Skip to content

fix(app-server): stabilize cross-platform CI regressions #434

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
andrei-hasna merged 4 commits into from
Jul 28, 2026
Merged

fix(app-server): stabilize cross-platform CI regressions #434

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
ff373c7
fix(app-server): stabilize cross-platform CI regressions
andrei-hasna Jul 28, 2026
275b16c
fix(app-server): preserve Windows read-only workers
andrei-hasna Jul 28, 2026
0ca397f
fix(app-server): preserve unmaterialized archive threads
andrei-hasna Jul 28, 2026
c22e73e
fix(app-server): stabilize mailbox and worker cleanup
andrei-hasna Jul 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
174 changes: 91 additions & 83 deletions codex-rs/app-server/src/message_processor_schedule_tests.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,6 +92,7 @@ use codex_rollout::state_db::StateDbHandle;
use core_test_support::responses;
use pretty_assertions::assert_eq;
use std::collections::BTreeMap;
use std::collections::VecDeque;
use std::future::Future;
use std::path::Path;
use std::sync::Arc;
Expand All @@ -117,6 +118,7 @@ struct ScheduleHarness {
state_db: StateDbHandle,
processor: Arc<MessageProcessor>,
outgoing_rx: mpsc::Receiver<OutgoingEnvelope>,
pending_notifications: VecDeque<ServerNotification>,
session: Arc<ConnectionSessionState>,
next_request_id: i64,
}
Expand Down Expand Up @@ -153,6 +155,7 @@ impl ScheduleHarness {
state_db,
processor,
outgoing_rx,
pending_notifications: VecDeque::new(),
session: Arc::new(ConnectionSessionState::new(ConnectionOrigin::WebSocket)),
next_request_id: 1,
};
Expand Down Expand Up @@ -250,6 +253,7 @@ impl ScheduleHarness {
params: ThreadStartParams {
cwd: Some(self.workspace_cwd()),
ephemeral: Some(ephemeral),
auth_profile: Some(None),
..ThreadStartParams::default()
},
})
Expand Down Expand Up @@ -302,17 +306,9 @@ impl ScheduleHarness {
.await
.expect("timed out waiting for response")
.expect("outgoing channel closed");
let OutgoingEnvelope::ToConnection {
connection_id,
message,
..
} = envelope
else {
let Some(message) = message_for_test_connection(envelope) else {
continue;
};
if connection_id != TEST_CONNECTION_ID {
continue;
}
match message {
OutgoingMessage::Response(response)
if response.id == RequestId::Integer(request_id) =>
Expand All @@ -323,6 +319,9 @@ impl ScheduleHarness {
OutgoingMessage::Error(error) if error.id == RequestId::Integer(request_id) => {
panic!("request {request_id} failed: {:?}", error.error);
}
OutgoingMessage::AppServerNotification(notification) => {
self.pending_notifications.push_back(notification);
}
_ => {
continue;
}
Expand All @@ -337,17 +336,9 @@ impl ScheduleHarness {
.await
.expect("timed out waiting for error")
.expect("outgoing channel closed");
let OutgoingEnvelope::ToConnection {
connection_id,
message,
..
} = envelope
else {
let Some(message) = message_for_test_connection(envelope) else {
continue;
};
if connection_id != TEST_CONNECTION_ID {
continue;
}
match message {
OutgoingMessage::Response(response)
if response.id == RequestId::Integer(request_id) =>
Expand All @@ -360,6 +351,9 @@ impl ScheduleHarness {
OutgoingMessage::Error(error) if error.id == RequestId::Integer(request_id) => {
return error.error;
}
OutgoingMessage::AppServerNotification(notification) => {
self.pending_notifications.push_back(notification);
}
_ => {
continue;
}
Expand Down Expand Up @@ -423,6 +417,43 @@ impl ScheduleHarness {
response.schedule
}

async fn seed_schedule_failure(&self, schedule_id: &str) -> Result<()> {
let now = Utc::now();
let local_active_fresh_after = self
.processor
.thread_schedule_runtime
.local_active_fresh_after(now);
let claim = self
.state_db
.thread_schedules()
.claim_thread_schedule_now_with_params(codex_state::ThreadScheduleNowClaimParams {
schedule_id,
now,
lease_id: "lease-fail",
lease_duration: std::time::Duration::from_secs(300),
local_active_owner_id: Some(
self.processor
.thread_schedule_runtime
.local_active_owner_id(),
),
local_active_fresh_after: Some(local_active_fresh_after),
})
.await?
.expect("schedule should claim for seeded failure");
self.state_db
.thread_schedules()
.fail_thread_schedule_run(
schedule_id,
claim.run.run_id.as_str(),
claim.run.lease_id.as_str(),
now,
/*next_run_at*/ None,
"model unavailable".to_string(),
)
.await?;
Ok(())
}

async fn read_schedule_deleted(&mut self, thread_id: &str, schedule_id: &str) {
loop {
let notification = self.read_server_notification().await;
Expand Down Expand Up @@ -706,6 +737,9 @@ impl ScheduleHarness {
}

async fn read_server_notification(&mut self) -> ServerNotification {
if let Some(notification) = self.pending_notifications.pop_front() {
return notification;
}
loop {
let envelope = tokio::time::timeout(
std::time::Duration::from_secs(/*secs*/ 20),
Expand All @@ -714,18 +748,8 @@ impl ScheduleHarness {
.await
.expect("timed out waiting for server notification")
.expect("outgoing channel closed");
let message = match envelope {
OutgoingEnvelope::ToConnection {
connection_id,
message,
..
} => {
if connection_id != TEST_CONNECTION_ID {
continue;
}
message
}
OutgoingEnvelope::Broadcast { message } => message,
let Some(message) = message_for_test_connection(envelope) else {
continue;
};
if let OutgoingMessage::AppServerNotification(notification) = message {
return notification;
Expand All @@ -734,6 +758,17 @@ impl ScheduleHarness {
}
}

fn message_for_test_connection(envelope: OutgoingEnvelope) -> Option<OutgoingMessage> {
match envelope {
OutgoingEnvelope::ToConnection {
connection_id,
message,
..
} => (connection_id == TEST_CONNECTION_ID).then_some(message),
OutgoingEnvelope::Broadcast { message } => Some(message),
}
}

async fn create_mock_responses_server_unauthorized() -> MockServer {
let server = MockServer::start().await;
Mock::given(method("POST"))
Expand All @@ -759,7 +794,10 @@ where
.name("schedule-harness".to_string())
.stack_size(16 * 1024 * 1024)
.spawn(|| {
tokio::runtime::Builder::new_current_thread()
// Message processing spawns runtime work that must remain runnable
// while Windows filesystem and SQLite operations block the caller.
tokio::runtime::Builder::new_multi_thread()
.worker_threads(2)
.enable_all()
.build()
.expect("schedule harness runtime should build")
Expand Down Expand Up @@ -1191,28 +1229,8 @@ fn thread_schedule_resume_recomputes_recurring_without_next_run_at() -> Result<(
.await;
harness.read_schedule_updated(&thread_id).await;

let claim = harness
.state_db
.thread_schedules()
.claim_thread_schedule_now(
create_response.schedule.schedule_id.as_str(),
Utc::now(),
"lease-fail",
std::time::Duration::from_secs(300),
)
.await?
.expect("schedule should claim for seeded failure");
harness
.state_db
.thread_schedules()
.fail_thread_schedule_run(
create_response.schedule.schedule_id.as_str(),
claim.run.run_id.as_str(),
"lease-fail",
Utc::now(),
/*next_run_at*/ None,
"model unavailable".to_string(),
)
.seed_schedule_failure(create_response.schedule.schedule_id.as_str())
.await?;
let failed_schedule = harness
.state_db
Expand Down Expand Up @@ -1281,28 +1299,8 @@ fn thread_schedule_update_to_active_resets_failure_count() -> Result<()> {
.await;
harness.read_schedule_updated(&thread_id).await;

let claim = harness
.state_db
.thread_schedules()
.claim_thread_schedule_now(
create_response.schedule.schedule_id.as_str(),
Utc::now(),
"lease-fail",
std::time::Duration::from_secs(300),
)
.await?
.expect("schedule should claim for seeded failure");
harness
.state_db
.thread_schedules()
.fail_thread_schedule_run(
create_response.schedule.schedule_id.as_str(),
claim.run.run_id.as_str(),
"lease-fail",
Utc::now(),
/*next_run_at*/ None,
"model unavailable".to_string(),
)
.seed_schedule_failure(create_response.schedule.schedule_id.as_str())
.await?;
let failed_schedule = harness
.state_db
Expand Down Expand Up @@ -1788,45 +1786,50 @@ fn thread_schedule_create_nests_loops_to_depth_five() -> Result<()> {
let thread_id = thread.thread.id.clone();

let root = harness
.create_interval_thread_schedule(&thread_id, "root loop", 1, None)
.create_interval_thread_schedule(
&thread_id,
"root loop",
/*amount_minutes*/ 1,
/*parent_schedule_id*/ None,
)
.await;
let level_2 = harness
.create_interval_thread_schedule(
&thread_id,
"level 2 loop",
2,
/*amount_minutes*/ 2,
Some(root.schedule_id.clone()),
)
.await;
let branch = harness
.create_interval_thread_schedule(
&thread_id,
"branch level 2 loop",
3,
/*amount_minutes*/ 3,
Some(root.schedule_id.clone()),
)
.await;
let level_3 = harness
.create_interval_thread_schedule(
&thread_id,
"level 3 loop",
3,
/*amount_minutes*/ 3,
Some(level_2.schedule_id.clone()),
)
.await;
let level_4 = harness
.create_interval_thread_schedule(
&thread_id,
"level 4 loop",
4,
/*amount_minutes*/ 4,
Some(level_3.schedule_id.clone()),
)
.await;
let level_5 = harness
.create_interval_thread_schedule(
&thread_id,
"level 5 loop",
5,
/*amount_minutes*/ 5,
Some(level_4.schedule_id.clone()),
)
.await;
Expand Down Expand Up @@ -1889,21 +1892,26 @@ fn thread_schedule_delete_parent_emits_descendant_delete_notifications() -> Resu
let thread = harness.start_materialized_thread().await;
let thread_id = thread.thread.id.clone();
let root = harness
.create_interval_thread_schedule(&thread_id, "root loop", 1, None)
.create_interval_thread_schedule(
&thread_id,
"root loop",
/*amount_minutes*/ 1,
/*parent_schedule_id*/ None,
)
.await;
let child = harness
.create_interval_thread_schedule(
&thread_id,
"child loop",
2,
/*amount_minutes*/ 2,
Some(root.schedule_id.clone()),
)
.await;
let grandchild = harness
.create_interval_thread_schedule(
&thread_id,
"grandchild loop",
3,
/*amount_minutes*/ 3,
Some(child.schedule_id.clone()),
)
.await;
Expand Down Expand Up @@ -2300,7 +2308,7 @@ fn schedule_create_materializes_fresh_thread_rollout_before_first_user_turn() ->
let rollout_path = codex_rollout::find_thread_path_by_id_str(
harness._codex_home.path(),
&thread_id,
Option::<&codex_state::StateRuntime>::None,
/*state_db_ctx*/ Option::<&codex_state::StateRuntime>::None,
)
.await?
.expect("fresh scheduled thread should have a materialized rollout");
Expand Down
7 changes: 5 additions & 2 deletions codex-rs/app-server/src/request_processors/active_session_processor.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -571,7 +571,10 @@ mod tests {
assert_eq!(named_api_peer.auth_profile.as_deref(), Some("work"));
assert_eq!(named_api_peer.auth_profile_kind, AuthProfileKind::Named);

let default_api_peer = api_active_session_peer(test_active_peer(ThreadId::new(), None));
let default_api_peer = api_active_session_peer(test_active_peer(
ThreadId::new(),
/*auth_profile*/ None,
));

assert_eq!(default_api_peer.auth_profile, None);
assert_eq!(default_api_peer.auth_profile_kind, AuthProfileKind::Default);
Expand Down Expand Up @@ -682,7 +685,7 @@ mod tests {
auth_profile,
process: None,
capabilities: ActivePeerCapabilities::codewith_session(),
last_seen_at: LastSeenAt::from_unix_seconds(100),
last_seen_at: LastSeenAt::from_unix_seconds(/*seconds*/ 100),
}
}
}
Loading
Loading