String,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub submission_request_id: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub segments: Option<Vec<protocol::Segment>>,
}
@@ -725,11 +732,22 @@ pub struct WorkerCompletionsResult {
pub diagnostics: Vec<RuntimeDiagnostic>,
}
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum WorkerInputDisposition {
Accepted,
Rejected,
Unknown,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct WorkerInputResult {
pub state: InternalWorkerOperationState,
pub disposition: WorkerInputDisposition,
#[serde(flatten)]
pub worker: RuntimeWorkerRef,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub runtime_run_id: Option<String>,
pub diagnostics: Vec<RuntimeDiagnostic>,
}
@@ -1200,7 +1218,9 @@ pub trait WorkspaceWorkerRuntime: Send + Sync {
fn send_input(&self, worker_id: &str, _request: WorkerInputRequest) -> WorkerInputResult {
WorkerInputResult {
state: InternalWorkerOperationState::Unsupported,
disposition: WorkerInputDisposition::Rejected,
worker: RuntimeWorkerRef::new(self.runtime_id().to_string(), worker_id.to_string()),
runtime_run_id: None,
diagnostics: vec![diagnostic(
"worker_input_pending",
HostDiagnosticSeverity::Info,
@@ -2985,6 +3005,7 @@ impl WorkspaceWorkerRuntime for EmbeddedWorkerRuntime {
return embedded_input_rejected(
&self.runtime_id,
worker_id,
WorkerInputDisposition::Rejected,
diagnostic(
"embedded_worker_execution_unavailable",
HostDiagnosticSeverity::Info,
@@ -2998,6 +3019,7 @@ impl WorkspaceWorkerRuntime for EmbeddedWorkerRuntime {
return embedded_input_rejected(
&self.runtime_id,
worker_id,
WorkerInputDisposition::Rejected,
diagnostic(
"embedded_worker_id_invalid",
HostDiagnosticSeverity::Warning,
@@ -3014,20 +3036,34 @@ impl WorkspaceWorkerRuntime for EmbeddedWorkerRuntime {
WorkerInputKind::RegisterPeer => EmbeddedWorkerInputKind::RegisterPeer,
},
content: request.content,
submission_request_id: None,
submission_request_id: request.submission_request_id,
segments: request.segments,
};
match self.runtime.send_input(&worker_ref, input) {
Ok(_) => WorkerInputResult {
Ok(ack) => WorkerInputResult {
state: InternalWorkerOperationState::Accepted,
disposition: WorkerInputDisposition::Accepted,
worker: RuntimeWorkerRef::new(self.runtime_id.clone(), worker_id.to_string()),
runtime_run_id: ack
.submission
.map(|submission| submission.submission_request_id),
diagnostics: Vec::new(),
},
Err(error) => embedded_input_rejected(
&self.runtime_id,
worker_id,
embedded_runtime_diagnostic(&error),
),
Err(error) => {
let disposition = match &error {
EmbeddedRuntimeError::WorkerExecutionRejected {
outcome: worker_runtime::execution::WorkerExecutionOutcome::Errored,
..
} => WorkerInputDisposition::Unknown,
_ => WorkerInputDisposition::Rejected,
};
embedded_input_rejected(
&self.runtime_id,
worker_id,
disposition,
embedded_runtime_diagnostic(&error),
)
}
}
}
@@ -4991,7 +5027,7 @@ impl WorkspaceWorkerRuntime for RemoteWorkerRuntime {
WorkerInputKind::RegisterPeer => EmbeddedWorkerInputKind::RegisterPeer,
},
content: request.content,
submission_request_id: None,
submission_request_id: request.submission_request_id,
segments: request.segments,
};
let input = match runtime_contract_convert(input) {
@@ -4999,7 +5035,9 @@ impl WorkspaceWorkerRuntime for RemoteWorkerRuntime {
Err(diagnostic) => {
return WorkerInputResult {
state: InternalWorkerOperationState::Rejected,
disposition: WorkerInputDisposition::Rejected,
worker: RuntimeWorkerRef::new(self.runtime_id.clone(), worker_id.to_string()),
runtime_run_id: None,
diagnostics: vec![diagnostic],
};
}
@@ -5010,9 +5048,14 @@ impl WorkspaceWorkerRuntime for RemoteWorkerRuntime {
MAX_REMOTE_RUNTIME_RESPONSE_BYTES,
move |client| async move { client.send_worker_input(worker_id_owned, input).await },
) {
Ok(_) => WorkerInputResult {
Ok(response) => WorkerInputResult {
state: InternalWorkerOperationState::Accepted,
disposition: WorkerInputDisposition::Accepted,
worker: RuntimeWorkerRef::new(self.runtime_id.clone(), worker_id.to_string()),
runtime_run_id: response
.ack
.submission
.map(|submission| submission.submission_request_id),
diagnostics: Vec::new(),
},
Err(diagnostic) => remote_input_rejected(&self.runtime_id, worker_id, diagnostic),
@@ -5396,11 +5439,14 @@ fn profile_display_name(profile_label: &str) -> String {
fn embedded_input_rejected(
runtime_id: &str,
worker_id: &str,
disposition: WorkerInputDisposition,
diagnostic: RuntimeDiagnostic,
) -> WorkerInputResult {
WorkerInputResult {
state: InternalWorkerOperationState::Rejected,
disposition,
worker: RuntimeWorkerRef::new(runtime_id.to_string(), worker_id.to_string()),
runtime_run_id: None,
diagnostics: vec![diagnostic],
}
}
@@ -5412,7 +5458,9 @@ fn remote_input_rejected(
) -> WorkerInputResult {
WorkerInputResult {
state: InternalWorkerOperationState::Rejected,
disposition: WorkerInputDisposition::Unknown,
worker: RuntimeWorkerRef::new(runtime_id.to_string(), worker_id.to_string()),
runtime_run_id: None,
diagnostics: vec![diagnostic],
}
}
@@ -5890,6 +5938,7 @@ fn worker_spawn_intent_label(intent: &WorkerSpawnIntent) -> &'static str {
WorkerSpawnIntent::WorkspaceCompanion => "workspace_companion",
WorkerSpawnIntent::WorkspaceOrchestrator => "workspace_orchestrator",
WorkerSpawnIntent::WorkspaceCoding => "workspace_coding",
WorkerSpawnIntent::BackendJob { .. } => "backend_job",
WorkerSpawnIntent::TicketRole { role, .. } => match role {
TicketWorkerRole::Intake => "ticket_intake",
TicketWorkerRole::Orchestrator => "ticket_orchestrator",
@@ -7088,10 +7137,13 @@ mod tests {
WorkerInputRequest {
kind: WorkerInputKind::User,
content: "hello".to_string(),
submission_request_id: Some("tracked-input-1".to_string()),
segments: None,
},
);
assert_eq!(input.state, InternalWorkerOperationState::Accepted);
assert_eq!(input.disposition, WorkerInputDisposition::Accepted);
assert_eq!(input.runtime_run_id.as_deref(), Some("tracked-input-1"));
let deadline = std::time::Instant::now() + std::time::Duration::from_secs(2);
loop {
@@ -7183,6 +7235,7 @@ mod tests {
WorkerInputRequest {
kind: WorkerInputKind::User,
content: "hello embedded runtime".to_string(),
submission_request_id: None,
segments: None,
},
)
@@ -7467,6 +7520,7 @@ mod tests {
WorkerInputRequest {
kind: WorkerInputKind::User,
content: "hello remote".to_string(),
submission_request_id: None,
segments: None,
},
)
@@ -806,6 +806,79 @@ CREATE TABLE "worker_registry" (
PRIMARY KEY (workspace_id, worker_id),
FOREIGN KEY (workspace_id) REFERENCES workspaces(workspace_id) ON DELETE CASCADE
);
CREATE TABLE backend_jobs (
workspace_id TEXT NOT NULL,
job_id TEXT NOT NULL,
purpose TEXT NOT NULL,
input_revision TEXT NOT NULL,
input_ref TEXT NOT NULL,
request_json TEXT NOT NULL,
intent_fingerprint TEXT NOT NULL,
state TEXT NOT NULL CHECK (state IN ('pending', 'completed', 'failed', 'unknown')),
current_attempt INTEGER NOT NULL CHECK (current_attempt > 0 AND current_attempt <= 3),
result_json TEXT,
result_digest TEXT,
failure_category TEXT,
failure_detail TEXT,
created_at TEXT NOT NULL,
updated_at TEXT NOT NULL,
completed_at TEXT,
PRIMARY KEY (workspace_id, job_id),
FOREIGN KEY (workspace_id) REFERENCES workspaces(workspace_id) ON DELETE CASCADE,
CHECK ((result_json IS NULL) = (result_digest IS NULL)),
CHECK ((state = 'completed') = (result_json IS NOT NULL))
);
CREATE INDEX backend_jobs_active
ON backend_jobs(workspace_id, state, updated_at);
CREATE TABLE backend_job_attempts (
workspace_id TEXT NOT NULL,
job_id TEXT NOT NULL,
attempt_id TEXT NOT NULL,
attempt INTEGER NOT NULL CHECK (attempt > 0 AND attempt <= 3),
input_revision TEXT NOT NULL,
state TEXT NOT NULL CHECK (state IN ('reserved', 'dispatched', 'completed', 'failed', 'unknown')),
runtime_id TEXT,
worker_id TEXT,
runtime_run_id TEXT,
dispatched_at TEXT,
deadline_at TEXT NOT NULL,
result_json TEXT,
result_digest TEXT,
failure_category TEXT,
failure_detail TEXT,
created_at TEXT NOT NULL,
updated_at TEXT NOT NULL,
completed_at TEXT,
PRIMARY KEY (workspace_id, job_id, attempt_id),
UNIQUE (workspace_id, job_id, attempt),
UNIQUE (workspace_id, runtime_id, worker_id),
FOREIGN KEY (workspace_id, job_id)
REFERENCES backend_jobs(workspace_id, job_id) ON DELETE CASCADE,
CHECK ((runtime_id IS NULL) = (worker_id IS NULL)),
CHECK ((result_json IS NULL) = (result_digest IS NULL)),
CHECK ((state = 'completed') = (result_json IS NOT NULL))
);
CREATE INDEX backend_job_attempts_recovery
ON backend_job_attempts(workspace_id, state, deadline_at);
CREATE TABLE backend_job_deliveries (
workspace_id TEXT NOT NULL,
delivery_id TEXT NOT NULL,
job_id TEXT NOT NULL,
attempt_id TEXT NOT NULL,
target_runtime_id TEXT NOT NULL,
target_worker_id TEXT NOT NULL,
state TEXT NOT NULL CHECK (state IN ('pending', 'sending', 'completed', 'failed', 'unknown')),
failure_category TEXT,
failure_detail TEXT,
created_at TEXT NOT NULL,
updated_at TEXT NOT NULL,
delivered_at TEXT,
PRIMARY KEY (workspace_id, delivery_id),
FOREIGN KEY (workspace_id, job_id, attempt_id)
REFERENCES backend_job_attempts(workspace_id, job_id, attempt_id) ON DELETE CASCADE
);
CREATE INDEX backend_job_deliveries_pending
ON backend_job_deliveries(workspace_id, state, updated_at);
CREATE TABLE worker_registry_observations (
workspace_id TEXT NOT NULL,
runtime_id TEXT NOT NULL,
+1
View File
@@ -6,6 +6,7 @@
pub mod auth;
pub mod authority;
pub mod backend_job;
pub mod companion;
pub mod config;
pub mod config_source;
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
@@ -18,6 +18,9 @@ use crate::{Error, Result};
const WORKSPACE_DELETION_PURGE_TABLES: &[&str] = &[
"artifacts",
"audit_events",
"backend_job_deliveries",
"backend_job_attempts",
"backend_jobs",
"external_workdir_grants",
"flow_source_revisions",
"flow_sources",
@@ -449,6 +449,7 @@ fn project_registry_worker(
workspace_id: Some(record.registry.workspace_id.clone()),
display_name: Some(record.registry.display_name.clone()),
profile: record.registry.profile.clone(),
job: None,
workdir_attachments: Vec::new(),
}
};
@@ -457,6 +458,13 @@ fn project_registry_worker(
worker.workspace_id = Some(record.registry.workspace_id);
worker.display_name = Some(record.registry.display_name);
worker.profile = record.registry.profile;
worker.job = record
.job
.map(|job| protocol::subscription::SubscriptionWorkerJob {
job_id: job.job_id,
attempt_id: job.attempt_id,
purpose: job.purpose,
});
if !project_working_directory(api, &record.registry.worker, &mut worker) {
return None;
}
@@ -616,6 +624,7 @@ mod tests {
workspace_id: Some("workspace-1".to_string()),
display_name: Some("Companion".to_string()),
profile: Some("builtin:companion".to_string()),
job: None,
workdir_attachments: vec![
protocol::subscription::SubscriptionWorkerWorkdirAttachment {
alias: "external".to_string(),