zsp/everything-claude-code
1
0
Fork 0
mirror of https://github.com/affaan-m/everything-claude-code.git synced 2026-04-10 11:23:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: track ecc2 chronic saturation streak

Browse Source
This commit is contained in:
Affaan Mustafa
2026-04-08 12:36:32 -07:00
parent 9d766af025
commit 10e34aa47a
3 changed files with 394 additions and 175 deletions
Download Patch File Download Diff File Expand all files Collapse all files

176
ecc2/src/session/daemon.rs
View File

@@ -97,15 +97,19 @@ fn check_sessions(db: &StateStore, timeout: Duration) -> Result<()> {
}
async fn maybe_auto_dispatch(db: &StateStore, cfg: &Config) -> Result<usize> {
let summary = maybe_auto_dispatch_with_recorder(cfg, || {
manager::auto_dispatch_backlog(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
}, |routed, deferred, leads| db.record_daemon_dispatch_pass(routed, deferred, leads))
let summary = maybe_auto_dispatch_with_recorder(
cfg,
|| {
manager::auto_dispatch_backlog(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
},
|routed, deferred, leads| db.record_daemon_dispatch_pass(routed, deferred, leads),
)
.await?;
Ok(summary.routed)
}
@@ -116,26 +120,34 @@ async fn coordinate_backlog_cycle(db: &StateStore, cfg: &Config) -> Result<()> {
cfg,
&activity,
|| {
maybe_auto_dispatch_with_recorder(cfg, || {
manager::auto_dispatch_backlog(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
}, |routed, deferred, leads| db.record_daemon_dispatch_pass(routed, deferred, leads))
maybe_auto_dispatch_with_recorder(
cfg,
|| {
manager::auto_dispatch_backlog(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
},
|routed, deferred, leads| db.record_daemon_dispatch_pass(routed, deferred, leads),
)
},
|| {
maybe_auto_rebalance_with_recorder(cfg, || {
manager::rebalance_all_teams(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
}, |rerouted, leads| db.record_daemon_rebalance_pass(rerouted, leads))
maybe_auto_rebalance_with_recorder(
cfg,
|| {
manager::rebalance_all_teams(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
},
|rerouted, leads| db.record_daemon_rebalance_pass(rerouted, leads),
)
},
|routed, leads| db.record_daemon_recovery_dispatch_pass(routed, leads),
)
@@ -163,7 +175,11 @@ where
tracing::warn!(
"Skipping immediate dispatch retry because chronic saturation cooloff is active"
);
return Ok((DispatchPassSummary::default(), rebalanced, DispatchPassSummary::default()));
return Ok((
DispatchPassSummary::default(),
rebalanced,
DispatchPassSummary::default(),
));
}
let first_dispatch = dispatch().await?;
if first_dispatch.routed > 0 {
@@ -206,7 +222,11 @@ where
F: Fn() -> Fut,
Fut: Future<Output = Result<Vec<manager::LeadDispatchOutcome>>>,
{
Ok(maybe_auto_dispatch_with_recorder(cfg, dispatch, |_, _, _| Ok(())).await?.routed)
Ok(
maybe_auto_dispatch_with_recorder(cfg, dispatch, |_, _, _| Ok(()))
.await?
.routed,
)
}
async fn maybe_auto_dispatch_with_recorder<F, Fut, R>(
@@ -254,9 +274,7 @@ where
);
}
if deferred > 0 {
tracing::warn!(
"Deferred {deferred} task handoff(s) because delegate teams were saturated"
);
tracing::warn!("Deferred {deferred} task handoff(s) because delegate teams were saturated");
}
Ok(DispatchPassSummary {
@@ -267,15 +285,19 @@ where
}
async fn maybe_auto_rebalance(db: &StateStore, cfg: &Config) -> Result<usize> {
maybe_auto_rebalance_with_recorder(cfg, || {
manager::rebalance_all_teams(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
}, |rerouted, leads| db.record_daemon_rebalance_pass(rerouted, leads))
maybe_auto_rebalance_with_recorder(
cfg,
|| {
manager::rebalance_all_teams(
db,
cfg,
&cfg.default_agent,
true,
cfg.max_parallel_sessions,
)
},
|rerouted, leads| db.record_daemon_rebalance_pass(rerouted, leads),
)
.await
}
@@ -528,7 +550,8 @@ mod tests {
}
#[tokio::test]
async fn coordinate_backlog_cycle_retries_after_rebalance_when_dispatch_deferred() -> Result<()> {
async fn coordinate_backlog_cycle_retries_after_rebalance_when_dispatch_deferred() -> Result<()>
{
let cfg = Config {
auto_dispatch_unread_handoffs: true,
..Config::default()
@@ -607,7 +630,8 @@ mod tests {
}
#[tokio::test]
async fn coordinate_backlog_cycle_records_recovery_dispatch_when_it_routes_work() -> Result<()> {
async fn coordinate_backlog_cycle_records_recovery_dispatch_when_it_routes_work() -> Result<()>
{
let cfg = Config {
auto_dispatch_unread_handoffs: true,
..Config::default()
@@ -653,7 +677,8 @@ mod tests {
}
#[tokio::test]
async fn coordinate_backlog_cycle_rebalances_first_after_unrecovered_deferred_pressure() -> Result<()> {
async fn coordinate_backlog_cycle_rebalances_first_after_unrecovered_deferred_pressure(
) -> Result<()> {
let cfg = Config {
auto_dispatch_unread_handoffs: true,
..Config::default()
@@ -664,6 +689,7 @@ mod tests {
last_dispatch_routed: 0,
last_dispatch_deferred: 2,
last_dispatch_leads: 1,
chronic_saturation_streak: 1,
last_recovery_dispatch_at: None,
last_recovery_dispatch_routed: 0,
last_recovery_dispatch_leads: 0,
@@ -708,7 +734,8 @@ mod tests {
}
#[tokio::test]
async fn coordinate_backlog_cycle_records_recovery_when_rebalance_first_dispatch_routes_work() -> Result<()> {
async fn coordinate_backlog_cycle_records_recovery_when_rebalance_first_dispatch_routes_work(
) -> Result<()> {
let cfg = Config {
auto_dispatch_unread_handoffs: true,
..Config::default()
@@ -719,6 +746,7 @@ mod tests {
last_dispatch_routed: 0,
last_dispatch_deferred: 2,
last_dispatch_leads: 1,
chronic_saturation_streak: 1,
last_recovery_dispatch_at: None,
last_recovery_dispatch_routed: 0,
last_recovery_dispatch_leads: 0,
@@ -755,7 +783,8 @@ mod tests {
}
#[tokio::test]
async fn coordinate_backlog_cycle_skips_dispatch_during_chronic_cooloff_when_rebalance_does_not_help() -> Result<()> {
async fn coordinate_backlog_cycle_skips_dispatch_during_chronic_cooloff_when_rebalance_does_not_help(
) -> Result<()> {
let cfg = Config {
auto_dispatch_unread_handoffs: true,
..Config::default()
@@ -766,6 +795,7 @@ mod tests {
last_dispatch_routed: 0,
last_dispatch_deferred: 3,
last_dispatch_leads: 1,
chronic_saturation_streak: 1,
last_recovery_dispatch_at: None,
last_recovery_dispatch_routed: 0,
last_recovery_dispatch_leads: 0,
@@ -803,7 +833,58 @@ mod tests {
}
#[tokio::test]
async fn coordinate_backlog_cycle_skips_rebalance_when_stabilized_and_dispatch_is_healthy() -> Result<()> {
async fn coordinate_backlog_cycle_skips_dispatch_when_persistent_saturation_streak_hits_cooloff(
) -> Result<()> {
let cfg = Config {
auto_dispatch_unread_handoffs: true,
..Config::default()
};
let now = chrono::Utc::now();
let activity = DaemonActivity {
last_dispatch_at: Some(now),
last_dispatch_routed: 0,
last_dispatch_deferred: 1,
last_dispatch_leads: 1,
chronic_saturation_streak: 3,
last_recovery_dispatch_at: None,
last_recovery_dispatch_routed: 0,
last_recovery_dispatch_leads: 0,
last_rebalance_at: Some(now - chrono::Duration::seconds(1)),
last_rebalance_rerouted: 0,
last_rebalance_leads: 1,
};
let calls = std::sync::Arc::new(std::sync::atomic::AtomicUsize::new(0));
let calls_clone = calls.clone();
let (first, rebalanced, recovery) = coordinate_backlog_cycle_with(
&cfg,
&activity,
move || {
let calls_clone = calls_clone.clone();
async move {
calls_clone.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
Ok(DispatchPassSummary {
routed: 1,
deferred: 0,
leads: 1,
})
}
},
|| async move { Ok(0) },
|_, _| Ok(()),
)
.await?;
assert_eq!(first, DispatchPassSummary::default());
assert_eq!(rebalanced, 0);
assert_eq!(recovery, DispatchPassSummary::default());
assert_eq!(calls.load(std::sync::atomic::Ordering::SeqCst), 0);
Ok(())
}
#[tokio::test]
async fn coordinate_backlog_cycle_skips_rebalance_when_stabilized_and_dispatch_is_healthy(
) -> Result<()> {
let cfg = Config {
auto_dispatch_unread_handoffs: true,
..Config::default()
@@ -814,6 +895,7 @@ mod tests {
last_dispatch_routed: 2,
last_dispatch_deferred: 0,
last_dispatch_leads: 1,
chronic_saturation_streak: 0,
last_recovery_dispatch_at: Some(now + chrono::Duration::seconds(1)),
last_recovery_dispatch_routed: 1,
last_recovery_dispatch_leads: 1,

122
ecc2/src/session/store.rs
View File

@@ -19,6 +19,7 @@ pub struct DaemonActivity {
pub last_dispatch_routed: usize,
pub last_dispatch_deferred: usize,
pub last_dispatch_leads: usize,
pub chronic_saturation_streak: usize,
pub last_recovery_dispatch_at: Option<chrono::DateTime<chrono::Utc>>,
pub last_recovery_dispatch_routed: usize,
pub last_recovery_dispatch_leads: usize,
@@ -44,12 +45,11 @@ impl DaemonActivity {
}
pub fn dispatch_cooloff_active(&self) -> bool {
self.prefers_rebalance_first() && self.last_dispatch_deferred >= 2
self.prefers_rebalance_first()
&& (self.last_dispatch_deferred >= 2 || self.chronic_saturation_streak >= 3)
}
pub fn chronic_saturation_cleared_at(
&self,
) -> Option<&chrono::DateTime<chrono::Utc>> {
pub fn chronic_saturation_cleared_at(&self) -> Option<&chrono::DateTime<chrono::Utc>> {
if self.prefers_rebalance_first() {
return None;
}
@@ -58,14 +58,14 @@ impl DaemonActivity {
self.last_dispatch_at.as_ref(),
self.last_recovery_dispatch_at.as_ref(),
) {
(Some(dispatch_at), Some(recovery_at)) if recovery_at > dispatch_at => Some(recovery_at),
(Some(dispatch_at), Some(recovery_at)) if recovery_at > dispatch_at => {
Some(recovery_at)
}
_ => None,
}
}
pub fn stabilized_after_recovery_at(
&self,
) -> Option<&chrono::DateTime<chrono::Utc>> {
pub fn stabilized_after_recovery_at(&self) -> Option<&chrono::DateTime<chrono::Utc>> {
if self.last_dispatch_deferred != 0 {
return None;
}
@@ -74,7 +74,9 @@ impl DaemonActivity {
self.last_dispatch_at.as_ref(),
self.last_recovery_dispatch_at.as_ref(),
) {
(Some(dispatch_at), Some(recovery_at)) if dispatch_at > recovery_at => Some(dispatch_at),
(Some(dispatch_at), Some(recovery_at)) if dispatch_at > recovery_at => {
Some(dispatch_at)
}
_ => None,
}
}
@@ -147,6 +149,7 @@ impl StateStore {
last_dispatch_routed INTEGER NOT NULL DEFAULT 0,
last_dispatch_deferred INTEGER NOT NULL DEFAULT 0,
last_dispatch_leads INTEGER NOT NULL DEFAULT 0,
chronic_saturation_streak INTEGER NOT NULL DEFAULT 0,
last_recovery_dispatch_at TEXT,
last_recovery_dispatch_routed INTEGER NOT NULL DEFAULT 0,
last_recovery_dispatch_leads INTEGER NOT NULL DEFAULT 0,
@@ -199,7 +202,9 @@ impl StateStore {
"ALTER TABLE daemon_activity ADD COLUMN last_recovery_dispatch_at TEXT",
[],
)
.context("Failed to add last_recovery_dispatch_at column to daemon_activity table")?;
.context(
"Failed to add last_recovery_dispatch_at column to daemon_activity table",
)?;
}
if !self.has_column("daemon_activity", "last_recovery_dispatch_routed")? {
@@ -220,6 +225,15 @@ impl StateStore {
.context("Failed to add last_recovery_dispatch_leads column to daemon_activity table")?;
}
if !self.has_column("daemon_activity", "chronic_saturation_streak")? {
self.conn
.execute(
"ALTER TABLE daemon_activity ADD COLUMN chronic_saturation_streak INTEGER NOT NULL DEFAULT 0",
[],
)
.context("Failed to add chronic_saturation_streak column to daemon_activity table")?;
}
Ok(())
}
@@ -550,9 +564,7 @@ impl StateStore {
})
})?;
messages
.collect::<Result<Vec<_>, _>>()
.map_err(Into::into)
messages.collect::<Result<Vec<_>, _>>().map_err(Into::into)
}
pub fn unread_task_handoff_count(&self, session_id: &str) -> Result<usize> {
@@ -582,9 +594,7 @@ impl StateStore {
Ok((row.get::<_, String>(0)?, row.get::<_, i64>(1)? as usize))
})?;
targets
.collect::<Result<Vec<_>, _>>()
.map_err(Into::into)
targets.collect::<Result<Vec<_>, _>>().map_err(Into::into)
}
pub fn mark_messages_read(&self, session_id: &str) -> Result<usize> {
@@ -624,6 +634,7 @@ impl StateStore {
self.conn
.query_row(
"SELECT last_dispatch_at, last_dispatch_routed, last_dispatch_deferred, last_dispatch_leads,
chronic_saturation_streak,
last_recovery_dispatch_at, last_recovery_dispatch_routed, last_recovery_dispatch_leads,
last_rebalance_at, last_rebalance_rerouted, last_rebalance_leads
FROM daemon_activity
@@ -652,12 +663,13 @@ impl StateStore {
last_dispatch_routed: row.get::<_, i64>(1)? as usize,
last_dispatch_deferred: row.get::<_, i64>(2)? as usize,
last_dispatch_leads: row.get::<_, i64>(3)? as usize,
last_recovery_dispatch_at: parse_ts(row.get(4)?)?,
last_recovery_dispatch_routed: row.get::<_, i64>(5)? as usize,
last_recovery_dispatch_leads: row.get::<_, i64>(6)? as usize,
last_rebalance_at: parse_ts(row.get(7)?)?,
last_rebalance_rerouted: row.get::<_, i64>(8)? as usize,
last_rebalance_leads: row.get::<_, i64>(9)? as usize,
chronic_saturation_streak: row.get::<_, i64>(4)? as usize,
last_recovery_dispatch_at: parse_ts(row.get(5)?)?,
last_recovery_dispatch_routed: row.get::<_, i64>(6)? as usize,
last_recovery_dispatch_leads: row.get::<_, i64>(7)? as usize,
last_rebalance_at: parse_ts(row.get(8)?)?,
last_rebalance_rerouted: row.get::<_, i64>(9)? as usize,
last_rebalance_leads: row.get::<_, i64>(10)? as usize,
})
},
)
@@ -675,7 +687,11 @@ impl StateStore {
SET last_dispatch_at = ?1,
last_dispatch_routed = ?2,
last_dispatch_deferred = ?3,
last_dispatch_leads = ?4
last_dispatch_leads = ?4,
chronic_saturation_streak = CASE
WHEN ?3 > 0 THEN chronic_saturation_streak + 1
ELSE 0
END
WHERE id = 1",
rusqlite::params![
chrono::Utc::now().to_rfc3339(),
@@ -693,7 +709,8 @@ impl StateStore {
"UPDATE daemon_activity
SET last_recovery_dispatch_at = ?1,
last_recovery_dispatch_routed = ?2,
last_recovery_dispatch_leads = ?3
last_recovery_dispatch_leads = ?3,
chronic_saturation_streak = 0
WHERE id = 1",
rusqlite::params![chrono::Utc::now().to_rfc3339(), routed as i64, leads as i64],
)?;
@@ -708,7 +725,11 @@ impl StateStore {
last_rebalance_rerouted = ?2,
last_rebalance_leads = ?3
WHERE id = 1",
rusqlite::params![chrono::Utc::now().to_rfc3339(), rerouted as i64, leads as i64],
rusqlite::params![
chrono::Utc::now().to_rfc3339(),
rerouted as i64,
leads as i64
],
)?;
Ok(())
@@ -1023,7 +1044,12 @@ mod tests {
db.insert_session(&build_session("planner", SessionState::Running))?;
db.insert_session(&build_session("worker", SessionState::Pending))?;
db.send_message("planner", "worker", "{\"question\":\"Need context\"}", "query")?;
db.send_message(
"planner",
"worker",
"{\"question\":\"Need context\"}",
"query",
)?;
db.send_message(
"worker",
"planner",
@@ -1066,17 +1092,11 @@ mod tests {
);
assert_eq!(
db.delegated_children("planner", 10)?,
vec![
"worker-3".to_string(),
"worker-2".to_string(),
]
vec!["worker-3".to_string(), "worker-2".to_string(),]
);
assert_eq!(
db.unread_task_handoff_targets(10)?,
vec![
("worker-2".to_string(), 1),
("worker-3".to_string(), 1),
]
vec![("worker-2".to_string(), 1), ("worker-3".to_string(), 1),]
);
Ok(())
@@ -1095,6 +1115,7 @@ mod tests {
assert_eq!(activity.last_dispatch_routed, 4);
assert_eq!(activity.last_dispatch_deferred, 1);
assert_eq!(activity.last_dispatch_leads, 2);
assert_eq!(activity.chronic_saturation_streak, 0);
assert_eq!(activity.last_recovery_dispatch_routed, 2);
assert_eq!(activity.last_recovery_dispatch_leads, 1);
assert_eq!(activity.last_rebalance_rerouted, 3);
@@ -1121,6 +1142,7 @@ mod tests {
last_dispatch_routed: 0,
last_dispatch_deferred: 2,
last_dispatch_leads: 1,
chronic_saturation_streak: 1,
last_recovery_dispatch_at: None,
last_recovery_dispatch_routed: 0,
last_recovery_dispatch_leads: 0,
@@ -1133,9 +1155,18 @@ mod tests {
assert!(unresolved.chronic_saturation_cleared_at().is_none());
assert!(unresolved.stabilized_after_recovery_at().is_none());
let persistent = DaemonActivity {
last_dispatch_deferred: 1,
chronic_saturation_streak: 3,
..unresolved.clone()
};
assert!(persistent.prefers_rebalance_first());
assert!(persistent.dispatch_cooloff_active());
let recovered = DaemonActivity {
last_recovery_dispatch_at: Some(now + chrono::Duration::seconds(1)),
last_recovery_dispatch_routed: 1,
chronic_saturation_streak: 0,
..unresolved
};
assert!(!recovered.prefers_rebalance_first());
@@ -1161,4 +1192,27 @@ mod tests {
stabilized.last_dispatch_at.as_ref()
);
}
#[test]
fn daemon_activity_tracks_chronic_saturation_streak() -> Result<()> {
let tempdir = TestDir::new("store-daemon-streak")?;
let db = StateStore::open(&tempdir.path().join("state.db"))?;
db.record_daemon_dispatch_pass(0, 1, 1)?;
db.record_daemon_dispatch_pass(0, 1, 1)?;
let saturated = db.daemon_activity()?;
assert_eq!(saturated.chronic_saturation_streak, 2);
assert!(!saturated.dispatch_cooloff_active());
db.record_daemon_dispatch_pass(0, 1, 1)?;
let chronic = db.daemon_activity()?;
assert_eq!(chronic.chronic_saturation_streak, 3);
assert!(chronic.dispatch_cooloff_active());
db.record_daemon_recovery_dispatch_pass(1, 1)?;
let recovered = db.daemon_activity()?;
assert_eq!(recovered.chronic_saturation_streak, 0);
Ok(())
}
}