feat(#1106): transient wake for bash tasks — bypass broker sqlite

This commit is contained in:
damocles 2026-06-03 10:40:33 +02:00 committed by mara
commit a1c6736ba5
8 changed files with 102 additions and 19 deletions

View file

@ -465,6 +465,11 @@ async fn send_wake(socket: &Path, id: &str, summary: &str, output: Option<(&str,
let req = hive_sh4re::AgentRequest::Wake {
from: format!("bash-task-{id}"),
body,
// Transient: do not persist to the message broker. Bash task
// completions are fire-and-forget — the output is already on disk
// in harness/bash-tasks/; persisting would cause duplicate delivery
// after a harness restart (issue #1103).
transient: true,
};
match crate::client::request::<_, hive_sh4re::AgentResponse>(socket, &req).await {
Ok(_) => tracing::info!(id = %id, "bash_runner: wake delivered"),

View file

@ -316,6 +316,7 @@ impl Surface for AgentSurface {
&AgentRequest::Wake {
from: "self".into(),
body: "continue".into(),
transient: false,
},
)
.await;
@ -365,7 +366,7 @@ impl Surface for AgentSurface {
async fn wake_external(socket: &Path, from: String, body: String) -> Result<()> {
let resp: AgentResponse =
client::request(socket, &AgentRequest::Wake { from, body }).await?;
client::request(socket, &AgentRequest::Wake { from, body, transient: false }).await?;
match resp {
AgentResponse::Ok => Ok(()),
AgentResponse::Err { message } => anyhow::bail!("wake: {message}"),
@ -456,6 +457,7 @@ impl Surface for ManagerSurface {
&ManagerRequest::Wake {
from: "self".into(),
body: "continue".into(),
transient: false,
},
)
.await;
@ -505,7 +507,7 @@ impl Surface for ManagerSurface {
async fn wake_external(socket: &Path, from: String, body: String) -> Result<()> {
let resp: ManagerResponse =
client::request(socket, &ManagerRequest::Wake { from, body }).await?;
client::request(socket, &ManagerRequest::Wake { from, body, transient: false }).await?;
match resp {
ManagerResponse::Ok => Ok(()),
ManagerResponse::Err { message } => anyhow::bail!("wake: {message}"),

View file

@ -684,6 +684,7 @@ async fn poll_once(
let req = hive_sh4re::Request::Wake {
from: "forge".to_owned(),
body,
transient: false,
};
let delivered = crate::client::request::<_, hive_sh4re::Response>(socket, &req)
.await

View file

@ -174,17 +174,31 @@ pub(crate) async fn dispatch_shared(
message: format!("{e:#}"),
},
},
hive_sh4re::Request::Wake { from, body } => match broker.send(&Message {
from: from.clone(),
to: agent.to_owned(),
body: body.clone(),
in_reply_to: None,
}) {
Ok(()) => hive_sh4re::Response::Ok,
Err(e) => hive_sh4re::Response::Err {
message: format!("{e:#}"),
},
},
hive_sh4re::Request::Wake {
from,
body,
transient,
} => {
if *transient {
// Transient wakes bypass sqlite — they fire the broadcast
// channel only. No redelivery on restart; no message history
// entry. Used by bash task completions.
broker.ping(agent, from, body);
hive_sh4re::Response::Ok
} else {
match broker.send(&Message {
from: from.clone(),
to: agent.to_owned(),
body: body.clone(),
in_reply_to: None,
}) {
Ok(()) => hive_sh4re::Response::Ok,
Err(e) => hive_sh4re::Response::Err {
message: format!("{e:#}"),
},
}
}
}
hive_sh4re::Request::Recent { limit } => match broker.recent_for(agent, *limit) {
Ok(rows) => hive_sh4re::Response::Recent { rows },
Err(e) => hive_sh4re::Response::Err {

View file

@ -117,6 +117,11 @@ pub enum MessageEvent {
at: i64,
in_reply_to: Option<i64>,
},
/// Transient wake signal — NOT persisted to sqlite. Wakes
/// `recv_blocking_batch` for the target agent but is not stored,
/// not re-delivered on restart, and not shown in message history.
/// Used for bash task completion notifications.
Ping { to: String, from: String, body: String },
}
/// Per-recipient in-memory bookkeeping for the deliver-then-ack
@ -191,6 +196,19 @@ impl Broker {
Ok(())
}
/// Deliver a transient wake signal to `to` without writing to sqlite.
/// The signal wakes a long-polling `recv_blocking_batch` for the target
/// agent but is not persisted, not redelivered on restart, and not shown
/// in message history. Use for ephemeral notifications (bash task
/// completions) where persistence would cause duplicate delivery.
pub fn ping(&self, to: &str, from: &str, body: &str) {
let _ = self.events.send(MessageEvent::Ping {
to: to.to_owned(),
from: from.to_owned(),
body: body.to_owned(),
});
}
/// Latest `limit` messages addressed to `recipient`, newest-first.
/// Includes delivered + undelivered alike — used for the operator
/// inbox view on the dashboard. Caller decides what to show.
@ -402,6 +420,32 @@ impl Broker {
}
// Lost a race (concurrent recv elsewhere). Keep waiting.
}
// Transient ping — not sqlite-backed. Return it directly as
// a Delivery with id=0 (sentinel: never pushed to unacked_ids
// so ack_turn silently ignores it).
Ok(Ok(MessageEvent::Ping {
to,
from,
body,
})) if to == recipient => {
// Also drain any real sqlite messages that may have landed
// concurrently; prepend the ping so the agent sees both.
let mut batch = self.recv_batch(recipient, max.saturating_sub(1))?;
batch.insert(
0,
Delivery {
id: 0,
redelivered: false,
message: Message {
from,
to,
body,
in_reply_to: None,
},
},
);
return Ok(batch);
}
Ok(Ok(_)) => {}
}
}

View file

@ -899,7 +899,7 @@ async fn dashboard_history(State(state): State<AppState>) -> Response {
messages.reverse();
let events: Vec<crate::dashboard_events::DashboardEvent> = messages
.into_iter()
.map(|m| match m {
.filter_map(|m| match m {
crate::broker::MessageEvent::Sent {
id,
from,
@ -909,7 +909,7 @@ async fn dashboard_history(State(state): State<AppState>) -> Response {
in_reply_to,
} => {
let file_refs = scan_validated_paths(&body);
crate::dashboard_events::DashboardEvent::Sent {
Some(crate::dashboard_events::DashboardEvent::Sent {
seq: 0,
id,
from,
@ -918,7 +918,7 @@ async fn dashboard_history(State(state): State<AppState>) -> Response {
at,
in_reply_to,
file_refs,
}
})
}
crate::broker::MessageEvent::Delivered {
id,
@ -929,7 +929,7 @@ async fn dashboard_history(State(state): State<AppState>) -> Response {
in_reply_to,
} => {
let file_refs = scan_validated_paths(&body);
crate::dashboard_events::DashboardEvent::Delivered {
Some(crate::dashboard_events::DashboardEvent::Delivered {
seq: 0,
id,
from,
@ -938,8 +938,11 @@ async fn dashboard_history(State(state): State<AppState>) -> Response {
at,
in_reply_to,
file_refs,
}
})
}
// Ping events are never persisted to sqlite — this arm is
// unreachable in practice but required for exhaustiveness.
crate::broker::MessageEvent::Ping { .. } => None,
})
.collect();
axum::Json(serde_json::json!({ "seq": seq, "events": events })).into_response()

View file

@ -408,6 +408,9 @@ fn spawn_broker_to_dashboard_forwarder(coord: Arc<Coordinator>) {
file_refs,
});
}
// Transient pings are not persisted and not shown in the
// dashboard message history — ignore silently.
Ok(MessageEvent::Ping { .. }) => {}
Err(tokio::sync::broadcast::error::RecvError::Lagged(n)) => {
tracing::warn!(skipped = n, "broker-to-dashboard forwarder lagged");
}

View file

@ -344,7 +344,18 @@ pub enum Request {
/// implicit (this agent); `from` is caller-chosen. See
/// `docs/conventions.md::Wake injection` for the trust model and
/// typical callers.
Wake { from: String, body: String },
///
/// When `transient` is `true` the server delivers the wake signal
/// through an in-process channel only — no sqlite write, no
/// redelivery on restart. Use this for ephemeral notifications (e.g.
/// bash task completions) where persistence is unnecessary and would
/// cause duplicate delivery after a harness restart.
Wake {
from: String,
body: String,
#[serde(default)]
transient: bool,
},
/// Last `limit` messages addressed to this agent, newest-first.
/// Non-mutating — pulls from the broker without delivering. The
/// per-agent web UI uses this to render its own inbox section.