feat: route backend worker controls over protocol
This commit is contained in:
@@ -2,9 +2,9 @@ use std::collections::VecDeque;
|
||||
use std::fmt;
|
||||
use std::time::Duration;
|
||||
|
||||
use futures::StreamExt;
|
||||
use protocol::{ErrorCode, Event, Method, Segment};
|
||||
use serde::{Deserialize, Serialize};
|
||||
use futures::{SinkExt, StreamExt};
|
||||
use protocol::{ErrorCode, Event, Method};
|
||||
use serde::Deserialize;
|
||||
use tokio::sync::mpsc;
|
||||
use tokio_tungstenite::connect_async;
|
||||
use tokio_tungstenite::tungstenite::Message as TungsteniteMessage;
|
||||
@@ -160,10 +160,11 @@ pub struct BackendWorkerSummary {
|
||||
#[derive(Debug)]
|
||||
pub struct BackendRuntimeClient {
|
||||
target: BackendRuntimeTarget,
|
||||
http: reqwest::Client,
|
||||
command_tx: mpsc::UnboundedSender<Method>,
|
||||
events: mpsc::UnboundedReceiver<Event>,
|
||||
diagnostics: VecDeque<Event>,
|
||||
_observation_task: tokio::task::JoinHandle<()>,
|
||||
_command_task: tokio::task::JoinHandle<()>,
|
||||
}
|
||||
|
||||
#[derive(Debug)]
|
||||
@@ -258,21 +259,28 @@ pub async fn list_backend_workers(
|
||||
impl BackendRuntimeClient {
|
||||
pub async fn connect(target: BackendRuntimeTarget) -> Result<Self, BackendRuntimeClientError> {
|
||||
validate_target(&target)?;
|
||||
let http = reqwest::Client::new();
|
||||
let (tx, rx) = mpsc::unbounded_channel();
|
||||
let (event_tx, rx) = mpsc::unbounded_channel();
|
||||
let (command_tx, command_rx) = mpsc::unbounded_channel();
|
||||
|
||||
let observation_target = target.clone();
|
||||
let observation_tx = tx.clone();
|
||||
let observation_tx = event_tx.clone();
|
||||
let observation_task = tokio::spawn(async move {
|
||||
observe_worker_events(observation_target, observation_tx).await;
|
||||
});
|
||||
|
||||
let command_target = target.clone();
|
||||
let command_event_tx = event_tx.clone();
|
||||
let command_task = tokio::spawn(async move {
|
||||
run_worker_protocol_commands(command_target, command_rx, command_event_tx).await;
|
||||
});
|
||||
|
||||
Ok(Self {
|
||||
target,
|
||||
http,
|
||||
command_tx,
|
||||
events: rx,
|
||||
diagnostics: VecDeque::new(),
|
||||
_observation_task: observation_task,
|
||||
_command_task: command_task,
|
||||
})
|
||||
}
|
||||
|
||||
@@ -291,163 +299,106 @@ impl BackendRuntimeClient {
|
||||
}
|
||||
|
||||
pub async fn send(&mut self, method: &Method) -> Result<(), BackendRuntimeClientError> {
|
||||
match backend_command_from_method(method) {
|
||||
BackendCommand::Input { kind, content } => {
|
||||
let url = self.worker_api_url("input");
|
||||
match self
|
||||
.http
|
||||
.post(url)
|
||||
.json(&WorkerInputRequest { kind, content })
|
||||
.send()
|
||||
.await
|
||||
.and_then(|response| response.error_for_status())
|
||||
{
|
||||
Ok(response) => match response.json::<WorkerInputResult>().await {
|
||||
Ok(result) => self.enqueue_operation_diagnostics(
|
||||
"input",
|
||||
result.state,
|
||||
result.diagnostics,
|
||||
),
|
||||
Err(error) => self.enqueue_diagnostic(format!(
|
||||
"Backend runtime input response could not be decoded for {}: {error}",
|
||||
self.target.display_label()
|
||||
)),
|
||||
},
|
||||
Err(error) => self.enqueue_diagnostic(format!(
|
||||
"Backend runtime input failed for {}: {error}",
|
||||
self.target.display_label()
|
||||
)),
|
||||
}
|
||||
}
|
||||
BackendCommand::Lifecycle { action, reason } => {
|
||||
let url = self.worker_api_url(action);
|
||||
match self
|
||||
.http
|
||||
.post(url)
|
||||
.json(&WorkerLifecycleRequest { reason })
|
||||
.send()
|
||||
.await
|
||||
.and_then(|response| response.error_for_status())
|
||||
{
|
||||
Ok(response) => match response.json::<WorkerLifecycleResult>().await {
|
||||
Ok(result) => self.enqueue_operation_diagnostics(
|
||||
action,
|
||||
result.state,
|
||||
result.diagnostics,
|
||||
),
|
||||
Err(error) => self.enqueue_diagnostic(format!(
|
||||
"Backend runtime {action} response could not be decoded for {}: {error}",
|
||||
self.target.display_label()
|
||||
)),
|
||||
},
|
||||
Err(error) => self.enqueue_diagnostic(format!(
|
||||
"Backend runtime {action} failed for {}: {error}",
|
||||
self.target.display_label()
|
||||
)),
|
||||
}
|
||||
}
|
||||
BackendCommand::Unsupported(message) => {
|
||||
self.enqueue_diagnostic(message);
|
||||
}
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn worker_api_url(&self, suffix: &str) -> String {
|
||||
let path = format!(
|
||||
"/api/runtimes/{}/workers/{}/{}",
|
||||
path_segment_encode(&self.target.runtime_id),
|
||||
path_segment_encode(&self.target.worker_id),
|
||||
suffix
|
||||
);
|
||||
join_base_and_path(&self.target.base_url, &path)
|
||||
}
|
||||
|
||||
fn enqueue_operation_diagnostics(
|
||||
&mut self,
|
||||
operation: &str,
|
||||
state: String,
|
||||
diagnostics: Vec<BackendDiagnostic>,
|
||||
) {
|
||||
if state != "accepted" {
|
||||
self.enqueue_diagnostic(format!(
|
||||
"Backend runtime {operation} was {state} for {}",
|
||||
self.command_tx.send(method.clone()).map_err(|_| {
|
||||
BackendRuntimeClientError::InvalidTarget(format!(
|
||||
"Backend protocol command stream is closed for {}",
|
||||
self.target.display_label()
|
||||
));
|
||||
}
|
||||
for diagnostic in diagnostics {
|
||||
self.enqueue_diagnostic(format!(
|
||||
"Backend runtime {operation} diagnostic [{}]: {}",
|
||||
diagnostic.code, diagnostic.message
|
||||
));
|
||||
}
|
||||
}
|
||||
|
||||
fn enqueue_diagnostic(&mut self, message: impl Into<String>) {
|
||||
self.diagnostics.push_back(diagnostic_event(message));
|
||||
))
|
||||
})?;
|
||||
Ok(())
|
||||
}
|
||||
}
|
||||
|
||||
impl Drop for BackendRuntimeClient {
|
||||
fn drop(&mut self) {
|
||||
self._observation_task.abort();
|
||||
self._command_task.abort();
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Debug, PartialEq, Eq)]
|
||||
enum BackendCommand {
|
||||
Input {
|
||||
kind: WorkerInputKind,
|
||||
content: String,
|
||||
},
|
||||
Lifecycle {
|
||||
action: &'static str,
|
||||
reason: Option<String>,
|
||||
},
|
||||
Unsupported(String),
|
||||
}
|
||||
|
||||
fn backend_command_from_method(method: &Method) -> BackendCommand {
|
||||
match method {
|
||||
Method::Run { input } => BackendCommand::Input {
|
||||
kind: WorkerInputKind::User,
|
||||
content: Segment::flatten_to_text(input),
|
||||
},
|
||||
Method::Notify { message, .. } => BackendCommand::Input {
|
||||
kind: WorkerInputKind::System,
|
||||
content: message.clone(),
|
||||
},
|
||||
Method::Cancel => BackendCommand::Lifecycle {
|
||||
action: "cancel",
|
||||
reason: Some("requested from TUI Backend Runtime API client".to_string()),
|
||||
},
|
||||
Method::Shutdown => BackendCommand::Lifecycle {
|
||||
action: "stop",
|
||||
reason: Some("requested from TUI Backend Runtime API client".to_string()),
|
||||
},
|
||||
Method::Pause => BackendCommand::Unsupported(
|
||||
"Backend Runtime API does not expose pause/resume for the TUI client yet; command was not sent".to_string(),
|
||||
),
|
||||
Method::Resume => BackendCommand::Unsupported(
|
||||
"Backend Runtime API does not expose resume for the TUI client yet; command was not sent".to_string(),
|
||||
),
|
||||
Method::Compact => BackendCommand::Unsupported(
|
||||
"Backend Runtime API does not expose compaction for the TUI client yet; command was not sent".to_string(),
|
||||
),
|
||||
Method::ListCompletions { .. } => BackendCommand::Unsupported(
|
||||
"Backend Runtime API does not expose completion lookup for the TUI client yet".to_string(),
|
||||
),
|
||||
Method::ListRewindTargets | Method::RewindTo { .. } => BackendCommand::Unsupported(
|
||||
"Backend Runtime API does not expose rewind controls for the TUI client yet; command was not sent".to_string(),
|
||||
),
|
||||
Method::ListWorkers | Method::RestoreWorker { .. } | Method::RegisterPeer { .. } => {
|
||||
BackendCommand::Unsupported(
|
||||
"Backend Runtime API worker-management controls are not available from this Console connection".to_string(),
|
||||
)
|
||||
async fn run_worker_protocol_commands(
|
||||
target: BackendRuntimeTarget,
|
||||
mut commands: mpsc::UnboundedReceiver<Method>,
|
||||
tx: mpsc::UnboundedSender<Event>,
|
||||
) {
|
||||
let url = protocol_ws_url(&target);
|
||||
match connect_async(&url).await {
|
||||
Ok((ws, _)) => {
|
||||
let (mut sink, mut stream) = ws.split();
|
||||
loop {
|
||||
tokio::select! {
|
||||
maybe_method = commands.recv() => {
|
||||
let Some(method) = maybe_method else {
|
||||
break;
|
||||
};
|
||||
match serde_json::to_string(&method) {
|
||||
Ok(text) => {
|
||||
if let Err(error) = sink.send(TungsteniteMessage::Text(text.into())).await {
|
||||
let _ = tx.send(diagnostic_event(format!(
|
||||
"Backend protocol command send failed for {}: {error}",
|
||||
target.display_label()
|
||||
)));
|
||||
break;
|
||||
}
|
||||
}
|
||||
Err(error) => {
|
||||
let _ = tx.send(diagnostic_event(format!(
|
||||
"Backend protocol command could not serialize method for {}: {error}",
|
||||
target.display_label()
|
||||
)));
|
||||
}
|
||||
}
|
||||
}
|
||||
frame = stream.next() => {
|
||||
match frame {
|
||||
Some(Ok(TungsteniteMessage::Text(text))) => {
|
||||
match serde_json::from_str::<Event>(&text) {
|
||||
Ok(event) => {
|
||||
let _ = tx.send(event);
|
||||
}
|
||||
Err(error) => {
|
||||
let _ = tx.send(diagnostic_event(format!(
|
||||
"Backend protocol response was not valid Event JSON for {}: {error}",
|
||||
target.display_label()
|
||||
)));
|
||||
}
|
||||
}
|
||||
}
|
||||
Some(Ok(TungsteniteMessage::Close(_))) | None => {
|
||||
let _ = tx.send(diagnostic_event(format!(
|
||||
"Backend protocol command stream closed for {}",
|
||||
target.display_label()
|
||||
)));
|
||||
break;
|
||||
}
|
||||
Some(Ok(TungsteniteMessage::Ping(_)))
|
||||
| Some(Ok(TungsteniteMessage::Pong(_)))
|
||||
| Some(Ok(TungsteniteMessage::Binary(_)))
|
||||
| Some(Ok(TungsteniteMessage::Frame(_))) => {}
|
||||
Some(Err(error)) => {
|
||||
let _ = tx.send(diagnostic_event(format!(
|
||||
"Backend protocol WebSocket error for {}: {error}",
|
||||
target.display_label()
|
||||
)));
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
Err(error) => {
|
||||
let _ = tx.send(diagnostic_event(format!(
|
||||
"Backend protocol WebSocket connect failed for {}: {error}",
|
||||
target.display_label()
|
||||
)));
|
||||
while commands.recv().await.is_some() {
|
||||
let _ = tx.send(diagnostic_event(format!(
|
||||
"Backend protocol command was not sent because command stream is unavailable for {}",
|
||||
target.display_label()
|
||||
)));
|
||||
}
|
||||
}
|
||||
Method::WorkerEvent(_) => BackendCommand::Unsupported(
|
||||
"Backend Runtime API does not accept child Worker lifecycle events from this Console connection".to_string(),
|
||||
),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -610,6 +561,15 @@ fn observation_ws_url(target: &BackendRuntimeTarget) -> String {
|
||||
join_base_and_path(&http_base_to_ws(&target.base_url), &path)
|
||||
}
|
||||
|
||||
fn protocol_ws_url(target: &BackendRuntimeTarget) -> String {
|
||||
let path = format!(
|
||||
"/api/runtimes/{}/workers/{}/protocol/ws",
|
||||
path_segment_encode(&target.runtime_id),
|
||||
path_segment_encode(&target.worker_id)
|
||||
);
|
||||
join_base_and_path(&http_base_to_ws(&target.base_url), &path)
|
||||
}
|
||||
|
||||
fn http_base_to_ws(base: &str) -> String {
|
||||
if let Some(rest) = base.strip_prefix("https://") {
|
||||
format!("wss://{rest}")
|
||||
@@ -643,38 +603,6 @@ fn percent_encode(input: &str, keep: impl Fn(u8) -> bool) -> String {
|
||||
encoded
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
|
||||
#[serde(rename_all = "snake_case")]
|
||||
enum WorkerInputKind {
|
||||
User,
|
||||
System,
|
||||
}
|
||||
|
||||
#[derive(Debug, Serialize)]
|
||||
struct WorkerInputRequest {
|
||||
kind: WorkerInputKind,
|
||||
content: String,
|
||||
}
|
||||
|
||||
#[derive(Debug, Serialize)]
|
||||
struct WorkerLifecycleRequest {
|
||||
reason: Option<String>,
|
||||
}
|
||||
|
||||
#[derive(Debug, Deserialize)]
|
||||
struct WorkerInputResult {
|
||||
state: String,
|
||||
#[serde(default)]
|
||||
diagnostics: Vec<BackendDiagnostic>,
|
||||
}
|
||||
|
||||
#[derive(Debug, Deserialize)]
|
||||
struct WorkerLifecycleResult {
|
||||
state: String,
|
||||
#[serde(default)]
|
||||
diagnostics: Vec<BackendDiagnostic>,
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone, Deserialize, PartialEq, Eq)]
|
||||
pub struct BackendDiagnostic {
|
||||
pub code: String,
|
||||
@@ -712,47 +640,16 @@ mod tests {
|
||||
use super::*;
|
||||
|
||||
#[test]
|
||||
fn backend_command_maps_run_to_user_input_without_runtime_endpoint() {
|
||||
let method = Method::Run {
|
||||
input: vec![
|
||||
Segment::text("hello"),
|
||||
Segment::FileRef {
|
||||
path: "src/lib.rs".into(),
|
||||
},
|
||||
],
|
||||
};
|
||||
assert_eq!(
|
||||
backend_command_from_method(&method),
|
||||
BackendCommand::Input {
|
||||
kind: WorkerInputKind::User,
|
||||
content: "hello@src/lib.rs".to_string(),
|
||||
}
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn backend_worker_list_paths_use_scoped_workspace_when_available() {
|
||||
assert_eq!(
|
||||
backend_runtimes_path(Some("workspace/one")),
|
||||
"/api/w/workspace%2Fone/runtimes"
|
||||
);
|
||||
assert_eq!(
|
||||
backend_runtime_workers_path(Some("workspace/one"), "runtime one"),
|
||||
"/api/w/workspace%2Fone/runtimes/runtime%20one/workers"
|
||||
);
|
||||
assert_eq!(
|
||||
backend_runtime_workers_path(None, "runtime one"),
|
||||
"/api/runtimes/runtime%20one/workers"
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn observation_url_uses_backend_runtime_worker_identity() {
|
||||
fn command_and_observation_urls_use_backend_protocol_paths() {
|
||||
let target =
|
||||
BackendRuntimeTarget::new("http://127.0.0.1:8787/", "runtime/one", "worker one");
|
||||
assert_eq!(
|
||||
observation_ws_url(&target),
|
||||
"ws://127.0.0.1:8787/api/runtimes/runtime%2Fone/workers/worker%20one/events/ws"
|
||||
);
|
||||
assert_eq!(
|
||||
protocol_ws_url(&target),
|
||||
"ws://127.0.0.1:8787/api/runtimes/runtime%2Fone/workers/worker%20one/protocol/ws"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -6,6 +6,7 @@ use crate::interaction::WorkerInput;
|
||||
#[cfg(feature = "ws-server")]
|
||||
use crate::observation::WorkerObservationEvent;
|
||||
use crate::working_directory::{WorkingDirectoryBinding, WorkingDirectoryDiagnostic};
|
||||
use protocol::Method;
|
||||
use serde::{Deserialize, Serialize};
|
||||
use std::fmt;
|
||||
use std::sync::Arc;
|
||||
@@ -63,6 +64,7 @@ pub enum WorkerExecutionOperation {
|
||||
Spawn,
|
||||
Restore,
|
||||
Input,
|
||||
ProtocolMethod,
|
||||
Stop,
|
||||
Cancel,
|
||||
}
|
||||
@@ -380,6 +382,17 @@ pub trait WorkerExecutionBackend: Send + Sync + 'static {
|
||||
input: WorkerInput,
|
||||
) -> WorkerExecutionResult;
|
||||
|
||||
fn dispatch_method(
|
||||
&self,
|
||||
_handle: &WorkerExecutionHandle,
|
||||
_method: Method,
|
||||
) -> WorkerExecutionResult {
|
||||
WorkerExecutionResult::unsupported(
|
||||
WorkerExecutionOperation::ProtocolMethod,
|
||||
"execution backend does not support direct Worker protocol methods",
|
||||
)
|
||||
}
|
||||
|
||||
fn stop_worker(&self, _handle: &WorkerExecutionHandle) -> WorkerExecutionResult {
|
||||
WorkerExecutionResult::unsupported(
|
||||
WorkerExecutionOperation::Stop,
|
||||
@@ -479,6 +492,14 @@ impl WorkerExecutionBackendRef {
|
||||
self.backend.dispatch_input(handle, input)
|
||||
}
|
||||
|
||||
pub(crate) fn dispatch_method(
|
||||
&self,
|
||||
handle: &WorkerExecutionHandle,
|
||||
method: Method,
|
||||
) -> WorkerExecutionResult {
|
||||
self.backend.dispatch_method(handle, method)
|
||||
}
|
||||
|
||||
pub(crate) fn stop_worker(&self, handle: &WorkerExecutionHandle) -> WorkerExecutionResult {
|
||||
self.backend.stop_worker(handle)
|
||||
}
|
||||
|
||||
@@ -154,7 +154,12 @@ pub fn runtime_http_router(runtime: Runtime, local_token: Option<String>) -> Rou
|
||||
.route("/v1/workers/{worker_id}/cancel", post(cancel_worker));
|
||||
|
||||
#[cfg(feature = "ws-server")]
|
||||
let router = router.route("/v1/workers/{worker_id}/events/ws", get(worker_events_ws));
|
||||
let router = router
|
||||
.route("/v1/workers/{worker_id}/events/ws", get(worker_events_ws))
|
||||
.route(
|
||||
"/v1/workers/{worker_id}/protocol/ws",
|
||||
get(worker_protocol_ws),
|
||||
);
|
||||
|
||||
router
|
||||
.with_state(state.clone())
|
||||
@@ -506,6 +511,94 @@ async fn create_worker(
|
||||
Ok(Json(RuntimeHttpWorkerResponse { worker }))
|
||||
}
|
||||
|
||||
#[cfg(feature = "ws-server")]
|
||||
async fn worker_protocol_ws(
|
||||
State(state): State<RuntimeHttpState>,
|
||||
Path(worker_id): Path<String>,
|
||||
ws: WebSocketUpgrade,
|
||||
) -> Result<Response, RuntimeHttpRestError> {
|
||||
let worker_ref = worker_ref_for(&state.runtime, worker_id)?;
|
||||
state
|
||||
.runtime
|
||||
.worker_detail(&worker_ref)
|
||||
.map_err(RuntimeHttpRestError::runtime)?;
|
||||
Ok(ws
|
||||
.on_upgrade(move |socket| worker_protocol_ws_session(state.runtime, worker_ref, socket))
|
||||
.into_response())
|
||||
}
|
||||
|
||||
#[cfg(feature = "ws-server")]
|
||||
async fn worker_protocol_ws_session(
|
||||
runtime: Runtime,
|
||||
worker_ref: WorkerRef,
|
||||
mut socket: WebSocket,
|
||||
) {
|
||||
while let Some(frame) = socket.next().await {
|
||||
match frame {
|
||||
Ok(WsMessage::Text(text)) => match serde_json::from_str::<protocol::Method>(&text) {
|
||||
Ok(method) => match runtime.send_protocol_method(&worker_ref, method) {
|
||||
Ok(events) => {
|
||||
for event in events {
|
||||
if !send_protocol_event(&mut socket, &event).await {
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
Err(error) => {
|
||||
let event = protocol_error_event(error.to_string());
|
||||
if !send_protocol_event(&mut socket, &event).await {
|
||||
return;
|
||||
}
|
||||
}
|
||||
},
|
||||
Err(error) => {
|
||||
let event =
|
||||
protocol_error_event(format!("malformed protocol method frame: {error}"));
|
||||
if !send_protocol_event(&mut socket, &event).await {
|
||||
return;
|
||||
}
|
||||
}
|
||||
},
|
||||
Ok(WsMessage::Close(_)) => return,
|
||||
Ok(WsMessage::Ping(payload)) => {
|
||||
if socket.send(WsMessage::Pong(payload)).await.is_err() {
|
||||
return;
|
||||
}
|
||||
}
|
||||
Ok(WsMessage::Pong(_)) | Ok(WsMessage::Binary(_)) => {}
|
||||
Err(error) => {
|
||||
let event = protocol_error_event(format!("protocol WebSocket error: {error}"));
|
||||
let _ = send_protocol_event(&mut socket, &event).await;
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(feature = "ws-server")]
|
||||
async fn send_protocol_event(socket: &mut WebSocket, event: &protocol::Event) -> bool {
|
||||
match serde_json::to_string(event) {
|
||||
Ok(text) => socket.send(WsMessage::Text(text.into())).await.is_ok(),
|
||||
Err(error) => {
|
||||
let fallback = protocol_error_event(format!(
|
||||
"failed to serialize protocol response event: {error}"
|
||||
));
|
||||
let Ok(text) = serde_json::to_string(&fallback) else {
|
||||
return false;
|
||||
};
|
||||
socket.send(WsMessage::Text(text.into())).await.is_ok()
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(feature = "ws-server")]
|
||||
fn protocol_error_event(message: impl Into<String>) -> protocol::Event {
|
||||
protocol::Event::Error {
|
||||
code: protocol::ErrorCode::Internal,
|
||||
message: message.into(),
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(feature = "ws-server")]
|
||||
async fn worker_events_ws(
|
||||
State(state): State<RuntimeHttpState>,
|
||||
|
||||
@@ -35,6 +35,7 @@ use crate::observation::{
|
||||
};
|
||||
#[cfg(feature = "ws-server")]
|
||||
use crate::observation::{WorkerObservationCursor, WorkerObservationEvent};
|
||||
use protocol::{Event, Method};
|
||||
use std::collections::BTreeMap;
|
||||
#[cfg(feature = "ws-server")]
|
||||
use std::collections::VecDeque;
|
||||
@@ -559,6 +560,71 @@ impl Runtime {
|
||||
Ok(backend.worker_completions(&handle, kind, prefix))
|
||||
}
|
||||
|
||||
/// Accept a protocol method for a Worker through a Backend/runtime transport.
|
||||
///
|
||||
/// Most methods are delivered to the execution backend unchanged. Methods with
|
||||
/// direct same-connection replies in the local socket protocol return those
|
||||
/// events from this function so WebSocket transports can write them back to the
|
||||
/// requesting client without rebroadcasting them.
|
||||
pub fn send_protocol_method(
|
||||
&self,
|
||||
worker_ref: &WorkerRef,
|
||||
method: Method,
|
||||
) -> Result<Vec<Event>, RuntimeError> {
|
||||
if let Method::ListCompletions { kind, prefix } = method {
|
||||
let entries = self.worker_completions(worker_ref, kind, &prefix)?;
|
||||
return Ok(vec![Event::Completions { kind, entries }]);
|
||||
}
|
||||
|
||||
let (backend, handle) = {
|
||||
let mut state = self.lock()?;
|
||||
state.ensure_running()?;
|
||||
state.ensure_worker_ref(worker_ref)?;
|
||||
let worker = state.worker(worker_ref)?;
|
||||
if !worker.status.is_active() {
|
||||
return Err(RuntimeError::InvalidRequest(format!(
|
||||
"worker {} is not running",
|
||||
worker_ref.worker_id
|
||||
)));
|
||||
}
|
||||
let backend = state.execution_backend.clone();
|
||||
let handle = worker.execution_handle.clone();
|
||||
match (backend, handle) {
|
||||
(Some(backend), Some(handle)) => (backend, handle),
|
||||
_ => {
|
||||
let result = WorkerExecutionResult::rejected(
|
||||
WorkerExecutionOperation::ProtocolMethod,
|
||||
"worker has no execution backend",
|
||||
);
|
||||
let worker = state.worker_mut(worker_ref)?;
|
||||
let mut execution = WorkerExecutionStatus::unconnected().with_result(result);
|
||||
execution.binding = worker.execution.binding.clone();
|
||||
worker.execution = execution;
|
||||
state.persist_worker(&worker_ref.worker_id)?;
|
||||
return Err(RuntimeError::WorkerExecutionUnavailable {
|
||||
worker_id: worker_ref.worker_id.clone(),
|
||||
message: "worker has no execution backend".to_string(),
|
||||
});
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
let dispatch_result = backend.dispatch_method(&handle, method);
|
||||
if !dispatch_result.is_accepted() {
|
||||
self.record_execution_result(worker_ref, dispatch_result.clone())?;
|
||||
return Err(RuntimeError::WorkerExecutionRejected {
|
||||
worker_id: worker_ref.worker_id.clone(),
|
||||
operation: dispatch_result.operation,
|
||||
outcome: dispatch_result.outcome,
|
||||
message: dispatch_result.message_or_default(),
|
||||
result: dispatch_result,
|
||||
});
|
||||
}
|
||||
|
||||
self.record_execution_result(worker_ref, dispatch_result)?;
|
||||
Ok(Vec::new())
|
||||
}
|
||||
|
||||
fn commit_created_worker(
|
||||
&self,
|
||||
worker_ref: &WorkerRef,
|
||||
@@ -641,7 +707,8 @@ impl Runtime {
|
||||
WorkerExecutionOperation::Cancel => backend.cancel_worker(&handle),
|
||||
WorkerExecutionOperation::Spawn
|
||||
| WorkerExecutionOperation::Restore
|
||||
| WorkerExecutionOperation::Input => return Ok(()),
|
||||
| WorkerExecutionOperation::Input
|
||||
| WorkerExecutionOperation::ProtocolMethod => return Ok(()),
|
||||
};
|
||||
if result.is_accepted() {
|
||||
self.record_execution_result(worker_ref, result)?;
|
||||
|
||||
@@ -722,6 +722,27 @@ impl<F> Drop for WorkerRuntimeExecutionBackend<F> {
|
||||
}
|
||||
}
|
||||
|
||||
fn method_starts_turn(method: &Method) -> bool {
|
||||
matches!(
|
||||
method,
|
||||
Method::Run { .. }
|
||||
| Method::Notify { auto_run: true, .. }
|
||||
| Method::Resume
|
||||
| Method::Compact
|
||||
)
|
||||
}
|
||||
|
||||
fn accepted_run_state_for_method(method: &Method) -> WorkerExecutionRunState {
|
||||
match method {
|
||||
Method::Run { .. }
|
||||
| Method::Notify { auto_run: true, .. }
|
||||
| Method::Resume
|
||||
| Method::Compact => WorkerExecutionRunState::Busy,
|
||||
Method::Shutdown => WorkerExecutionRunState::Stopped,
|
||||
_ => WorkerExecutionRunState::Idle,
|
||||
}
|
||||
}
|
||||
|
||||
impl<F> WorkerExecutionBackend for WorkerRuntimeExecutionBackend<F>
|
||||
where
|
||||
F: RuntimeWorkerFactory,
|
||||
@@ -1039,6 +1060,48 @@ where
|
||||
result
|
||||
}
|
||||
|
||||
fn dispatch_method(
|
||||
&self,
|
||||
handle: &WorkerExecutionHandle,
|
||||
method: Method,
|
||||
) -> WorkerExecutionResult {
|
||||
let (worker, busy) = match self.get_execution(handle) {
|
||||
Ok(execution) => execution,
|
||||
Err(mut result) => {
|
||||
result.operation = WorkerExecutionOperation::ProtocolMethod;
|
||||
return result;
|
||||
}
|
||||
};
|
||||
|
||||
let starts_turn = method_starts_turn(&method);
|
||||
if starts_turn
|
||||
&& (worker.shared_state.get_status() != WorkerStatus::Idle
|
||||
|| busy
|
||||
.compare_exchange(false, true, Ordering::SeqCst, Ordering::SeqCst)
|
||||
.is_err())
|
||||
{
|
||||
return WorkerExecutionResult::busy(
|
||||
WorkerExecutionOperation::ProtocolMethod,
|
||||
"Worker is already running; runtime adapter v0 does not queue protocol methods",
|
||||
);
|
||||
}
|
||||
|
||||
let accepted_run_state = accepted_run_state_for_method(&method);
|
||||
let accepted_is_idle = accepted_run_state == WorkerExecutionRunState::Idle;
|
||||
let result = self.send_method(
|
||||
WorkerExecutionOperation::ProtocolMethod,
|
||||
worker,
|
||||
method,
|
||||
accepted_run_state,
|
||||
);
|
||||
if (starts_turn && accepted_is_idle)
|
||||
|| (starts_turn && result.outcome != crate::execution::WorkerExecutionOutcome::Accepted)
|
||||
{
|
||||
busy.store(false, Ordering::SeqCst);
|
||||
}
|
||||
result
|
||||
}
|
||||
|
||||
fn stop_worker(&self, handle: &WorkerExecutionHandle) -> WorkerExecutionResult {
|
||||
if handle.backend_id() != self.backend_id() {
|
||||
return WorkerExecutionResult::rejected(
|
||||
|
||||
@@ -514,6 +514,21 @@ pub enum RuntimeRegistryError {
|
||||
}
|
||||
|
||||
impl RuntimeRegistryError {
|
||||
pub fn message(&self) -> String {
|
||||
match self {
|
||||
Self::InvalidIdentifier { kind, value } => {
|
||||
format!("invalid {kind} identifier `{value}`")
|
||||
}
|
||||
Self::UnknownRuntime(runtime_id) => format!("unknown runtime `{runtime_id}`"),
|
||||
Self::UnknownHost(host_id) => format!("unknown host `{host_id}`"),
|
||||
Self::UnknownWorker {
|
||||
runtime_id,
|
||||
worker_id,
|
||||
} => format!("unknown worker `{worker_id}` in runtime `{runtime_id}`"),
|
||||
Self::RuntimeOperationFailed { message, .. } => message.clone(),
|
||||
}
|
||||
}
|
||||
|
||||
pub fn into_error(self) -> Error {
|
||||
match self {
|
||||
Self::InvalidIdentifier { kind, value } => Error::InvalidRuntimeIdentifier {
|
||||
@@ -658,6 +673,18 @@ pub trait WorkspaceWorkerRuntime: Send + Sync {
|
||||
}
|
||||
}
|
||||
|
||||
fn send_protocol_method(
|
||||
&self,
|
||||
_worker_id: &str,
|
||||
_method: protocol::Method,
|
||||
) -> Result<Vec<protocol::Event>, RuntimeRegistryError> {
|
||||
Err(RuntimeRegistryError::RuntimeOperationFailed {
|
||||
runtime_id: self.runtime_id().to_string(),
|
||||
code: "worker_protocol_method_unsupported".to_string(),
|
||||
message: "runtime does not support Worker protocol command transport".to_string(),
|
||||
})
|
||||
}
|
||||
|
||||
fn stop_worker(
|
||||
&self,
|
||||
worker_id: &str,
|
||||
@@ -1062,6 +1089,26 @@ impl RuntimeRegistry {
|
||||
Ok(runtime.list_config_bundles())
|
||||
}
|
||||
|
||||
pub fn send_protocol_method(
|
||||
&self,
|
||||
runtime_id: &str,
|
||||
worker_id: &str,
|
||||
method: protocol::Method,
|
||||
) -> Result<Vec<protocol::Event>, RuntimeRegistryError> {
|
||||
validate_backend_identifier("runtime_id", runtime_id)?;
|
||||
validate_backend_identifier("worker_id", worker_id)?;
|
||||
let runtime = self.runtime(runtime_id)?;
|
||||
let lookup = runtime.worker(worker_id);
|
||||
if lookup.worker.is_none() {
|
||||
return Err(operation_failed_or_unknown_worker(
|
||||
runtime_id,
|
||||
worker_id,
|
||||
lookup.diagnostics,
|
||||
));
|
||||
}
|
||||
runtime.send_protocol_method(worker_id, method)
|
||||
}
|
||||
|
||||
pub fn send_input(
|
||||
&self,
|
||||
runtime_id: &str,
|
||||
@@ -1807,6 +1854,35 @@ impl WorkspaceWorkerRuntime for EmbeddedWorkerRuntime {
|
||||
))
|
||||
}
|
||||
|
||||
fn send_protocol_method(
|
||||
&self,
|
||||
worker_id: &str,
|
||||
method: protocol::Method,
|
||||
) -> Result<Vec<protocol::Event>, RuntimeRegistryError> {
|
||||
if !self.execution_enabled {
|
||||
return Err(RuntimeRegistryError::RuntimeOperationFailed {
|
||||
runtime_id: self.runtime_id.clone(),
|
||||
code: "embedded_worker_execution_unavailable".to_string(),
|
||||
message: format!(
|
||||
"worker protocol command for '{worker_id}' requires an embedded execution backend"
|
||||
),
|
||||
});
|
||||
}
|
||||
let Some(worker_ref) = self.worker_ref(worker_id) else {
|
||||
return Err(RuntimeRegistryError::UnknownWorker {
|
||||
runtime_id: self.runtime_id.clone(),
|
||||
worker_id: worker_id.to_string(),
|
||||
});
|
||||
};
|
||||
self.runtime
|
||||
.send_protocol_method(&worker_ref, method)
|
||||
.map_err(|error| RuntimeRegistryError::RuntimeOperationFailed {
|
||||
runtime_id: self.runtime_id.clone(),
|
||||
code: "embedded_worker_protocol_command_failed".to_string(),
|
||||
message: error.to_string(),
|
||||
})
|
||||
}
|
||||
|
||||
fn send_input(&self, worker_id: &str, request: WorkerInputRequest) -> WorkerInputResult {
|
||||
if !self.execution_enabled {
|
||||
return embedded_input_rejected(
|
||||
|
||||
@@ -555,6 +555,14 @@ pub fn build_router(api: WorkspaceApi) -> Router {
|
||||
"/api/w/{workspace_id}/runtimes/{runtime_id}/workers/{worker_id}/events/ws",
|
||||
get(scoped_worker_observation_ws),
|
||||
)
|
||||
.route(
|
||||
"/api/runtimes/{runtime_id}/workers/{worker_id}/protocol/ws",
|
||||
get(worker_protocol_ws),
|
||||
)
|
||||
.route(
|
||||
"/api/w/{workspace_id}/runtimes/{runtime_id}/workers/{worker_id}/protocol/ws",
|
||||
get(scoped_worker_protocol_ws),
|
||||
)
|
||||
.route("/api/hosts/{host_id}/workers", get(list_host_workers))
|
||||
.route(
|
||||
"/api/w/{workspace_id}/hosts/{host_id}/workers",
|
||||
@@ -2457,6 +2465,19 @@ async fn scoped_worker_observation_ws(
|
||||
.into_response()
|
||||
}
|
||||
|
||||
async fn scoped_worker_protocol_ws(
|
||||
ws: WebSocketUpgrade,
|
||||
State(api): State<WorkspaceApi>,
|
||||
AxumPath(path): AxumPath<ScopedRuntimeWorkerPath>,
|
||||
) -> Response {
|
||||
if let Err(err) = validate_workspace_scope(&api, &path.workspace_id) {
|
||||
return err.into_response();
|
||||
}
|
||||
worker_protocol_ws(State(api), AxumPath((path.runtime_id, path.worker_id)), ws)
|
||||
.await
|
||||
.into_response()
|
||||
}
|
||||
|
||||
async fn scoped_list_host_workers(
|
||||
State(api): State<WorkspaceApi>,
|
||||
AxumPath(path): AxumPath<ScopedHostPath>,
|
||||
@@ -3388,6 +3409,89 @@ async fn cancel_runtime_worker(
|
||||
Ok(Json(result))
|
||||
}
|
||||
|
||||
async fn worker_protocol_ws(
|
||||
State(api): State<WorkspaceApi>,
|
||||
AxumPath((runtime_id, worker_id)): AxumPath<(String, String)>,
|
||||
ws: WebSocketUpgrade,
|
||||
) -> impl IntoResponse {
|
||||
match api.runtime.worker(&runtime_id, &worker_id) {
|
||||
Ok(_) => ws.on_upgrade(move |socket| {
|
||||
worker_protocol_ws_session(api.runtime, runtime_id, worker_id, socket)
|
||||
}),
|
||||
Err(error) => ApiError::from(error.into_error()).into_response(),
|
||||
}
|
||||
}
|
||||
|
||||
async fn worker_protocol_ws_session(
|
||||
runtime: Arc<RuntimeRegistry>,
|
||||
runtime_id: String,
|
||||
worker_id: String,
|
||||
mut socket: WebSocket,
|
||||
) {
|
||||
while let Some(frame) = socket.next().await {
|
||||
match frame {
|
||||
Ok(WsMessage::Text(text)) => match serde_json::from_str::<protocol::Method>(&text) {
|
||||
Ok(method) => match runtime.send_protocol_method(&runtime_id, &worker_id, method) {
|
||||
Ok(events) => {
|
||||
for event in events {
|
||||
if !send_protocol_event(&mut socket, &event).await {
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
Err(error) => {
|
||||
let event = protocol_error_event(error.message());
|
||||
if !send_protocol_event(&mut socket, &event).await {
|
||||
return;
|
||||
}
|
||||
}
|
||||
},
|
||||
Err(error) => {
|
||||
let event =
|
||||
protocol_error_event(format!("malformed protocol method frame: {error}"));
|
||||
if !send_protocol_event(&mut socket, &event).await {
|
||||
return;
|
||||
}
|
||||
}
|
||||
},
|
||||
Ok(WsMessage::Close(_)) => return,
|
||||
Ok(WsMessage::Ping(payload)) => {
|
||||
if socket.send(WsMessage::Pong(payload)).await.is_err() {
|
||||
return;
|
||||
}
|
||||
}
|
||||
Ok(WsMessage::Pong(_)) | Ok(WsMessage::Binary(_)) => {}
|
||||
Err(error) => {
|
||||
let event = protocol_error_event(format!("protocol WebSocket error: {error}"));
|
||||
let _ = send_protocol_event(&mut socket, &event).await;
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
async fn send_protocol_event(socket: &mut WebSocket, event: &protocol::Event) -> bool {
|
||||
match serde_json::to_string(event) {
|
||||
Ok(text) => socket.send(WsMessage::Text(text.into())).await.is_ok(),
|
||||
Err(error) => {
|
||||
let fallback = protocol_error_event(format!(
|
||||
"failed to serialize protocol response event: {error}"
|
||||
));
|
||||
let Ok(text) = serde_json::to_string(&fallback) else {
|
||||
return false;
|
||||
};
|
||||
socket.send(WsMessage::Text(text.into())).await.is_ok()
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
fn protocol_error_event(message: impl Into<String>) -> protocol::Event {
|
||||
protocol::Event::Error {
|
||||
code: protocol::ErrorCode::Internal,
|
||||
message: message.into(),
|
||||
}
|
||||
}
|
||||
|
||||
async fn worker_observation_ws(
|
||||
State(api): State<WorkspaceApi>,
|
||||
AxumPath((runtime_id, worker_id)): AxumPath<(String, String)>,
|
||||
|
||||
Reference in New Issue
Block a user