blomios/IdeA
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(orchestrator,agent): ask_agent survit au restart de la cible same-project et cross-project — #289 (RestartRendezvousCoordinator stable par RuntimeAgentKey (projectId, agentId) partagé entre le rendez-vous ask_structured et le scheduler AppAgentSelfRestarter: l'ask actif garde verrou de tour, ticket logique et conversation_id, abandonne le drain de l'ancienne session sans no-reply puis capture le Final de continuation de la session relancée (SELF_RESTART_CONTINUATION_PROMPT partagé, un seul tour de continuation); branche delegated du scheduler: pas de preempt/cancel_head/mark_idle locaux, ready + wait_for_handoff, échecs restart propagés explicitement et cancel en cascade sans relance de continuation; watchdog d'inactivité réarmé par l'activité de contrôle restart; corrélation cross-project (requesterProjectId, requesterAgentId, correlationId) conservée jusqu'au résultat post-restart puis supprimée; stop_agent d'une cible en restart annule le rendez-vous au lieu d'un NotFound; le scheduler standalone conserve son comportement historique sans ask actif; QA verte: orchestrator_service 87/87 dont 6 tests restart (final post-restart, nettoyage busy/workstate/registre, corrélation cross-project, failure explicite, cancel cross-project + stop same-project pendant restart), application lib 171/171 dont watchdog réarmé, backend 89+7 ignored, cargo check -p backend, fmt --check; réserves: pas d'E2E AppImage/vrai provider, 7 tests backend ignorés bind socket local; scripts/ hors périmètre; aucun push)

Browse Source
This commit is contained in:
Blomios
2026-09-08 22:53:52 +02:00
parent db91b00699
commit d6d989187b
9 changed files with 1204 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

197
.ideai/tickets/289/carnet.md Normal file
View File

@ -0,0 +1,197 @@
---
issueRef: "#289"
version: 4
updatedBy: {"kind":"agent","agent_id":"a6c6ea12-bfc6-4bdc-8031-324102dfa34d"}
updatedAt: 1788900709736
---
## 2026-09-08 — Main
Demande utilisateur : garantir que si une cible appelée via `ask_agent` se restart, le demandeur attend la fin du restart et reçoit le travail post-restart rattaché au même agent, aussi bien intra-projet qu'en cross-project.
Cadrage Architect :
- Identité stable par `RuntimeAgentKey` `(projectId, agentId)`.
- Le `conversation_id` logique et le verrou de tour doivent survivre au restart.
- Un restart accepté est une activité de contrôle du rendez-vous : ne pas classifier la fermeture de l'ancienne session comme no-reply/timeout d'inactivité.
- Cross-project : conserver la corrélation demandeur `(requesterProjectId, requesterAgentId, correlationId)` jusqu'au résultat post-restart.
- Cancel pendant restart doit cascader et nettoyer rendez-vous/workstate/registre cross-project.
- Échec de restart => erreur explicite, pas silence.
Branche Git retenue : `fix/289-ask-agent-during-target-restart` depuis `develop` @ `db91b00`.
Prochaines étapes : DevBackend implémente le contrat backend et tests ; QA valide same-project, cross-project, watchdog, failure/cancel.
## 2026-09-08 — DevBackend
Correctif backend implémenté sur `fix/289-ask-agent-during-target-restart`.
Fichiers modifiés :
- `crates/application/src/orchestrator/service.rs`
- `crates/application/src/orchestrator/mod.rs`
- `crates/application/src/lib.rs`
- `crates/application/tests/orchestrator_service.rs`
- `crates/backend/src/lib.rs`
Comportement livré :
- coordination stable par `RuntimeAgentKey (projectId, agentId)` entre le rendez-vous applicatif et le scheduler de self-restart ;
- conservation du verrou de tour, du ticket logique et du `conversation_id` pendant le remplacement de la session concrète ;
- abandon du drain de l'ancienne session sans no-reply, puis capture du `Final` du tour de continuation sur la session relancée ;
- activité de contrôle restart intégrée au watchdog d'inactivité ;
- corrélation cross-project conservée jusqu'au résultat post-restart ;
- échec de restart remonté explicitement ; cancel pendant la fenêtre sans runtime réveille l'ask, nettoie la corrélation et empêche la relance de continuation ;
- le scheduler standalone conserve son comportement historique lorsque aucun ask n'est actif.
Tests exécutés :
- `cargo test -p application --test orchestrator_service` — 85 passés ;
- `cargo test -p application --lib` — 171 passés ;
- `cargo test -p backend --lib` — 89 passés, 7 ignorés (tests socket nécessitant une permission locale) ;
- `cargo check -p backend` — OK ;
- `cargo fmt --all -- --check` — OK.
Tests de régression ajoutés : same-project Final post-restart, cross-project + nettoyage corrélation, faux no-reply interdit, failure explicite, cancel pendant le gap de restart, watchdog réarmé par activité de contrôle.
Limites : pas de test E2E avec un vrai processus fournisseur/AppImage dans ce lot ; la validation est faite avec les ports/sessions structurés et le scheduler backend compilé. Aucun fichier frontend modifié. Aucun commit effectué par DevBackend.
## 2026-09-08 — QA
Verdict : VERT pour le ticket #289 sur `fix/289-ask-agent-during-target-restart`.
Tests QA ajoutés dans `crates/application/tests/orchestrator_service.rs` :
- `ask_target_restart_cleans_busy_workstate_and_replaces_structured_registry_entry` : succès same-project post-restart, cible revenue `Idle`, FIFO vide, live-state `Done`, ancienne session structurée absente, session de remplacement enregistrée sur le même agent logique.
- `same_project_stop_during_restart_releases_ask_and_leaves_no_replacement_session` : cancel/stop same-project pendant le gap de restart, ask réveillé en erreur explicite de cancel, busy/FIFO/live-state nettoyés, aucune session de remplacement enregistrée.
Commandes exécutées et sorties pertinentes :
```bash
git status --short --branch
```
```text
## fix/289-ask-agent-during-target-restart
M .ideai/tickets/counter.json
M .ideai/tickets/index.json
M crates/application/src/lib.rs
M crates/application/src/orchestrator/mod.rs
M crates/application/src/orchestrator/service.rs
M crates/application/tests/orchestrator_service.rs
M crates/backend/src/lib.rs
?? .ideai/tickets/289/
?? scripts/
```
```bash
cargo test -p application --test orchestrator_service ask_target_restart_keeps_logical_turn_and_returns_post_restart_final -- --nocapture
```
```text
test ask_target_restart_keeps_logical_turn_and_returns_post_restart_final ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 84 filtered out; finished in 0.02s
```
```bash
cargo test -p application --test orchestrator_service cross_project_ask_preserves_correlation_across_target_restart -- --nocapture
```
```text
test cross_project_ask_preserves_correlation_across_target_restart ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 84 filtered out; finished in 0.02s
```
```bash
cargo test -p application --test orchestrator_service ask_target_restart_failure_is_explicit_not_no_reply -- --nocapture
```
```text
test ask_target_restart_failure_is_explicit_not_no_reply ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 84 filtered out; finished in 0.01s
```
```bash
cargo test -p application --test orchestrator_service cross_project_cancel_during_restart_releases_ask_and_correlation -- --nocapture
```
```text
test cross_project_cancel_during_restart_releases_ask_and_correlation ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 84 filtered out; finished in 0.00s
```
```bash
cargo test -p application --lib restart_control_activity_rearms_inactivity_watchdog -- --nocapture
```
```text
test orchestrator::service::tests::restart_control_activity_rearms_inactivity_watchdog ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 170 filtered out; finished in 0.05s
```
```bash
cargo test -p application --test orchestrator_service ask_target_restart_cleans_busy_workstate_and_replaces_structured_registry_entry -- --nocapture
```
```text
test ask_target_restart_cleans_busy_workstate_and_replaces_structured_registry_entry ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 86 filtered out; finished in 0.02s
```
```bash
cargo test -p application --test orchestrator_service same_project_stop_during_restart_releases_ask_and_leaves_no_replacement_session -- --nocapture
```
```text
test same_project_stop_during_restart_releases_ask_and_leaves_no_replacement_session ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 86 filtered out; finished in 0.00s
```
```bash
cargo fmt --all -- --check
```
```text
OK (aucune sortie)
```
```bash
cargo test -p application --test orchestrator_service
```
```text
running 87 tests
test result: ok. 87 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.09s
```
```bash
cargo test -p application --lib
```
```text
running 171 tests
test result: ok. 171 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.15s
```
```bash
cargo test -p backend --lib
```
```text
running 96 tests
test result: ok. 89 passed; 0 failed; 7 ignored; 0 measured; 0 filtered out; finished in 0.04s
```
```bash
cargo check -p backend
```
```text
Finished `dev` profile [unoptimized + debuginfo] target(s) in 3.37s
```
Couverture validée :
- same-project : Final récupéré après restart et test QA de nettoyage busy/workstate/registre.
- cross-project : corrélation conservée jusqu'au Final post-restart puis supprimée.
- pas de faux no-reply/timeout pendant restart : failure restart reste `PROCESS` explicite, watchdog réarmé par activité de contrôle restart.
- failure restart : erreur explicite contenant la cause provider.
- cancel pendant restart : cross-project et same-project couverts.
- nettoyage busy/workstate/registry : FIFO vide, `AgentBusyState::Idle`, live-state `Done`, registre structuré remplacé après succès et non recréé après cancel.
Risques résiduels :
- Pas d'E2E AppImage / vrai provider relancé : la preuve porte sur les ports applicatifs, les sessions structurées de test et le câblage backend compilé.
- Les 7 tests backend ignorés demandent une permission de bind socket locale ; ils restent hors preuve dans cet environnement.

17
.ideai/tickets/289/issue.md Normal file
View File

@ -0,0 +1,17 @@
---
id: "ae7f4f57-087e-463e-9390-33727e751d4f"
number: 289
title: "Garantir ask_agent pendant restart de la cible, same-project et cross-project"
status: "qa"
priority: "high"
sprint: null
links: []
agentRefs: []
attachments: []
createdBy: {"kind":"agent","agent_id":"a6c6ea12-bfc6-4bdc-8031-324102dfa34d"}
updatedBy: {"kind":"agent","agent_id":"a6c6ea12-bfc6-4bdc-8031-324102dfa34d"}
createdAt: 1788899102915
updatedAt: 1788900574986
version: 3
---
Vérifier et corriger le comportement suivant : lorsqu'un agent cible a été appelé via ask_agent puis se restart, le demandeur doit attendre la fin du restart et le travail repris/après restart doit rester rattaché à cette même cible. Le contrat doit couvrir les appels intra-projet et cross-project, sans perte de résultat ni attribution à un mauvais agent.

2
.ideai/tickets/counter.json
View File

@ -1,3 +1,3 @@
{
"nextNumber": 289
"nextNumber": 290
}

14
.ideai/tickets/index.json
View File

@ -3949,6 +3949,20 @@
"agent_id": "a6c6ea12-bfc6-4bdc-8031-324102dfa34d"
},
"updatedAt": 1788898181382
},
{
"issueRef": "#289",
"path": "289",
"title": "Garantir ask_agent pendant restart de la cible, same-project et cross-project",
"status": "qa",
"priority": "high",
"sprint": null,
"assignedAgentIds": [],
"createdBy": {
"kind": "agent",
"agent_id": "a6c6ea12-bfc6-4bdc-8031-324102dfa34d"
},
"updatedAt": 1788900574986
}
]
}

5
crates/application/src/lib.rs
View File

@ -168,8 +168,9 @@ pub use orchestrator::{
AgentRestartScheduled, AgentSelfRestartScheduler, AgentWakeService, AskLivenessProbe,
ContextGuardUseCases, LiveStateProvider, LiveStateReadProvider, McpRuntimeProvider,
OrchestratorOutcome, OrchestratorService, ProposeContext, ReadContext, ReadMemory,
RecordTurnProvider, WakeSessionProvider, WatchdogOutcome, WriteMemory,
DEFAULT_RENDEZVOUS_CEILING, DEFAULT_RENDEZVOUS_WINDOW,
RecordTurnProvider, RestartRendezvousCoordinator, RestartRendezvousState, WakeSessionProvider,
WatchdogOutcome, WriteMemory, DEFAULT_RENDEZVOUS_CEILING, DEFAULT_RENDEZVOUS_WINDOW,
SELF_RESTART_CONTINUATION_PROMPT,
};
pub use permission::{
GetProjectPermissions, GetProjectPermissionsInput, GetProjectPermissionsOutput,

3
crates/application/src/orchestrator/mod.rs
View File

@ -21,6 +21,7 @@ pub use rendezvous::{
pub use service::{
AgentRestartScheduled, AgentSelfRestartScheduler, AskLivenessProbe, ContextGuardUseCases,
LiveStateProvider, LiveStateReadProvider, McpRuntimeProvider, OrchestratorOutcome,
OrchestratorService, RecordTurnProvider,
OrchestratorService, RecordTurnProvider, RestartRendezvousCoordinator, RestartRendezvousState,
SELF_RESTART_CONTINUATION_PROMPT,
};
pub use wake::{AgentWakeService, WakeSessionProvider};

556
crates/application/src/orchestrator/service.rs
View File

@ -19,7 +19,7 @@ use std::sync::atomic::{AtomicU64, Ordering as AtomicOrdering};
use std::sync::{Arc, Mutex as StdMutex};
use std::time::{Duration, Instant};
use tokio::sync::Mutex as AsyncMutex;
use tokio::sync::{watch, Mutex as AsyncMutex};
use domain::conversation::{ConversationParty, ConversationRegistry, SessionRef, WaitForGraph};
use domain::conversation_log::{ConversationTurn, TurnId, TurnRole};
@ -82,6 +82,330 @@ pub trait AgentSelfRestartScheduler: Send + Sync {
) -> Result<AgentRestartScheduled, AppError>;
}
/// Internal continuation prompt used when a self-restart is handed back to the
/// synchronous inter-agent rendezvous that was driving the old runtime.
pub const SELF_RESTART_CONTINUATION_PROMPT: &str =
"IdeA vient de te redémarrer. Reprends immédiatement le travail en cours là où tu l'avais laissé.";
/// Control-plane state of a self-restart affecting an active logical rendezvous.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum RestartRendezvousState {
/// No restart currently affects the logical turn.
Idle,
/// The old runtime is being replaced.
Restarting {
/// Stable restart correlation id.
restart_id: String,
},
/// The replacement structured session is available.
Ready {
/// Stable restart correlation id.
restart_id: String,
},
/// Runtime replacement failed.
Failed {
/// Stable restart correlation id.
restart_id: String,
/// Explicit adapter failure.
message: String,
},
/// The owning logical rendezvous was cancelled.
Cancelled {
/// Stable restart correlation id.
restart_id: String,
},
}
struct RestartRendezvousEntry {
active_waiters: usize,
activity: u64,
state: RestartRendezvousState,
tx: watch::Sender<RestartRendezvousState>,
}
/// Stable, runtime-keyed seam shared by the orchestrator and the detached restart
/// adapter. It keeps a logical ask alive while its concrete provider session is
/// replaced.
#[derive(Default)]
pub struct RestartRendezvousCoordinator {
entries: StdMutex<HashMap<RuntimeAgentKey, RestartRendezvousEntry>>,
}
impl RestartRendezvousCoordinator {
/// Creates an empty runtime-keyed coordinator.
#[must_use]
pub fn new() -> Self {
Self::default()
}
fn register(self: &Arc<Self>, key: RuntimeAgentKey) -> RestartRendezvousRegistration {
let mut entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let entry = entries.entry(key).or_insert_with(|| {
let state = RestartRendezvousState::Idle;
let (tx, _) = watch::channel(state.clone());
RestartRendezvousEntry {
active_waiters: 0,
activity: 0,
state,
tx,
}
});
entry.active_waiters = entry.active_waiters.saturating_add(1);
RestartRendezvousRegistration {
coordinator: Arc::clone(self),
key,
rx: entry.tx.subscribe(),
}
}
/// Announces a restart synchronously. Returns whether an active ask owns the
/// continuation and therefore must receive the post-restart `Final`.
pub fn begin(&self, key: RuntimeAgentKey, restart_id: String) -> bool {
let mut entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let Some(entry) = entries.get_mut(&key) else {
return false;
};
if entry.active_waiters == 0 {
return false;
}
entry.activity = entry.activity.saturating_add(1);
entry.state = RestartRendezvousState::Restarting { restart_id };
entry.tx.send_replace(entry.state.clone());
true
}
/// Publishes that the replacement structured session can accept continuation.
pub fn ready(&self, key: RuntimeAgentKey, restart_id: &str) {
self.transition(key, restart_id, |restart_id| {
RestartRendezvousState::Ready { restart_id }
});
}
fn resume_taken(&self, key: RuntimeAgentKey, restart_id: &str) {
let mut entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let Some(entry) = entries.get_mut(&key) else {
return;
};
if !matches!(
&entry.state,
RestartRendezvousState::Ready { restart_id: current } if current == restart_id
) {
return;
}
entry.state = RestartRendezvousState::Idle;
entry.activity = entry.activity.saturating_add(1);
entry.tx.send_replace(entry.state.clone());
}
/// Publishes an explicit restart failure to the awaiting logical rendezvous.
pub fn fail(&self, key: RuntimeAgentKey, restart_id: &str, message: String) {
self.transition(key, restart_id, |restart_id| {
RestartRendezvousState::Failed {
restart_id,
message,
}
});
}
/// Cancels the restart-aware continuation for a target being explicitly stopped.
/// Returns whether a restart handoff was active.
pub fn cancel(&self, key: RuntimeAgentKey) -> bool {
let mut entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let Some(entry) = entries.get_mut(&key) else {
return false;
};
let restart_id = match &entry.state {
RestartRendezvousState::Restarting { restart_id }
| RestartRendezvousState::Ready { restart_id } => restart_id.clone(),
_ => return false,
};
entry.state = RestartRendezvousState::Cancelled { restart_id };
entry.activity = entry.activity.saturating_add(1);
entry.tx.send_replace(entry.state.clone());
true
}
fn transition(
&self,
key: RuntimeAgentKey,
restart_id: &str,
next: impl FnOnce(String) -> RestartRendezvousState,
) {
let mut entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let Some(entry) = entries.get_mut(&key) else {
return;
};
let matches = match &entry.state {
RestartRendezvousState::Restarting {
restart_id: current,
}
| RestartRendezvousState::Ready {
restart_id: current,
} => current == restart_id,
_ => false,
};
if !matches {
return;
}
entry.activity = entry.activity.saturating_add(1);
entry.state = next(restart_id.to_owned());
entry.tx.send_replace(entry.state.clone());
}
#[must_use]
/// Reports whether the owning rendezvous was cancelled during this restart.
pub fn is_cancelled(&self, key: RuntimeAgentKey, restart_id: &str) -> bool {
self.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.get(&key)
.is_some_and(|entry| {
matches!(
&entry.state,
RestartRendezvousState::Cancelled { restart_id: current }
if current == restart_id
)
})
}
/// Waits until the active ask has accepted the replacement session handoff.
pub async fn wait_for_handoff(
&self,
key: RuntimeAgentKey,
restart_id: &str,
) -> Result<(), String> {
let mut rx = {
let entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
entries
.get(&key)
.ok_or_else(|| "restart rendezvous disappeared".to_owned())?
.tx
.subscribe()
};
loop {
let state = rx.borrow_and_update().clone();
match state {
RestartRendezvousState::Idle => return Ok(()),
RestartRendezvousState::Cancelled {
restart_id: current,
} if current == restart_id => {
return Err("restart rendezvous was cancelled".to_owned());
}
RestartRendezvousState::Failed {
restart_id: current,
message,
} if current == restart_id => return Err(message),
_ => {}
}
rx.changed()
.await
.map_err(|_| "restart rendezvous channel closed".to_owned())?;
}
}
/// Releases terminal coordination state after the detached scheduler observed it.
pub fn finish(&self, key: RuntimeAgentKey, restart_id: &str) {
let mut entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let remove = entries.get(&key).is_some_and(|entry| {
entry.active_waiters == 0
&& matches!(
&entry.state,
RestartRendezvousState::Cancelled { restart_id: current }
| RestartRendezvousState::Failed { restart_id: current, .. }
if current == restart_id
)
});
if remove {
entries.remove(&key);
}
}
/// A restarting target is control-plane activity on every watchdog probe, even
/// when provider/transcript activity is temporarily impossible during teardown.
fn activity_token(&self, key: RuntimeAgentKey) -> Option<u64> {
let mut entries = self
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let entry = entries.get_mut(&key)?;
if matches!(
entry.state,
RestartRendezvousState::Restarting { .. } | RestartRendezvousState::Ready { .. }
) {
entry.activity = entry.activity.saturating_add(1);
}
Some(entry.activity)
}
}
struct RestartRendezvousRegistration {
coordinator: Arc<RestartRendezvousCoordinator>,
key: RuntimeAgentKey,
rx: watch::Receiver<RestartRendezvousState>,
}
impl RestartRendezvousRegistration {
fn state(&mut self) -> RestartRendezvousState {
self.rx.borrow_and_update().clone()
}
async fn changed(&mut self) -> Result<(), watch::error::RecvError> {
self.rx.changed().await
}
}
impl Drop for RestartRendezvousRegistration {
fn drop(&mut self) {
let mut entries = self
.coordinator
.entries
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
let Some(entry) = entries.get_mut(&self.key) else {
return;
};
entry.active_waiters = entry.active_waiters.saturating_sub(1);
if entry.active_waiters == 0 {
let restart_id = match &entry.state {
RestartRendezvousState::Restarting { restart_id }
| RestartRendezvousState::Ready { restart_id } => Some(restart_id.clone()),
_ => None,
};
if let Some(restart_id) = restart_id {
entry.state = RestartRendezvousState::Cancelled { restart_id };
entry.activity = entry.activity.saturating_add(1);
entry.tx.send_replace(entry.state.clone());
} else if matches!(
entry.state,
RestartRendezvousState::Idle | RestartRendezvousState::Failed { .. }
) {
entries.remove(&self.key);
}
}
}
}
/// Default terminal geometry for an orchestrator-launched agent cell. The UI
/// resizes the PTY to the real cell size on attach; these are sane starting rows
/// /cols so the spawn never fails on a zero-sized terminal.
@ -706,6 +1030,8 @@ pub struct OrchestratorService {
spawn_background_command: Option<Arc<SpawnBackgroundCommand>>,
/// Detached self-restart coordinator, wired at the composition root.
self_restart: Option<Arc<dyn AgentSelfRestartScheduler>>,
/// Stable control-plane handoff between an active ask and a detached restart.
restart_rendezvous: Arc<RestartRendezvousCoordinator>,
}
/// Bundle des quatre use cases C7 sous [`domain::fileguard::FileGuard`], injectés
@ -806,6 +1132,7 @@ impl OrchestratorService {
session_limits: None,
spawn_background_command: None,
self_restart: None,
restart_rendezvous: Arc::new(RestartRendezvousCoordinator::new()),
}
}
@ -837,6 +1164,16 @@ impl OrchestratorService {
self
}
#[must_use]
/// Shares the runtime-keyed restart handoff with the detached restart adapter.
pub fn with_restart_rendezvous(
mut self,
coordinator: Arc<RestartRendezvousCoordinator>,
) -> Self {
self.restart_rendezvous = coordinator;
self
}
/// Branche le store des tâches de fond pour tracer les rendez-vous inter-agents
/// comme [`BackgroundTaskKind::HeadlessRendezvous`] (B6).
///
@ -2393,7 +2730,7 @@ impl OrchestratorService {
conversation_id,
requester,
task,
session.as_ref(),
session,
)
.await
}
@ -2419,7 +2756,7 @@ impl OrchestratorService {
conversation_id: domain::conversation::ConversationId,
requester: Option<AgentId>,
task: String,
session: &dyn domain::ports::AgentSession,
session: Arc<dyn domain::ports::AgentSession>,
) -> Result<OrchestratorOutcome, AppError> {
let (input, mailbox) = match (&self.input, &self.mailbox) {
(Some(i), Some(m)) => (i, m),
@ -2524,57 +2861,131 @@ impl OrchestratorService {
ticket: ticket_id,
});
let local_activity = Arc::new(AtomicU64::new(0));
let drain = drain_with_readiness_announcements_and_activity_outcome(
session,
&task,
None,
input.as_ref(),
agent_key,
announcement_publisher,
Some(Arc::clone(&local_activity)),
let mut restart = self.restart_rendezvous.register(agent_key);
let structured = Arc::clone(
self.structured
.as_ref()
.expect("ask_structured requires the structured registry"),
);
let wait_activity = Arc::clone(&local_activity);
// L'attente du rendez-vous structured, rendue comme `Result<String, AppError>`
// pour être enveloppée par le watchdog. Un flux clos sans `Final` correspond
// au no-reply du chemin headless : la cible a rendu la main sans réponse
// exploitable, on expose donc l'erreur métier retryable plutôt qu'un PROCESS.
// The concrete provider session may disappear under us when the target calls
// `idea_restart_self`. The logical rendezvous remains registered by
// RuntimeAgentKey; on the restart control signal we drop only the old drain,
// wait for the replacement session, then capture its continuation `Final`.
let wait = async {
match drain.await {
Ok(TurnOutcome::Completed(content)) => Ok(content),
Ok(TurnOutcome::RateLimited { resets_at_ms }) => {
let conversation_id = session.conversation_id();
let (conversation_id, resets_at_ms) = match (conversation_id, resets_at_ms) {
(Some(conversation_id), resets_at_ms) => {
(Some(conversation_id), resets_at_ms)
let mut current_session = session;
let mut prompt = task.clone();
loop {
let turn = {
let drain = drain_with_readiness_announcements_and_activity_outcome(
current_session.as_ref(),
&prompt,
None,
input.as_ref(),
agent_key,
announcement_publisher.clone(),
Some(Arc::clone(&wait_activity)),
);
tokio::pin!(drain);
tokio::select! {
biased;
changed = restart.changed() => {
changed.map_err(|_| AppError::Process(
format!("restart coordination closed for target {target}")
))?;
None
}
(None, None) => (None, None),
// Reset connu mais conversation moteur absente : ne pas armer
// une reprise non reprenable ; surfacer le fallback humain.
(None, Some(_)) => (None, None),
};
if let (Some(service), Some(structured)) =
(&self.session_limits, &self.structured)
{
if let Some(node_id) =
structured.node_for_agent_in_project(project.id, &agent_id)
{
service.on_rate_limited(
project.id,
agent_id,
node_id,
conversation_id,
resets_at_ms,
);
outcome = &mut drain => Some(outcome),
}
};
let control_active = !matches!(restart.state(), RestartRendezvousState::Idle);
if let Some(outcome) = turn.filter(|_| !control_active) {
match outcome {
Ok(TurnOutcome::Completed(content)) => return Ok(content),
Ok(TurnOutcome::RateLimited { resets_at_ms }) => {
let conversation_id = current_session.conversation_id();
let (conversation_id, resets_at_ms) =
match (conversation_id, resets_at_ms) {
(Some(conversation_id), resets_at_ms) => {
(Some(conversation_id), resets_at_ms)
}
(None, None) => (None, None),
// Reset connu mais conversation moteur absente : ne pas armer
// une reprise non reprenable ; surfacer le fallback humain.
(None, Some(_)) => (None, None),
};
if let (Some(service), Some(structured)) =
(&self.session_limits, &self.structured)
{
if let Some(node_id) =
structured.node_for_agent_in_project(project.id, &agent_id)
{
service.on_rate_limited(
project.id,
agent_id,
node_id,
conversation_id,
resets_at_ms,
);
}
}
return Err(AppError::Process(
"le tour s'est clos en limite de session, sans contenu Final"
.to_owned(),
));
}
Err(err) if structured_no_reply_error(&err) => {
return Err(AppError::TargetReturnedNoReply(target.to_owned()));
}
Err(err) => return Err(AppError::from(err)),
}
}
loop {
match restart.state() {
RestartRendezvousState::Restarting { .. } => {
restart.changed().await.map_err(|_| {
AppError::Process(format!(
"restart coordination closed for target {target}"
))
})?;
}
RestartRendezvousState::Ready { restart_id } => {
current_session = structured
.session_for_agent_in_project(project.id, &agent_id)
.ok_or_else(|| {
AppError::Process(format!(
"restart completed without a live structured session for target {target}"
))
})?;
self.bind_conversation_session(conversation_id, current_session.id());
self.restart_rendezvous.resume_taken(agent_key, &restart_id);
let _ = restart.state();
prompt = SELF_RESTART_CONTINUATION_PROMPT.to_owned();
wait_activity.fetch_add(1, AtomicOrdering::Relaxed);
break;
}
RestartRendezvousState::Failed { message, .. } => {
return Err(AppError::Process(format!(
"restart failed for target {target}: {message}"
)));
}
RestartRendezvousState::Cancelled { .. } => {
return Err(AppError::Process(format!(
"restart cancelled for target {target}"
)));
}
RestartRendezvousState::Idle => {
restart.changed().await.map_err(|_| {
AppError::Process(format!(
"restart coordination closed for target {target}"
))
})?;
}
}
Err(AppError::Process(
"le tour s'est clos en limite de session, sans contenu Final".to_owned(),
))
}
Err(err) if structured_no_reply_error(&err) => {
Err(AppError::TargetReturnedNoReply(target.to_owned()))
}
Err(err) => Err(AppError::from(err)),
}
};
@ -2590,6 +3001,7 @@ impl OrchestratorService {
started,
local: local_activity,
},
Some(agent_key),
)
.await
{
@ -3278,6 +3690,10 @@ impl OrchestratorService {
}
}
let restart_cancelled = self
.restart_rendezvous
.cancel(runtime_key(project, agent_id));
let structured = self
.structured
.clone()
@ -3317,6 +3733,12 @@ impl OrchestratorService {
.await?
.is_empty()
{
if restart_cancelled {
return Ok(OrchestratorOutcome {
detail: format!("cancelled restarting work for agent {label}"),
reply: None,
});
}
return Err(AppError::NotFound(format!(
"running session for agent {label}"
)));
@ -3547,6 +3969,7 @@ impl OrchestratorService {
agent_id: AgentId,
target: &str,
activity: AskWatchdogActivity,
restart_key: Option<RuntimeAgentKey>,
) -> WatchdogOutcome<R> {
let AskWatchdogActivity {
started,
@ -3559,11 +3982,13 @@ impl OrchestratorService {
#[derive(Default)]
struct CompositeActivity {
local: u64,
restart: u64,
transcript: Option<u64>,
token: u64,
}
let probe = self.ask_liveness_probe.clone();
let restart_rendezvous = Arc::clone(&self.restart_rendezvous);
let root = project_root.clone();
let composite = Arc::new(StdMutex::new(CompositeActivity::default()));
let probe_fn = move || {
@ -3571,20 +3996,26 @@ impl OrchestratorService {
let root = root.clone();
let local_activity = Arc::clone(&local_activity);
let composite = Arc::clone(&composite);
let restart_rendezvous = Arc::clone(&restart_rendezvous);
async move {
let transcript = match &probe {
Some(p) => p(root, agent_id).await,
None => None,
};
let local = local_activity.load(AtomicOrdering::Relaxed);
let restart = restart_key
.and_then(|key| restart_rendezvous.activity_token(key))
.unwrap_or(0);
let mut state = composite.lock().unwrap();
let local_delta = local.saturating_sub(state.local);
let restart_delta = restart.saturating_sub(state.restart);
let transcript_delta = match (state.transcript, transcript) {
(Some(previous), Some(current)) => current.saturating_sub(previous),
(None, Some(_)) => 1,
_ => 0,
};
state.local = state.local.max(local);
state.restart = state.restart.max(restart);
if let Some(current) = transcript {
state.transcript =
Some(state.transcript.map_or(current, |old| old.max(current)));
@ -3592,6 +4023,7 @@ impl OrchestratorService {
state.token = state
.token
.saturating_add(local_delta)
.saturating_add(restart_delta)
.saturating_add(transcript_delta);
Some(state.token)
}
@ -4057,4 +4489,34 @@ mod tests {
"pas de cancel_head après disarm"
);
}
#[tokio::test]
async fn restart_control_activity_rearms_inactivity_watchdog() {
let coordinator = Arc::new(RestartRendezvousCoordinator::new());
let key = rkey(9);
let _registration = coordinator.register(key);
assert!(coordinator.begin(key, "restart-watchdog".to_owned()));
let probe = Arc::clone(&coordinator);
let outcome = run_inactivity_watchdog(
async {
tokio::time::sleep(Duration::from_millis(45)).await;
"continued"
},
Duration::from_millis(10),
Duration::from_millis(100),
Instant::now(),
true,
move || {
let probe = Arc::clone(&probe);
async move { probe.activity_token(key) }
},
|_| {},
|_| {},
|_| {},
)
.await;
assert!(matches!(outcome, WatchdogOutcome::Resolved("continued")));
}
}

387
crates/application/tests/orchestrator_service.rs
View File

@ -2022,6 +2022,7 @@ struct AskFixture {
/// assert the target's Working/Done transitions and the upsert count.
live: Arc<RecordingLiveState>,
structured: Arc<StructuredSessions>,
restart_rendezvous: Arc<application::RestartRendezvousCoordinator>,
}
struct CompletionSession {
@ -2198,6 +2199,7 @@ fn ask_fixture_full(
let bus = SpyBus::default();
let mailbox = Arc::new(TestMailbox::new());
let structured = Arc::new(StructuredSessions::new());
let restart_rendezvous = Arc::new(application::RestartRendezvousCoordinator::new());
let completion_factory = completion_factory_for_contexts(&contexts, mailbox.completions());
seed_structured_sessions(&structured, &contexts, mailbox.completions());
let memories = Arc::new(HarvestMemories {
@ -2270,6 +2272,7 @@ fn ask_fixture_full(
.with_conversations(conversations)
.with_events(Arc::new(bus.clone()))
.with_structured(Arc::clone(&structured))
.with_restart_rendezvous(Arc::clone(&restart_rendezvous))
.with_memory_harvest(Arc::new(HarvestMemoryFromTurn::new(
Arc::clone(&memories) as Arc<dyn MemoryStore>,
Arc::new(bus.clone()),
@ -2302,6 +2305,7 @@ fn ask_fixture_full(
memories,
live,
structured,
restart_rendezvous,
}
}
@ -4412,6 +4416,7 @@ fn ask_fixture_with_record(
let bus = SpyBus::default();
let mailbox = Arc::new(TestMailbox::new());
let structured = Arc::new(StructuredSessions::new());
let restart_rendezvous = Arc::new(application::RestartRendezvousCoordinator::new());
let completion_factory = completion_factory_for_contexts(&contexts, mailbox.completions());
seed_structured_sessions(&structured, &contexts, mailbox.completions());
@ -4473,6 +4478,7 @@ fn ask_fixture_with_record(
.with_conversations(conversations)
.with_events(Arc::new(bus.clone()))
.with_structured(Arc::clone(&structured))
.with_restart_rendezvous(Arc::clone(&restart_rendezvous))
.with_record_turn(provider, Arc::new(FixedClock(clock_ms)) as Arc<dyn Clock>),
);
@ -4489,6 +4495,7 @@ fn ask_fixture_with_record(
// Live-state likewise unwired here (RecordTurn-focused fixtures).
live: Arc::new(RecordingLiveState::default()),
structured,
restart_rendezvous,
}
}
@ -4868,6 +4875,106 @@ struct FailingStructuredSession {
message: String,
}
struct RestartingStructuredSession {
id: SessionId,
project_id: ProjectId,
agent_id: AgentId,
coordinator: Arc<application::RestartRendezvousCoordinator>,
structured: Arc<StructuredSessions>,
replacement: Arc<dyn AgentSession>,
delay: Duration,
failure: Option<String>,
began: Arc<std::sync::atomic::AtomicBool>,
}
#[async_trait]
impl AgentSession for RestartingStructuredSession {
fn id(&self) -> SessionId {
self.id
}
fn conversation_id(&self) -> Option<String> {
Some("stable-engine-conversation".to_owned())
}
async fn send(&self, _prompt: &str) -> Result<ReplyStream, AgentSessionError> {
let key = domain::RuntimeAgentKey::new(self.project_id, self.agent_id);
let restart_id = format!("restart-{}", self.id);
assert!(
self.coordinator.begin(key, restart_id.clone()),
"the logical rendezvous must be registered before provider send"
);
self.began.store(true, Ordering::SeqCst);
let coordinator = Arc::clone(&self.coordinator);
let structured = Arc::clone(&self.structured);
let replacement = Arc::clone(&self.replacement);
let replacement_project_id = self.project_id;
let old_id = self.id;
let delay = self.delay;
let failure = self.failure.clone();
structured.remove(&old_id);
tokio::spawn(async move {
tokio::time::sleep(delay).await;
if let Some(message) = failure {
coordinator.fail(key, &restart_id, message);
return;
}
if coordinator.is_cancelled(key, &restart_id) {
return;
}
structured.insert_in_project(
replacement_project_id,
replacement,
key.agent_id,
nid(799),
);
coordinator.ready(key, &restart_id);
});
Err(AgentSessionError::Io(
"old provider session closed during restart".to_owned(),
))
}
async fn shutdown(&self) -> Result<(), AgentSessionError> {
Ok(())
}
async fn cancel_current_turn(&self) -> Result<(), AgentSessionError> {
Ok(())
}
}
struct CapturingFinalSession {
id: SessionId,
prompts: Arc<Mutex<Vec<String>>>,
final_text: String,
}
#[async_trait]
impl AgentSession for CapturingFinalSession {
fn id(&self) -> SessionId {
self.id
}
fn conversation_id(&self) -> Option<String> {
Some("stable-engine-conversation".to_owned())
}
async fn send(&self, prompt: &str) -> Result<ReplyStream, AgentSessionError> {
self.prompts.lock().unwrap().push(prompt.to_owned());
Ok(Box::new(
vec![ReplyEvent::Final {
content: self.final_text.clone(),
}]
.into_iter(),
))
}
async fn shutdown(&self) -> Result<(), AgentSessionError> {
Ok(())
}
}
#[async_trait]
impl AgentSession for FailingStructuredSession {
fn id(&self) -> SessionId {
@ -5329,6 +5436,286 @@ async fn ask_warm_structured_target_reuses_session_without_relaunch() {
);
}
fn install_restarting_session(
fx: &AskFixture,
project_id: ProjectId,
agent_id: AgentId,
delay: Duration,
failure: Option<&str>,
) -> (Arc<Mutex<Vec<String>>>, Arc<std::sync::atomic::AtomicBool>) {
if let Some(existing) = fx
.structured
.session_for_agent_in_project(project_id, &agent_id)
{
fx.structured.remove(&existing.id());
}
let prompts = Arc::new(Mutex::new(Vec::new()));
let began = Arc::new(std::sync::atomic::AtomicBool::new(false));
let replacement = Arc::new(CapturingFinalSession {
id: sid(9951),
prompts: Arc::clone(&prompts),
final_text: "final after restart".to_owned(),
}) as Arc<dyn AgentSession>;
fx.structured.insert_in_project(
project_id,
Arc::new(RestartingStructuredSession {
id: sid(9950),
project_id,
agent_id,
coordinator: Arc::clone(&fx.restart_rendezvous),
structured: Arc::clone(&fx.structured),
replacement,
delay,
failure: failure.map(str::to_owned),
began: Arc::clone(&began),
}),
agent_id,
nid(798),
);
(prompts, began)
}
#[tokio::test]
async fn ask_target_restart_keeps_logical_turn_and_returns_post_restart_final() {
let target = scratch_agent(aid(1), "architect", "agents/architect.md");
let fx = ask_fixture(FakeContexts::with_agent(&target, "# persona"));
let (prompts, _) = install_restarting_session(
&fx,
project_id(),
target.id,
Duration::from_millis(20),
None,
);
let out = fx
.service
.dispatch(&project(), cmd(ASK_JSON))
.await
.expect("restart-aware ask must return the replacement session Final");
assert_eq!(out.reply.as_deref(), Some("final after restart"));
assert_eq!(
prompts.lock().unwrap().as_slice(),
[application::SELF_RESTART_CONTINUATION_PROMPT],
"the replacement runtime receives one continuation turn, not the original task twice"
);
assert_eq!(
fx.mailbox.pending(&target.id),
0,
"logical ticket is retired"
);
}
#[tokio::test]
async fn ask_target_restart_cleans_busy_workstate_and_replaces_structured_registry_entry() {
let target = scratch_agent(aid(1), "architect", "agents/architect.md");
let fx = ask_fixture(FakeContexts::with_agent(&target, "# persona"));
install_restarting_session(
&fx,
project_id(),
target.id,
Duration::from_millis(20),
None,
);
let out = fx
.service
.dispatch(&project(), cmd(ASK_JSON))
.await
.expect("restart-aware ask must complete");
assert_eq!(out.reply.as_deref(), Some("final after restart"));
assert_eq!(
fx.mediator.busy_state(rkey(1)),
AgentBusyState::Idle,
"the logical target is not left busy after the post-restart Final"
);
assert_eq!(fx.mailbox.pending(&target.id), 0, "FIFO is empty");
let entry = fx
.live
.entry_for(target.id)
.expect("target live-state entry is written");
assert_eq!(entry.status, WorkStatus::Done);
assert_eq!(entry.last_delegation, entry.ticket);
assert!(
fx.structured.remove(&sid(9950)).is_none(),
"old provider session was removed during restart"
);
let current = fx
.structured
.session_for_agent_in_project(project_id(), &target.id)
.expect("replacement session remains registered for the same logical agent");
assert_eq!(current.id(), sid(9951));
}
#[tokio::test]
async fn cross_project_ask_preserves_correlation_across_target_restart() {
let target = scratch_agent(aid(1), "architect", "agents/architect.md").with_contactable(true);
let requester = scratch_agent(aid(99), "requester", "agents/requester.md");
let contexts = FakeContexts::with_agent(&target, "# target");
contexts.add_agent(&requester, "# requester");
let fx = ask_fixture(contexts);
let target_project = other_project();
let (prompts, _) = install_restarting_session(
&fx,
target_project.id,
target.id,
Duration::from_millis(20),
None,
);
let out = fx
.service
.ask_project_agent_for_mcp(
target_project.id,
target.id,
"cross-project work".to_owned(),
project_id(),
requester.id,
"corr-restart".to_owned(),
)
.await
.expect("cross-project rendezvous survives target restart");
assert_eq!(out.reply.as_deref(), Some("final after restart"));
assert_eq!(
prompts.lock().unwrap().as_slice(),
[application::SELF_RESTART_CONTINUATION_PROMPT]
);
let cleaned = fx
.service
.cancel_project_agent_for_mcp(
target_project.id,
target.id,
project_id(),
requester.id,
"corr-restart".to_owned(),
)
.await
.expect_err("completed cross-project correlation must be removed");
assert_eq!(cleaned.code(), "NOT_FOUND");
}
#[tokio::test]
async fn ask_target_restart_failure_is_explicit_not_no_reply() {
let target = scratch_agent(aid(1), "architect", "agents/architect.md");
let fx = ask_fixture(FakeContexts::with_agent(&target, "# persona"));
install_restarting_session(
&fx,
project_id(),
target.id,
Duration::from_millis(10),
Some("replacement launch exploded"),
);
let err = fx
.service
.dispatch(&project(), cmd(ASK_JSON))
.await
.expect_err("restart failure must terminate the awaiting ask explicitly");
assert_eq!(err.code(), "PROCESS");
assert!(err.to_string().contains("replacement launch exploded"));
assert!(!err.to_string().contains("structured Final response"));
assert_eq!(fx.mailbox.pending(&target.id), 0);
}
#[tokio::test]
async fn cross_project_cancel_during_restart_releases_ask_and_correlation() {
let target = scratch_agent(aid(1), "architect", "agents/architect.md").with_contactable(true);
let requester = scratch_agent(aid(99), "requester", "agents/requester.md");
let contexts = FakeContexts::with_agent(&target, "# target");
contexts.add_agent(&requester, "# requester");
let fx = ask_fixture(contexts);
let target_project = other_project();
let (_, began) = install_restarting_session(
&fx,
target_project.id,
target.id,
Duration::from_secs(5),
None,
);
let service = Arc::clone(&fx.service);
let ask = tokio::spawn(async move {
service
.ask_project_agent_for_mcp(
target_project.id,
target.id,
"cross-project work".to_owned(),
project_id(),
requester.id,
"corr-cancel-restart".to_owned(),
)
.await
});
await_until(|| began.load(Ordering::SeqCst)).await;
fx.service
.cancel_project_agent_for_mcp(
other_project().id,
aid(1),
project_id(),
aid(99),
"corr-cancel-restart".to_owned(),
)
.await
.expect("cancel routes through the stable cross-project correlation");
let err = ask
.await
.expect("ask task joins")
.expect_err("cancelled restart-aware ask must fail");
assert!(err.to_string().to_ascii_lowercase().contains("cancel"));
assert_eq!(fx.mailbox.pending(&aid(1)), 0);
}
#[tokio::test]
async fn same_project_stop_during_restart_releases_ask_and_leaves_no_replacement_session() {
let target = scratch_agent(aid(1), "architect", "agents/architect.md");
let fx = ask_fixture(FakeContexts::with_agent(&target, "# persona"));
let (_, began) =
install_restarting_session(&fx, project_id(), target.id, Duration::from_secs(5), None);
let service = Arc::clone(&fx.service);
let ask = tokio::spawn(async move { service.dispatch(&project(), cmd(ASK_JSON)).await });
await_until(|| began.load(Ordering::SeqCst)).await;
let stopped = fx
.service
.dispatch(
&project(),
cmd(r#"{ "action":"stop_agent", "name":"architect" }"#),
)
.await
.expect("same-project stop cancels the restart-aware logical turn");
assert_eq!(
stopped.detail,
"cancelled restarting work for agent architect"
);
let err = ask
.await
.expect("ask task joins")
.expect_err("cancelled restart-aware ask must fail");
assert!(err.to_string().to_ascii_lowercase().contains("cancel"));
assert_eq!(
fx.mediator.busy_state(rkey(1)),
AgentBusyState::Idle,
"cancel cleanup frees the target busy state"
);
assert_eq!(fx.mailbox.pending(&target.id), 0, "FIFO is empty");
await_until(|| {
fx.live
.entry_for(target.id)
.is_some_and(|entry| entry.status == WorkStatus::Done)
})
.await;
assert!(
fx.structured
.session_for_agent_in_project(project_id(), &target.id)
.is_none(),
"cancel during restart prevents a replacement runtime from being registered"
);
}
/// Régression #287 : la sonde transcript peut être stable (notamment inexistante pour
/// un provider non-Claude), tandis que `send_with_tap` publie réellement l'activité du
/// tour. Chaque événement non terminal doit réarmer la fenêtre jusqu'au `Final` tardif.

84
crates/backend/src/lib.rs
View File

@ -1037,9 +1037,6 @@ struct AppAgentResumer {
contexts: ResumeContexts,
}
const SELF_RESTART_PROMPT: &str =
"IdeA vient de te redémarrer. Reprends immédiatement le travail en cours là où tu l'avais laissé.";
struct AppAgentSelfRestarter {
launch_agent: Arc<LaunchAgent>,
close_terminal: Arc<CloseTerminal>,
@ -1049,6 +1046,7 @@ struct AppAgentSelfRestarter {
events: Arc<dyn EventBus>,
contexts: ResumeContexts,
active: Arc<Mutex<HashSet<domain::RuntimeAgentKey>>>,
rendezvous: Arc<application::RestartRendezvousCoordinator>,
}
fn self_restart_launch_input(
@ -1134,6 +1132,7 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
}
let restart_id = Uuid::new_v4().to_string();
let delegated = self.rendezvous.begin(key, restart_id.clone());
self.events.publish(DomainEvent::AgentRestartRequested {
agent_id: agent,
restart_id: restart_id.clone(),
@ -1148,6 +1147,7 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
let events = Arc::clone(&self.events);
let contexts = Arc::clone(&self.contexts);
let active = Arc::clone(&self.active);
let rendezvous = Arc::clone(&self.rendezvous);
let restart_id_task = restart_id.clone();
let project = project.clone();
spawn_detached(async move {
@ -1176,7 +1176,9 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
});
publish_phase(AgentRestartPhase::Stopping);
input.preempt(key);
if !delegated {
input.preempt(key);
}
let stop_result = application::terminal::stop_live_agent_runtime_in_project(
&project,
agent,
@ -1186,12 +1188,17 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
.await;
publish_phase(AgentRestartPhase::ReleasingInput);
if let Some(ticket) = input.busy_state(key).ticket() {
mailbox.cancel_head(key, ticket);
if !delegated {
if let Some(ticket) = input.busy_state(key).ticket() {
mailbox.cancel_head(key, ticket);
}
input.mark_idle(key);
}
input.mark_idle(key);
let fail = |phase, error: AppError| {
if delegated {
rendezvous.fail(key, &restart_id_task, error.to_string());
}
events.publish(DomainEvent::AgentRestartFailed {
agent_id: agent,
restart_id: restart_id_task.clone(),
@ -1199,6 +1206,11 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
message: error.to_string(),
});
};
let finish_coordination = || {
if delegated {
rendezvous.finish(key, &restart_id_task);
}
};
match stop_result {
Ok(stopped) if !stopped.is_empty() => {}
Ok(_) => {
@ -1206,16 +1218,28 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
AgentRestartPhase::Stopping,
AppError::NotFound(format!("running runtime for agent {agent}")),
);
finish_coordination();
active.lock().ok().map(|mut set| set.remove(&key));
return;
}
Err(error) => {
fail(AgentRestartPhase::Stopping, error);
finish_coordination();
active.lock().ok().map(|mut set| set.remove(&key));
return;
}
}
if delegated && rendezvous.is_cancelled(key, &restart_id_task) {
fail(
AgentRestartPhase::ReleasingInput,
AppError::Process("restart cancelled with its active rendezvous".to_owned()),
);
finish_coordination();
active.lock().ok().map(|mut set| set.remove(&key));
return;
}
publish_phase(AgentRestartPhase::Launching);
let mcp_runtime = crate::mcp_endpoint::idea_exe_path().map(|exe| McpRuntime {
exe,
@ -1232,11 +1256,48 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
Ok(output) => output,
Err(error) => {
fail(AgentRestartPhase::Launching, error);
finish_coordination();
active.lock().ok().map(|mut set| set.remove(&key));
return;
}
};
if delegated {
if rendezvous.is_cancelled(key, &restart_id_task) {
let _ = application::terminal::stop_live_agent_runtime_in_project(
&project,
agent,
live.as_ref(),
close_terminal.as_ref(),
)
.await;
fail(
AgentRestartPhase::Launching,
AppError::Process(
"restart cancelled before the continuation turn".to_owned(),
),
);
finish_coordination();
active.lock().ok().map(|mut set| set.remove(&key));
return;
}
publish_phase(AgentRestartPhase::Resuming);
rendezvous.ready(key, &restart_id_task);
if let Err(message) = rendezvous.wait_for_handoff(key, &restart_id_task).await {
fail(AgentRestartPhase::Resuming, AppError::Process(message));
finish_coordination();
active.lock().ok().map(|mut set| set.remove(&key));
return;
}
events.publish(DomainEvent::AgentRestartCompleted {
agent_id: agent,
restart_id: restart_id_task.clone(),
});
finish_coordination();
active.lock().ok().map(|mut set| set.remove(&key));
return;
}
publish_phase(AgentRestartPhase::Resuming);
let resume_result = if let Some(session) = live
.structured
@ -1247,10 +1308,10 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
Arc::clone(&mailbox),
project.id,
agent,
SELF_RESTART_PROMPT,
application::SELF_RESTART_CONTINUATION_PROMPT,
);
let result = session
.send(SELF_RESTART_PROMPT)
.send(application::SELF_RESTART_CONTINUATION_PROMPT)
.await
.map(|stream| stream.for_each(drop))
.map_err(|error| AppError::Process(error.to_string()));
@ -1272,7 +1333,7 @@ impl AgentSelfRestartScheduler for AppAgentSelfRestarter {
let ticket = domain::mailbox::Ticket::new(
domain::mailbox::TicketId::new_random(),
"IdeA",
SELF_RESTART_PROMPT,
application::SELF_RESTART_CONTINUATION_PROMPT,
);
let _pending = input.enqueue(key, ticket);
Ok(())
@ -3496,6 +3557,7 @@ impl BackendCore {
Arc::clone(&events_port),
resumer,
));
let restart_rendezvous = Arc::new(application::RestartRendezvousCoordinator::new());
let self_restart = Arc::new(AppAgentSelfRestarter {
launch_agent: Arc::clone(&launch_agent),
close_terminal: Arc::clone(&close_terminal),
@ -3505,6 +3567,7 @@ impl BackendCore {
events: Arc::clone(&events_port),
contexts: Arc::clone(&resume_contexts),
active: Arc::new(Mutex::new(HashSet::new())),
rendezvous: Arc::clone(&restart_rendezvous),
}) as Arc<dyn AgentSelfRestartScheduler>;
// Drain du scheduler (§21.5-b) : à chaque réveil tiré, `TokioScheduler` pousse une
// `ScheduledTask::ResumeAgent` ; on l'exécute via le service (relance + AgentResumed).
@ -3576,6 +3639,7 @@ impl BackendCore {
.with_conversations(Arc::clone(&conversation_registry))
.with_events(Arc::clone(&events_port))
.with_self_restart(self_restart)
.with_restart_rendezvous(restart_rendezvous)
// Faits OS/runtime (exe $APPIMAGE/current_exe + endpoint loopback) pour
// que les (re)lancements issus du chemin `ask` écrivent la déclaration MCP
// réelle ⇒ le pont MCP de la cible se spawne et `idea_reply` débloque le round-trip.