Retry failed Codex Apps MCP startup (#29920)

## Problem

The built-in Codex Apps MCP client shares a future for the full startup
operation: connect, complete `initialize`, fetch the initial tools, and
return a usable client. Sharing deduplicates startup work, but it also
memoizes terminal errors.

After a transient connection, handshake, or initial `tools/list`
failure, later tool builds observe the same failed future. The thread
cannot reconnect after the backend recovers and continues serving its
startup-time cached tool snapshot, which may be empty or stale.

## Fix

When Apps MCP startup ends in an error, Codex starts bounded recovery
without putting startup latency on tool-router construction:

1. The current tool build immediately continues with the cached startup
snapshot.
2. After the initial failure is reported, Codex starts one fresh full
startup attempt in the background.
3. Concurrent tool builds share that in-flight attempt and also continue
with cached tools.
4. On success, the recovered client becomes active, refreshes the Apps
tools cache, emits a `Ready` startup status, and is reused by later
operations.
5. On failure, the cache remains unchanged and later tool builds may
start another background attempt after exponential cooldown: 1s, 2s, 4s,
8s, 16s, then 30s maximum.

Each recreated startup performs a fresh MCP `initialize` and uncached
`tools/list`. The MCP client retains its existing bounded retries for
retryable `initialize` and `tools/list` failures.

This avoids adding the Apps startup timeout to every request during a
sustained outage.

## Scope

This is limited to the built-in Codex Apps MCP client:

- no reconnects for user-configured MCP servers;
- no cache deletion; and
- no proactive refresh for a healthy client with stale tools.

## Tests

Coverage verifies:

- tool builds return cached tools without waiting for a blocked
reconnect;
- concurrent tool builds start only one background reconnect;
- failed reconnects preserve cached tools and respect exponential
cooldown;
- a recovered client is retained and reused; and
- a long-lived thread exposes recovered app tools on a later follow-up.

Validation:

- `just test -p codex-mcp` — 95 passed
- `just test -p codex-core
later_follow_up_uses_background_recovered_apps_after_mid_thread_startup_failures
--no-capture` — passed
- `just fix -p codex-mcp`
- `just fmt`
This commit is contained in:
kbazzi
2026-06-25 21:31:12 -07:00
committed by GitHub
Unverified
parent f5f812389e
commit 92d2e1df70
5 changed files with 767 additions and 51 deletions
@@ -208,6 +208,7 @@ impl McpConnectionManager {
};
let async_managed_client = AsyncManagedClient::new(
server_name.clone(),
startup_submit_id.clone(),
server,
store_mode,
keyring_backend_kind,
@@ -257,6 +258,10 @@ impl McpConnectionManager {
)
.await;
if matches!(&outcome, Err(StartupOutcomeError::Failed { .. })) {
async_managed_client.reconnect_failed_startup().await;
}
(server_name, outcome)
});
}
@@ -479,6 +484,7 @@ impl McpConnectionManager {
pub async fn list_all_tools(&self) -> Vec<ToolInfo> {
let mut tools = Vec::new();
for (server_name, managed_client) in &self.clients {
managed_client.reconnect_failed_startup().await;
let has_cached_tools = managed_client.has_cached_tools();
let startup_complete = managed_client
.startup_complete
@@ -6,7 +6,10 @@ use crate::elicitation::ElicitationRequestManager;
use crate::elicitation::ElicitationRequestRouter;
use crate::elicitation::elicitation_is_rejected_by_policy;
use crate::rmcp_client::AsyncManagedClient;
use crate::rmcp_client::CODEX_APPS_RECONNECT_INITIAL_BACKOFF;
use crate::rmcp_client::CodexAppsStartupReconnect;
use crate::rmcp_client::ManagedClient;
use crate::rmcp_client::ManagedClientFuture;
use crate::rmcp_client::StartupOutcomeError;
use crate::server::EffectiveMcpServer;
use crate::server::McpServerMetadata;
@@ -21,6 +24,7 @@ use codex_config::Constrained;
use codex_config::McpServerConfig;
use codex_config::McpServerToolConfig;
use codex_config::types::AuthKeyringBackendKind;
use codex_config::types::OAuthCredentialsStoreMode;
use codex_exec_server::EnvironmentManager;
use codex_protocol::ToolName;
use codex_protocol::mcp::McpServerInfo;
@@ -41,6 +45,7 @@ use rmcp::model::Tool;
use std::collections::HashSet;
use std::io;
use std::sync::Arc;
use std::sync::atomic::AtomicUsize;
use tempfile::tempdir;
use tokio::io::DuplexStream;
@@ -101,9 +106,8 @@ impl InProcessTransportFactory for TestInProcessTransportFactory {
}
}
async fn create_ready_async_managed_client(tools: Vec<ToolInfo>) -> AsyncManagedClient {
let tool_filter = ToolFilter::default();
let managed_client = ManagedClient {
async fn create_test_managed_client(tools: Vec<ToolInfo>) -> ManagedClient {
ManagedClient {
client: Arc::new(
RmcpClient::new_in_process_client(Arc::new(TestInProcessTransportFactory))
.await
@@ -111,29 +115,73 @@ async fn create_ready_async_managed_client(tools: Vec<ToolInfo>) -> AsyncManaged
),
server_info: create_test_server_info("Ready"),
tools,
tool_filter: tool_filter.clone(),
tool_filter: ToolFilter::default(),
tool_timeout: None,
server_instructions: None,
server_supports_sandbox_state_meta_capability: false,
codex_apps_tools_cache_context: None,
};
}
}
async fn create_ready_async_managed_client(tools: Vec<ToolInfo>) -> AsyncManagedClient {
AsyncManagedClient {
client: futures::future::ready::<Result<ManagedClient, StartupOutcomeError>>(Ok(
managed_client,
create_test_managed_client(tools).await,
))
.boxed()
.shared(),
is_codex_apps_mcp_server: false,
cached_server_info: None,
codex_apps_tools_cache_context: None,
tool_filter,
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(true)),
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
}
}
fn create_test_manager_with_failed_apps_startup(
cached_tools: Vec<ToolInfo>,
reconnect_factory: Arc<dyn Fn() -> ManagedClientFuture + Send + Sync>,
) -> McpConnectionManager {
let client: ManagedClientFuture = futures::future::ready(Err(StartupOutcomeError::Failed {
error: "startup failed".to_string(),
is_authentication_required: false,
}))
.boxed()
.shared();
let codex_home = tempdir().expect("tempdir");
let cache_context = create_codex_apps_tools_cache_context(
codex_home.path().to_path_buf(),
Some("reconnect-test-account"),
Some("reconnect-test-user"),
);
cache_context.store_current_tools_for_test(cached_tools);
let approval_policy = Constrained::allow_any(AskForApproval::OnRequest);
let permission_profile = Constrained::allow_any(PermissionProfile::default());
let mut manager = McpConnectionManager::new_uninitialized(
&approval_policy,
&permission_profile,
/*prefix_mcp_tool_names*/ true,
);
manager.clients.insert(
CODEX_APPS_MCP_SERVER_NAME.to_string(),
AsyncManagedClient {
client,
is_codex_apps_mcp_server: true,
cached_server_info: None,
codex_apps_tools_cache_context: Some(cache_context),
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(true)),
startup_reconnect: Some(Arc::new(CodexAppsStartupReconnect::new(reconnect_factory))),
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
},
);
manager
}
fn model_tool_names(tools: &[ToolInfo]) -> HashSet<ToolName> {
tools
.iter()
@@ -721,6 +769,7 @@ async fn list_all_tools_uses_shared_codex_apps_cache_while_client_is_pending() {
codex_apps_tools_cache_context: Some(cache_context),
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(false)),
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
},
@@ -760,6 +809,7 @@ async fn list_available_server_infos_uses_cache_while_client_is_pending() {
codex_apps_tools_cache_context: None,
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(false)),
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
},
@@ -860,6 +910,7 @@ async fn list_all_tools_blocks_while_client_is_pending_without_cached_tools() {
codex_apps_tools_cache_context: None,
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(false)),
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
},
@@ -898,6 +949,7 @@ async fn shutdown_cancels_pending_tool_listing() {
codex_apps_tools_cache_context: None,
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(false)),
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token,
},
@@ -944,6 +996,7 @@ async fn shutdown_continues_after_caller_is_aborted() {
codex_apps_tools_cache_context: None,
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(false)),
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
},
@@ -996,6 +1049,7 @@ async fn list_all_tools_does_not_block_when_shared_codex_apps_cache_is_empty() {
codex_apps_tools_cache_context: Some(cache_context),
tool_filter: ToolFilter::default(),
startup_complete: Arc::new(std::sync::atomic::AtomicBool::new(false)),
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
},
@@ -1045,6 +1099,7 @@ async fn list_all_tools_uses_shared_codex_apps_cache_when_client_startup_fails()
codex_apps_tools_cache_context: Some(cache_context),
tool_filter: ToolFilter::default(),
startup_complete,
startup_reconnect: None,
tool_plugin_provenance: Arc::new(ToolPluginProvenance::default()),
cancel_token: CancellationToken::new(),
},
@@ -1069,6 +1124,235 @@ async fn list_all_tools_uses_shared_codex_apps_cache_when_client_startup_fails()
);
}
#[tokio::test]
async fn list_all_tools_reconnects_failed_codex_apps_startup_and_reuses_client() {
let recovered_client = create_test_managed_client(vec![create_test_tool(
CODEX_APPS_MCP_SERVER_NAME,
"drive_search",
)])
.await;
let attempts = Arc::new(AtomicUsize::new(0));
let attempts_for_reconnect = Arc::clone(&attempts);
let reconnect_finished = Arc::new(tokio::sync::Notify::new());
let reconnect_finished_for_factory = Arc::clone(&reconnect_finished);
let reconnect_factory = Arc::new(move || {
attempts_for_reconnect.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
let reconnect_finished = Arc::clone(&reconnect_finished_for_factory);
let recovered_client = recovered_client.clone();
async move {
reconnect_finished.notify_one();
Ok(recovered_client)
}
.boxed()
.shared()
});
let manager = create_test_manager_with_failed_apps_startup(Vec::new(), reconnect_factory);
let reconnect_finished_wait = reconnect_finished.notified();
let tools = manager.list_all_tools().await;
assert!(tools.is_empty());
reconnect_finished_wait.await;
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["drive_search"]
);
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 1);
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["drive_search"]
);
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 1);
}
#[tokio::test(start_paused = true)]
async fn later_tool_list_retries_after_failed_reconnect_and_keeps_cached_tools() {
let recovered_client = create_test_managed_client(vec![create_test_tool(
CODEX_APPS_MCP_SERVER_NAME,
"drive_search",
)])
.await;
let attempts = Arc::new(AtomicUsize::new(0));
let attempts_for_reconnect = Arc::clone(&attempts);
let reconnect_finished = Arc::new(tokio::sync::Notify::new());
let reconnect_finished_for_factory = Arc::clone(&reconnect_finished);
let reconnect_factory = Arc::new(move || {
let attempt = attempts_for_reconnect.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
let reconnect_finished = Arc::clone(&reconnect_finished_for_factory);
let recovered_client = recovered_client.clone();
async move {
let result = if attempt < 2 {
Err(StartupOutcomeError::Failed {
error: "recreated startup failed".to_string(),
is_authentication_required: false,
})
} else {
Ok(recovered_client)
};
reconnect_finished.notify_one();
result
}
.boxed()
.shared()
});
let manager = create_test_manager_with_failed_apps_startup(
vec![create_test_tool(
CODEX_APPS_MCP_SERVER_NAME,
"cached_drive_search",
)],
reconnect_factory,
);
let first_reconnect_finished = reconnect_finished.notified();
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["cached_drive_search"]
);
first_reconnect_finished.await;
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 1);
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["cached_drive_search"]
);
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 1);
tokio::time::advance(CODEX_APPS_RECONNECT_INITIAL_BACKOFF).await;
let second_reconnect_finished = reconnect_finished.notified();
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["cached_drive_search"]
);
second_reconnect_finished.await;
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 2);
tokio::time::advance(CODEX_APPS_RECONNECT_INITIAL_BACKOFF).await;
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["cached_drive_search"]
);
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 2);
tokio::time::advance(CODEX_APPS_RECONNECT_INITIAL_BACKOFF).await;
let third_reconnect_finished = reconnect_finished.notified();
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["cached_drive_search"]
);
third_reconnect_finished.await;
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 3);
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["drive_search"]
);
}
#[tokio::test]
async fn tool_lists_do_not_block_and_share_codex_apps_startup_reconnect() {
let recovered_client = create_test_managed_client(vec![create_test_tool(
CODEX_APPS_MCP_SERVER_NAME,
"drive_search",
)])
.await;
let attempts = Arc::new(AtomicUsize::new(0));
let attempts_for_reconnect = Arc::clone(&attempts);
let reconnect_started = Arc::new(tokio::sync::Notify::new());
let reconnect_started_for_factory = Arc::clone(&reconnect_started);
let release_reconnect = Arc::new(tokio::sync::Notify::new());
let release_reconnect_for_factory = Arc::clone(&release_reconnect);
let reconnect_factory = Arc::new(move || {
let recovered_client = recovered_client.clone();
let attempts = Arc::clone(&attempts_for_reconnect);
let reconnect_started = Arc::clone(&reconnect_started_for_factory);
let release_reconnect = Arc::clone(&release_reconnect_for_factory);
async move {
attempts.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
reconnect_started.notify_one();
release_reconnect.notified().await;
Ok(recovered_client)
}
.boxed()
.shared()
});
let manager = Arc::new(create_test_manager_with_failed_apps_startup(
vec![create_test_tool(
CODEX_APPS_MCP_SERVER_NAME,
"cached_drive_search",
)],
reconnect_factory,
));
let reconnect_started_wait = reconnect_started.notified();
let first_tools = tokio::time::timeout(Duration::from_millis(10), manager.list_all_tools())
.await
.expect("cached tools should not wait for reconnect");
reconnect_started_wait.await;
let second_tools = tokio::time::timeout(Duration::from_millis(10), manager.list_all_tools())
.await
.expect("concurrent cached tools should not wait for reconnect");
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 1);
assert_eq!(
first_tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["cached_drive_search"]
);
assert_eq!(
second_tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["cached_drive_search"]
);
release_reconnect.notify_one();
tokio::task::yield_now().await;
let tools = manager.list_all_tools().await;
assert_eq!(
tools
.iter()
.map(|tool| tool.callable_name.as_str())
.collect::<Vec<_>>(),
vec!["drive_search"]
);
assert_eq!(attempts.load(std::sync::atomic::Ordering::SeqCst), 1);
}
#[tokio::test]
async fn list_all_tools_adds_server_metadata_to_tools() {
let server_name = "docs";
+271 -44
View File
@@ -12,6 +12,7 @@ use std::collections::HashMap;
use std::env;
use std::ffi::OsString;
use std::sync::Arc;
use std::sync::Mutex as StdMutex;
use std::sync::atomic::AtomicBool;
use std::sync::atomic::Ordering;
use std::time::Duration;
@@ -48,6 +49,9 @@ use codex_exec_server::HttpClient;
use codex_exec_server::ReqwestHttpClient;
use codex_protocol::mcp::McpServerInfo;
use codex_protocol::protocol::Event;
use codex_protocol::protocol::EventMsg;
use codex_protocol::protocol::McpStartupStatus;
use codex_protocol::protocol::McpStartupUpdateEvent;
use codex_rmcp_client::ExecutorStdioServerLauncher;
use codex_rmcp_client::LocalStdioServerLauncher;
use codex_rmcp_client::RmcpClient;
@@ -64,6 +68,7 @@ use rmcp::model::JsonObject;
use rmcp::model::ProtocolVersion;
use rmcp::model::Tool as RmcpTool;
use rmcp::transport::auth::AuthError;
use tokio::time::Instant as TokioInstant;
use tokio_util::sync::CancellationToken;
use tracing::Instrument;
use tracing::instrument;
@@ -80,6 +85,9 @@ pub(crate) const MCP_TOOLS_FETCH_UNCACHED_DURATION_METRIC: &str =
pub(crate) const DEFAULT_STARTUP_TIMEOUT: Duration = Duration::from_secs(30);
pub(crate) const DEFAULT_TOOL_TIMEOUT: Duration = Duration::from_secs(300);
pub(crate) const CODEX_APPS_RECONNECT_INITIAL_BACKOFF: Duration = Duration::from_secs(1);
const CODEX_APPS_RECONNECT_MAX_BACKOFF: Duration = Duration::from_secs(30);
const UNTRUSTED_CONNECTOR_META_KEYS: &[&str] = &[
"connector_id",
"connector_name",
@@ -128,56 +136,174 @@ impl ManagedClient {
}
}
#[derive(Clone)]
pub(crate) struct AsyncManagedClient {
pub(crate) client: Shared<BoxFuture<'static, Result<ManagedClient, StartupOutcomeError>>>,
pub(crate) is_codex_apps_mcp_server: bool,
pub(crate) cached_server_info: Option<McpServerInfo>,
pub(crate) codex_apps_tools_cache_context: Option<CodexAppsToolsCacheContext>,
pub(crate) tool_filter: ToolFilter,
pub(crate) startup_complete: Arc<AtomicBool>,
pub(crate) tool_plugin_provenance: Arc<ToolPluginProvenance>,
pub(crate) cancel_token: CancellationToken,
pub(crate) type ManagedClientFuture =
Shared<BoxFuture<'static, Result<ManagedClient, StartupOutcomeError>>>;
#[derive(Default)]
struct CodexAppsStartupReconnectState {
current_client: Option<ManagedClient>,
reconnect_in_flight: bool,
consecutive_failures: u32,
retry_not_before: Option<TokioInstant>,
}
impl AsyncManagedClient {
// Keep this constructor flat so the startup inputs remain readable at the
// single call site instead of introducing a one-off params wrapper.
#[instrument(level = "trace", skip_all, fields(server_name = %server_name))]
#[allow(clippy::too_many_arguments)]
pub(crate) fn new(
#[derive(Clone)]
struct CodexAppsStartupStatusContext {
submit_id: String,
server_name: String,
tx_event: Sender<Event>,
}
pub(crate) struct CodexAppsStartupReconnect {
factory: Arc<dyn Fn() -> ManagedClientFuture + Send + Sync>,
state: StdMutex<CodexAppsStartupReconnectState>,
startup_status_context: Option<CodexAppsStartupStatusContext>,
}
impl CodexAppsStartupReconnect {
pub(crate) fn new(factory: Arc<dyn Fn() -> ManagedClientFuture + Send + Sync>) -> Self {
Self {
factory,
state: StdMutex::new(CodexAppsStartupReconnectState::default()),
startup_status_context: None,
}
}
fn with_startup_status_context(
mut self,
submit_id: String,
server_name: String,
server: EffectiveMcpServer,
store_mode: OAuthCredentialsStoreMode,
keyring_backend_kind: AuthKeyringBackendKind,
cancel_token: CancellationToken,
tx_event: Sender<Event>,
elicitation_requests: ElicitationRequestManager,
codex_apps_tools_cache_context: Option<CodexAppsToolsCacheContext>,
tool_plugin_provenance: Arc<ToolPluginProvenance>,
runtime_context: McpRuntimeContext,
runtime_auth_provider: Option<SharedAuthProvider>,
client_elicitation_capability: ElicitationCapability,
supports_openai_form_elicitation: bool,
) -> Self {
self.startup_status_context = Some(CodexAppsStartupStatusContext {
submit_id,
server_name,
tx_event,
});
self
}
fn current_client(&self) -> Option<ManagedClient> {
self.state
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.current_client
.clone()
}
fn reconnect_in_background(self: &Arc<Self>) {
{
let mut state = self
.state
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
if state.current_client.is_some() || state.reconnect_in_flight {
return;
}
if state
.retry_not_before
.is_some_and(|retry_not_before| TokioInstant::now() < retry_not_before)
{
return;
}
state.reconnect_in_flight = true;
}
let reconnect = Arc::clone(self);
tokio::spawn(async move {
let result = (reconnect.factory)().await;
let startup_status_context = reconnect.startup_status_context.clone();
let recovered = {
let mut state = reconnect
.state
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
state.reconnect_in_flight = false;
match result {
Ok(client) => {
state.current_client = Some(client);
state.consecutive_failures = 0;
state.retry_not_before = None;
true
}
Err(error) => {
state.consecutive_failures = state.consecutive_failures.saturating_add(1);
let retry_after = codex_apps_reconnect_backoff(state.consecutive_failures);
state.retry_not_before = Some(TokioInstant::now() + retry_after);
warn!(
error = %error,
retry_after_ms = retry_after.as_millis(),
"Apps MCP startup reconnect failed; continuing with cached tools"
);
false
}
}
};
if recovered && let Some(context) = startup_status_context {
let _ = context
.tx_event
.send(Event {
id: context.submit_id,
msg: EventMsg::McpStartupUpdate(McpStartupUpdateEvent {
server: context.server_name,
status: McpStartupStatus::Ready,
}),
})
.await;
}
});
}
}
fn codex_apps_reconnect_backoff(consecutive_failures: u32) -> Duration {
let exponent = consecutive_failures.saturating_sub(1).min(5);
CODEX_APPS_RECONNECT_INITIAL_BACKOFF
.saturating_mul(1 << exponent)
.min(CODEX_APPS_RECONNECT_MAX_BACKOFF)
}
#[derive(Clone)]
struct ManagedClientStartup {
server_name: String,
server: EffectiveMcpServer,
store_mode: OAuthCredentialsStoreMode,
keyring_backend_kind: AuthKeyringBackendKind,
tx_event: Sender<Event>,
elicitation_requests: ElicitationRequestManager,
codex_apps_tools_cache_context: Option<CodexAppsToolsCacheContext>,
runtime_context: McpRuntimeContext,
runtime_auth_provider: Option<SharedAuthProvider>,
client_elicitation_capability: ElicitationCapability,
supports_openai_form_elicitation: bool,
cancel_token: CancellationToken,
startup_complete: Arc<AtomicBool>,
}
impl ManagedClientStartup {
fn start(&self) -> ManagedClientFuture {
let Self {
server_name,
server,
store_mode,
keyring_backend_kind,
tx_event,
elicitation_requests,
codex_apps_tools_cache_context,
runtime_context,
runtime_auth_provider,
client_elicitation_capability,
supports_openai_form_elicitation,
cancel_token,
startup_complete,
} = self.clone();
let is_codex_apps_mcp_server = server_name == CODEX_APPS_MCP_SERVER_NAME;
let tool_filter = server
.configured_config()
.map(ToolFilter::from_config)
.unwrap_or_default();
let cached_server_info = if is_codex_apps_mcp_server {
codex_apps_tools_cache_context
.as_ref()
.and_then(load_startup_cached_codex_apps_server_info)
} else {
None
};
let startup_tool_filter = tool_filter.clone();
let codex_apps_tools_cache_context_for_fut = codex_apps_tools_cache_context.clone();
let startup_complete = Arc::new(AtomicBool::new(false));
let startup_complete_for_fut = Arc::clone(&startup_complete);
let cancel_token_for_fut = cancel_token.clone();
let fut = async move {
let cancel_token_for_fut = cancel_token;
async move {
let outcome = match async {
if let Err(error) = validate_mcp_server_name(&server_name) {
return Err(error.into());
@@ -207,10 +333,10 @@ impl AsyncManagedClient {
.configured_config()
.and_then(|config| config.tool_timeout_sec)
.unwrap_or(DEFAULT_TOOL_TIMEOUT),
tool_filter: startup_tool_filter,
tool_filter,
tx_event,
elicitation_requests,
codex_apps_tools_cache_context: codex_apps_tools_cache_context_for_fut,
codex_apps_tools_cache_context,
client_elicitation_capability,
supports_openai_form_elicitation,
},
@@ -224,10 +350,91 @@ impl AsyncManagedClient {
Err(CancelErr::Cancelled) => Err(StartupOutcomeError::Cancelled),
};
startup_complete_for_fut.store(true, Ordering::Release);
startup_complete.store(true, Ordering::Release);
outcome
}
.in_current_span()
.boxed()
.shared()
}
}
#[derive(Clone)]
pub(crate) struct AsyncManagedClient {
pub(crate) client: ManagedClientFuture,
pub(crate) is_codex_apps_mcp_server: bool,
pub(crate) cached_server_info: Option<McpServerInfo>,
pub(crate) codex_apps_tools_cache_context: Option<CodexAppsToolsCacheContext>,
pub(crate) tool_filter: ToolFilter,
pub(crate) startup_complete: Arc<AtomicBool>,
pub(crate) startup_reconnect: Option<Arc<CodexAppsStartupReconnect>>,
pub(crate) tool_plugin_provenance: Arc<ToolPluginProvenance>,
pub(crate) cancel_token: CancellationToken,
}
impl AsyncManagedClient {
// Keep this constructor flat so the startup inputs remain readable at the
// single call site instead of introducing a one-off params wrapper.
#[instrument(level = "trace", skip_all, fields(server_name = %server_name))]
#[allow(clippy::too_many_arguments)]
pub(crate) fn new(
server_name: String,
startup_submit_id: String,
server: EffectiveMcpServer,
store_mode: OAuthCredentialsStoreMode,
keyring_backend_kind: AuthKeyringBackendKind,
cancel_token: CancellationToken,
tx_event: Sender<Event>,
elicitation_requests: ElicitationRequestManager,
codex_apps_tools_cache_context: Option<CodexAppsToolsCacheContext>,
tool_plugin_provenance: Arc<ToolPluginProvenance>,
runtime_context: McpRuntimeContext,
runtime_auth_provider: Option<SharedAuthProvider>,
client_elicitation_capability: ElicitationCapability,
supports_openai_form_elicitation: bool,
) -> Self {
let is_codex_apps_mcp_server = server_name == CODEX_APPS_MCP_SERVER_NAME;
let reconnect_server_name = server_name.clone();
let reconnect_tx_event = tx_event.clone();
let tool_filter = server
.configured_config()
.map(ToolFilter::from_config)
.unwrap_or_default();
let cached_server_info = if is_codex_apps_mcp_server {
codex_apps_tools_cache_context
.as_ref()
.and_then(load_startup_cached_codex_apps_server_info)
} else {
None
};
let client = fut.in_current_span().boxed().shared();
let startup_complete = Arc::new(AtomicBool::new(false));
let startup = Arc::new(ManagedClientStartup {
server_name,
server,
store_mode,
keyring_backend_kind,
tx_event,
elicitation_requests,
codex_apps_tools_cache_context: codex_apps_tools_cache_context.clone(),
runtime_context,
runtime_auth_provider,
client_elicitation_capability,
supports_openai_form_elicitation,
cancel_token: cancel_token.clone(),
startup_complete: Arc::clone(&startup_complete),
});
let client = startup.start();
let startup_reconnect = is_codex_apps_mcp_server.then(|| {
let startup = Arc::clone(&startup);
Arc::new(
CodexAppsStartupReconnect::new(Arc::new(move || startup.start()))
.with_startup_status_context(
startup_submit_id,
reconnect_server_name,
reconnect_tx_event,
),
)
});
if codex_apps_tools_cache_context
.as_ref()
.is_some_and(CodexAppsToolsCacheContext::has_current_tools)
@@ -245,15 +452,35 @@ impl AsyncManagedClient {
codex_apps_tools_cache_context,
tool_filter,
startup_complete,
startup_reconnect,
tool_plugin_provenance,
cancel_token,
}
}
pub(crate) async fn client(&self) -> Result<ManagedClient, StartupOutcomeError> {
if let Some(client) = self
.startup_reconnect
.as_ref()
.and_then(|reconnect| reconnect.current_client())
{
return Ok(client);
}
self.client.clone().await
}
pub(crate) async fn reconnect_failed_startup(&self) {
let Some(startup_reconnect) = self.startup_reconnect.as_ref() else {
return;
};
if !self.startup_complete.load(Ordering::Acquire) {
return;
}
if matches!(self.client().await, Err(StartupOutcomeError::Failed { .. })) {
startup_reconnect.reconnect_in_background();
}
}
pub(crate) async fn shutdown(&self) {
self.cancel_token.cancel();
match self.client().await {
@@ -7,6 +7,7 @@ use codex_login::CodexAuth;
use codex_models_manager::bundled_models_response;
use serde_json::Value;
use serde_json::json;
use std::sync::Arc;
use std::sync::atomic::AtomicUsize;
use std::sync::atomic::Ordering;
use wiremock::Mock;
@@ -55,6 +56,23 @@ pub struct AppsTestServer {
pub chatgpt_base_url: String,
}
#[derive(Clone)]
pub struct AppsTestServerStartupControl {
initialize_attempts: Arc<AtomicUsize>,
remaining_initialize_failures: Arc<AtomicUsize>,
}
impl AppsTestServerStartupControl {
pub fn fail_next_initialize_attempts(&self, attempts: usize) {
self.remaining_initialize_failures
.store(attempts, Ordering::SeqCst);
}
pub fn initialize_attempts(&self) -> usize {
self.initialize_attempts.load(Ordering::SeqCst)
}
}
#[derive(Clone, Copy)]
pub enum AppsTestToolLoading {
Direct,
@@ -130,6 +148,34 @@ impl AppsTestServer {
})
}
pub async fn mount_with_startup_control(
server: &MockServer,
) -> Result<(Self, AppsTestServerStartupControl)> {
mount_oauth_metadata(server).await;
mount_connectors_directory(server).await;
let control = AppsTestServerStartupControl {
initialize_attempts: Arc::new(AtomicUsize::new(0)),
remaining_initialize_failures: Arc::new(AtomicUsize::new(0)),
};
mount_streamable_http_json_rpc_with_startup_control(
server,
CONNECTOR_NAME.to_string(),
CONNECTOR_DESCRIPTION.to_string(),
/*searchable*/ true,
/*include_app_only_tool*/ false,
AppsTestToolsListBehavior::AlwaysAvailable,
Some(Arc::clone(&control.initialize_attempts)),
Some(Arc::clone(&control.remaining_initialize_failures)),
)
.await;
Ok((
Self {
chatgpt_base_url: server.uri(),
},
control,
))
}
pub async fn mount_with_tools_available_after_initial_list(
server: &MockServer,
) -> Result<Self> {
@@ -312,6 +358,30 @@ async fn mount_streamable_http_json_rpc(
searchable: bool,
include_app_only_tool: bool,
tools_list_behavior: AppsTestToolsListBehavior,
) {
mount_streamable_http_json_rpc_with_startup_control(
server,
connector_name,
connector_description,
searchable,
include_app_only_tool,
tools_list_behavior,
/*initialize_attempts*/ None,
/*remaining_initialize_failures*/ None,
)
.await;
}
#[allow(clippy::too_many_arguments)]
async fn mount_streamable_http_json_rpc_with_startup_control(
server: &MockServer,
connector_name: String,
connector_description: String,
searchable: bool,
include_app_only_tool: bool,
tools_list_behavior: AppsTestToolsListBehavior,
initialize_attempts: Option<Arc<AtomicUsize>>,
remaining_initialize_failures: Option<Arc<AtomicUsize>>,
) {
Mock::given(method("POST"))
.and(path_regex("^/api/codex/apps/?$"))
@@ -322,6 +392,8 @@ async fn mount_streamable_http_json_rpc(
include_app_only_tool,
tools_list_behavior,
tools_list_calls: AtomicUsize::new(0),
initialize_attempts,
remaining_initialize_failures,
})
.mount(server)
.await;
@@ -334,6 +406,8 @@ struct CodexAppsJsonRpcResponder {
include_app_only_tool: bool,
tools_list_behavior: AppsTestToolsListBehavior,
tools_list_calls: AtomicUsize,
initialize_attempts: Option<Arc<AtomicUsize>>,
remaining_initialize_failures: Option<Arc<AtomicUsize>>,
}
impl Respond for CodexAppsJsonRpcResponder {
@@ -355,6 +429,24 @@ impl Respond for CodexAppsJsonRpcResponder {
match method {
"initialize" => {
if let Some(initialize_attempts) = &self.initialize_attempts {
initialize_attempts.fetch_add(1, Ordering::SeqCst);
}
if self
.remaining_initialize_failures
.as_ref()
.is_some_and(|remaining| {
remaining
.fetch_update(Ordering::SeqCst, Ordering::SeqCst, |remaining| {
remaining.checked_sub(1)
})
.is_ok()
})
{
return ResponseTemplate::new(400).set_body_json(json!({
"error": "simulated non-retryable Apps MCP startup failure",
}));
}
let id = body.get("id").cloned().unwrap_or(Value::Null);
let protocol_version = body
.pointer("/params/protocolVersion")
@@ -1,5 +1,8 @@
use anyhow::Result;
use codex_features::Feature;
use codex_mcp::CODEX_APPS_MCP_SERVER_NAME;
use codex_protocol::protocol::McpServerRefreshConfig;
use codex_protocol::protocol::Op;
use core_test_support::apps_test_server::AppsTestServer;
use core_test_support::apps_test_server::SEARCH_CALENDAR_CREATE_TOOL;
use core_test_support::apps_test_server::SEARCH_CALENDAR_NAMESPACE;
@@ -8,10 +11,13 @@ use core_test_support::responses;
use core_test_support::responses::ev_assistant_message;
use core_test_support::responses::ev_completed;
use core_test_support::responses::ev_response_created;
use core_test_support::responses::mount_sse_sequence;
use core_test_support::responses::namespace_child_tool;
use core_test_support::responses::sse;
use core_test_support::skip_if_no_network;
use core_test_support::wait_for_mcp_server;
use serde_json::Value;
use std::time::Duration;
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn code_mode_only_exposes_direct_model_only_mcp_namespaces() -> Result<()> {
@@ -79,3 +85,104 @@ async fn code_mode_only_exposes_direct_model_only_mcp_namespaces() -> Result<()>
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn later_follow_up_uses_background_recovered_apps_after_mid_thread_startup_failures()
-> Result<()> {
skip_if_no_network!(Ok(()));
let server = responses::start_mock_server().await;
let (apps_server, startup_control) =
AppsTestServer::mount_with_startup_control(&server).await?;
let response = mount_sse_sequence(
&server,
vec![
sse(vec![
ev_response_created("resp-1"),
ev_assistant_message("msg-1", "initial turn"),
ev_completed("resp-1"),
]),
sse(vec![
ev_response_created("resp-2"),
ev_assistant_message("msg-2", "recovery-trigger turn"),
ev_completed("resp-2"),
]),
sse(vec![
ev_response_created("resp-3"),
ev_assistant_message("msg-3", "recovered follow-up turn"),
ev_completed("resp-3"),
]),
],
)
.await;
let mut builder = search_capable_apps_builder(apps_server.chatgpt_base_url.clone())
.with_config(move |config| {
config
.features
.enable(Feature::CodeModeOnly)
.expect("test config should allow feature update");
config.code_mode.direct_only_tool_namespaces =
vec![SEARCH_CALENDAR_NAMESPACE.to_string()];
});
let test = builder.build(&server).await?;
wait_for_mcp_server(&test.codex, CODEX_APPS_MCP_SERVER_NAME).await?;
test.submit_turn("use Calendar before refreshing MCP")
.await?;
let initial_request = response.requests()[0].body_json();
assert!(
namespace_child_tool(
&initial_request,
SEARCH_CALENDAR_NAMESPACE,
SEARCH_CALENDAR_CREATE_TOOL,
)
.is_some(),
"Calendar should be available before the MCP refresh: {initial_request}"
);
tokio::fs::remove_dir_all(test.codex_home_path().join("cache/codex_apps_tools")).await?;
startup_control.fail_next_initialize_attempts(/*attempts*/ 1);
let runtime_mcp_config = test.codex.runtime_mcp_config(&test.config).await;
let refresh_config = McpServerRefreshConfig {
mcp_servers: serde_json::to_value(codex_mcp::configured_mcp_servers(&runtime_mcp_config))?,
mcp_oauth_credentials_store_mode: serde_json::to_value(
runtime_mcp_config.mcp_oauth_credentials_store_mode,
)?,
auth_keyring_backend_kind: serde_json::to_value(
runtime_mcp_config.auth_keyring_backend_kind,
)?,
};
test.codex
.submit(Op::RefreshMcpServers {
config: refresh_config,
})
.await?;
test.submit_turn("use Calendar after transient Apps startup failures")
.await?;
tokio::time::timeout(Duration::from_secs(1), async {
while startup_control.initialize_attempts() < 3 {
tokio::time::sleep(Duration::from_millis(1)).await;
}
})
.await
.expect("background Apps reconnect should complete");
test.submit_turn("use Calendar after background Apps recovery")
.await?;
let requests = response.requests();
assert_eq!(requests.len(), 3);
let recovered_request = requests[2].body_json();
assert!(
namespace_child_tool(
&recovered_request,
SEARCH_CALENDAR_NAMESPACE,
SEARCH_CALENDAR_CREATE_TOOL,
)
.is_some(),
"Calendar should recover on the follow-up turn: {recovered_request}",
);
assert_eq!(startup_control.initialize_attempts(), 3);
Ok(())
}