Avoid paginated resume requests for verified legacy rollouts (#38604)

## Why

Resuming a known legacy rollout with `excludeTurns` first requires the TUI to
retry after the app server rejects paginated history loading.

## What changed

- Carry the history mode reported by the session picker into the resume flow.
- For embedded app-server sessions, resume a legacy rollout with its turns when
  the rollout is still legacy and background migration cannot race the check.
- Keep requesting paginated history when the picker metadata may be stale,
  migration is enabled or active, or the app server is remote.

## Testing

Added coverage for picker server replacement, background migration, maintenance
lock contention, stale history metadata, and paginated resume requests.

GitOrigin-RevId: 2300d1ad93a817eeeeb8f8ed60b073400ddf7106
This commit is contained in:
Charlie Marsh authored and copyberry committed 2026-08-14 16:33:28 +00:00
1 parent 588ee89324
commit 1bb6384c19
13 files changed
+455 -90

No files matched your search

@@ -0,0 +1,150 @@
//! Resume history policy for local rollouts that may migrate between formats.
use super::AppServerSession;
use super::AppServerStartedThread;
use super::HistoryHydrationScope;
use super::ResumeModelSettings;
use super::ThreadHistorySupport;
use super::bootstrap_request_error;
use super::is_history_pagination_unsupported;
use super::started_thread_from_resume_response;
use super::thread_resume_params_from_config;
use crate::legacy_core::config::Config;
use codex_app_server_client::TypedRequestError;
use codex_app_server_protocol::ClientRequest;
use codex_app_server_protocol::ThreadHistoryMode;
use codex_app_server_protocol::ThreadResumeResponse;
use codex_features::Feature;
use codex_protocol::ThreadId;
use color_eyre::eyre::Result;
impl AppServerSession {
/// Captures the server's startup migration policy before workspace config can change.
///
/// Sessions without a recorded startup config conservatively assume migration is enabled.
pub(crate) fn with_startup_config(mut self, config: &Config) -> Self {
self.background_rollout_migration_enabled = config
.features
.enabled(Feature::BackgroundPaginatedRolloutMigration);
self
}
/// Retains trusted embedded-server history metadata for a later resume.
///
/// Remote hints are ignored because another process can migrate the thread before resume.
pub(crate) fn remember_thread_history_mode(
&mut self,
thread_id: ThreadId,
history_mode: ThreadHistoryMode,
) {
if !self.uses_embedded_app_server() {
return;
}
self.history_pagination
.entry(thread_id)
.or_default()
.history_mode = history_mode;
}
pub(crate) async fn resume_thread(
&mut self,
config: Config,
thread_id: ThreadId,
model_settings: ResumeModelSettings,
) -> Result<AppServerStartedThread> {
let session_config = if model_settings == ResumeModelSettings::RestoreFromThread {
config.clone()
} else {
self.session_config_with_effective_service_tier(&config)
};
let mut params = thread_resume_params_from_config(
session_config,
thread_id,
self.thread_params_mode(),
self.remote_cwd_override.as_deref(),
model_settings,
);
let mut rollout_maintenance_guard = None;
params.exclude_turns = if self.history_support == ThreadHistorySupport::Paginated {
let known_legacy_history = self
.history_pagination
.get(&thread_id)
.is_some_and(|state| state.history_mode == ThreadHistoryMode::Legacy)
&& (!self.uses_embedded_app_server()
|| (!self.background_rollout_migration_enabled
&& !config
.features
.enabled(Feature::BackgroundPaginatedRolloutMigration)
&& {
rollout_maintenance_guard =
codex_rollout::try_acquire_rollout_maintenance_lock(
config.codex_home.as_path(),
)
.ok()
.flatten();
rollout_maintenance_guard.is_some()
}
&& self
.thread_read(thread_id, /*include_turns*/ false)
.await
.is_ok_and(|thread| thread.history_mode == ThreadHistoryMode::Legacy)));
!known_legacy_history
} else {
false
};
if params.exclude_turns {
rollout_maintenance_guard = None;
}
let request_id = self.next_request_id();
let resume_response = self
.client
.request_typed(ClientRequest::ThreadResume {
request_id,
params: params.clone(),
})
.await;
drop(rollout_maintenance_guard);
let mut response: ThreadResumeResponse = match resume_response {
Ok(response) => response,
Err(TypedRequestError::Server { source, .. })
if params.exclude_turns && is_history_pagination_unsupported(&source) =>
{
self.history_support = ThreadHistorySupport::LegacyOnly;
params.exclude_turns = false;
let request_id = self.next_request_id();
self.client
.request_typed(ClientRequest::ThreadResume { request_id, params })
.await
.map_err(|err| {
bootstrap_request_error("thread/resume failed during TUI bootstrap", err)
})?
}
Err(err) => {
return Err(bootstrap_request_error(
"thread/resume failed during TUI bootstrap",
err,
));
}
};
self.hydrate_initial_thread_history(
&mut response.thread,
response.turns_backwards_cursor.clone(),
response.items_backwards_cursor.clone(),
Some(&config),
HistoryHydrationScope::Initial,
)
.await?;
let fork_parent_title = self
.fork_parent_title_from_app_server(response.thread.forked_from_id.as_deref())
.await;
let mut started =
started_thread_from_resume_response(response, &config, self.thread_params_mode())
.await?;
started.session.fork_parent_title = fork_parent_title;
Ok(started)
}
}
#[cfg(test)]
#[path = "rollout_history_tests.rs"]
mod tests;
@@ -0,0 +1,225 @@
use super::super::ResumeModelSettings;
use crate::legacy_core::config::Config;
use crate::legacy_core::config::ConfigBuilder;
use app_test_support::create_fake_paginated_rollout;
use app_test_support::create_fake_rollout;
use codex_app_server_protocol::ThreadHistoryMode;
use codex_features::Feature;
use codex_protocol::ThreadId;
use color_eyre::eyre::Result;
use pretty_assertions::assert_eq;
use tempfile::TempDir;
async fn build_config(temp_dir: &TempDir) -> Config {
ConfigBuilder::default()
.codex_home(temp_dir.path().to_path_buf())
.build()
.await
.expect("config should build")
}
#[tokio::test]
async fn legacy_resume_preserves_history_mode_after_picker_server_replacement() -> Result<()> {
let codex_home = tempfile::tempdir().expect("tempdir");
let config = build_config(&codex_home).await;
let thread_id = ThreadId::from_string(
&create_fake_rollout(
codex_home.path(),
"2025-01-05T12-00-00",
"2025-01-05T12:00:00Z",
"Saved user message",
Some(config.model_provider_id.as_str()),
/*git_info*/ None,
)
.expect("create source rollout"),
)?;
let mut picker_app_server = crate::start_embedded_app_server_for_picker(&config).await?;
let history_mode = picker_app_server
.thread_read(thread_id, /*include_turns*/ false)
.await?
.history_mode;
picker_app_server.shutdown().await?;
let mut app_server = crate::start_embedded_app_server_for_picker(&config).await?;
app_server.remember_thread_history_mode(thread_id, history_mode);
let next_request_id = app_server.next_request_id;
let resumed = app_server
.resume_thread(config, thread_id, ResumeModelSettings::RestoreFromThread)
.await?;
assert_eq!(app_server.next_request_id, next_request_id + 2);
assert!(!resumed.turns.is_empty());
app_server.shutdown().await?;
Ok(())
}
#[tokio::test]
async fn background_migration_disables_cached_legacy_resume_shortcut() -> Result<()> {
let codex_home = tempfile::tempdir().expect("tempdir");
let config = build_config(&codex_home).await;
let legacy_thread_id = ThreadId::from_string(
&create_fake_rollout(
codex_home.path(),
"2025-01-05T12-00-00",
"2025-01-05T12:00:00Z",
"Saved legacy user message",
Some(config.model_provider_id.as_str()),
/*git_info*/ None,
)
.expect("create legacy rollout"),
)?;
let thread_id = ThreadId::from_string(
&create_fake_paginated_rollout(
codex_home.path(),
"2025-01-05T12-00-01",
"2025-01-05T12:00:01Z",
"Saved paginated user message",
Some(config.model_provider_id.as_str()),
/*git_info*/ None,
)
.expect("create paginated rollout"),
)?;
let mut migration_config = config.clone();
migration_config
.features
.enable(Feature::BackgroundPaginatedRolloutMigration)?;
for (startup_config, resume_config) in
[(&config, &migration_config), (&migration_config, &config)]
{
// Keep the real worker disabled while exercising both startup/request mismatches.
let mut app_server = crate::start_embedded_app_server_for_picker(&config)
.await?
.with_startup_config(startup_config);
app_server.remember_thread_history_mode(legacy_thread_id, ThreadHistoryMode::Legacy);
let next_request_id = app_server.next_request_id;
let legacy = app_server
.resume_thread(
resume_config.clone(),
legacy_thread_id,
ResumeModelSettings::RestoreFromThread,
)
.await?;
assert_eq!(app_server.next_request_id, next_request_id + 2);
assert!(!legacy.turns.is_empty());
app_server.remember_thread_history_mode(thread_id, ThreadHistoryMode::Legacy);
let next_request_id = app_server.next_request_id;
let resumed = app_server
.resume_thread(
resume_config.clone(),
thread_id,
ResumeModelSettings::RestoreFromThread,
)
.await?;
assert!(app_server.next_request_id > next_request_id + 1);
assert_eq!(resumed.session.thread_id, thread_id);
assert_eq!(
app_server
.history_pagination
.get(&thread_id)
.map(|state| state.history_mode),
Some(ThreadHistoryMode::Paginated)
);
app_server.shutdown().await?;
}
Ok(())
}
#[tokio::test]
async fn rollout_maintenance_contention_disables_cached_legacy_resume_shortcut() -> Result<()> {
let codex_home = tempfile::tempdir().expect("tempdir");
let config = build_config(&codex_home).await;
let thread_id = ThreadId::from_string(
&create_fake_paginated_rollout(
codex_home.path(),
"2025-01-05T12-00-00",
"2025-01-05T12:00:00Z",
"Saved paginated user message",
Some(config.model_provider_id.as_str()),
/*git_info*/ None,
)
.expect("create paginated rollout"),
)?;
let mut app_server = crate::start_embedded_app_server_for_picker(&config).await?;
app_server.remember_thread_history_mode(thread_id, ThreadHistoryMode::Legacy);
let _maintenance_guard =
codex_rollout::try_acquire_rollout_maintenance_lock(codex_home.path())?
.expect("acquire rollout maintenance lock");
let next_request_id = app_server.next_request_id;
let resumed = app_server
.resume_thread(config, thread_id, ResumeModelSettings::RestoreFromThread)
.await?;
assert_eq!(app_server.next_request_id, next_request_id + 3);
assert_eq!(resumed.session.thread_id, thread_id);
assert_eq!(
app_server
.history_pagination
.get(&thread_id)
.map(|state| state.history_mode),
Some(ThreadHistoryMode::Paginated)
);
app_server.shutdown().await?;
Ok(())
}
#[tokio::test]
async fn stale_legacy_history_mode_is_revalidated_before_resume() -> Result<()> {
let codex_home = tempfile::tempdir().expect("tempdir");
let config = build_config(&codex_home).await;
let thread_id = ThreadId::from_string(
&create_fake_paginated_rollout(
codex_home.path(),
"2025-01-05T12-00-00",
"2025-01-05T12:00:00Z",
"Saved paginated user message",
Some(config.model_provider_id.as_str()),
/*git_info*/ None,
)
.expect("create paginated rollout"),
)?;
let mut app_server = crate::start_embedded_app_server_for_picker(&config).await?;
app_server.remember_thread_history_mode(thread_id, ThreadHistoryMode::Legacy);
let next_request_id = app_server.next_request_id;
let resumed = app_server
.resume_thread(
config.clone(),
thread_id,
ResumeModelSettings::RestoreFromThread,
)
.await?;
assert_eq!(resumed.session.thread_id, thread_id);
assert!(app_server.next_request_id >= next_request_id + 4);
assert_eq!(
app_server
.history_pagination
.get(&thread_id)
.map(|state| state.history_mode),
Some(ThreadHistoryMode::Paginated)
);
let missing_thread_id = ThreadId::new();
app_server.remember_thread_history_mode(missing_thread_id, ThreadHistoryMode::Legacy);
let next_request_id = app_server.next_request_id;
assert!(
app_server
.resume_thread(
config,
missing_thread_id,
ResumeModelSettings::RestoreFromThread
)
.await
.is_err()
);
assert_eq!(app_server.next_request_id, next_request_id + 2);
app_server.shutdown().await?;
Ok(())
}