Propagate Apps tool refreshes to existing threads (#43900)

## Why

Refreshing installed Apps without a thread should update the tools available to existing threads on their next turn.

## What changed

- Publish live tool catalogs to clients with matching transport, auth, protocol, and listing settings within the same account and home directory. Ignore the thread attribution header when matching scopes.
- Adopt updated tools before catalog reads and new calls, preserving running calls and rejecting calls prepared against an outdated catalog revision.
- Keep the newest successful fetch per scope so older refreshes cannot overwrite newer tools, and exclude disk snapshots from live updates.

## Testing

Add regression coverage for refresh propagation to an existing thread without another tools listing, scope and account isolation, out-of-order refreshes, late client startup, and running versus stale prepared calls.

GitOrigin-RevId: 7a5ee34e23742ce374c6647dc8928b76ea622448
This commit is contained in:
Matthew Zeng
2026-09-08 19:28:15 +00:00
committed by copyberry
parent f31bd3adff
commit 6d377e96eb
11 changed files with 370 additions and 22 deletions

View File

@@ -4,6 +4,8 @@
//! workspace. Disk is best-effort cold-start persistence; a context reads it
//! once when created and never rereads it. Full connector metadata is
//! owned by the connector metadata store, not by this module.
//! Live catalog subscriptions publish only successful fetches from a matching
//! discovery scope, never disk snapshots or another scope's discovery winner.
use std::collections::HashMap;
use std::path::Path;
@@ -22,6 +24,7 @@ use codex_protocol::mcp::McpServerInfo;
use serde::Deserialize;
use serde::Serialize;
use serde::de::DeserializeOwned;
use tokio::sync::watch;
use self::persistence::load_cached_codex_apps_server_info;
use self::persistence::load_cached_connector_runtime_for_identity;
@@ -31,6 +34,8 @@ use self::persistence::tools_cache_path;
const MCP_TOOLS_CACHE_PUBLISH_DURATION_METRIC: &str = "codex.mcp.tools.cache_publish.duration_ms";
type LiveCatalog<T> = watch::Sender<Option<Arc<ConnectorRuntimeSnapshot<T>>>>;
/// Values stored in the connector runtime's persisted tool snapshot.
///
/// The runtime uses the connector-owned Codex Apps cache layout for every
@@ -93,6 +98,7 @@ pub fn connector_runtime_cache_path(codex_home: &Path, auth: Option<&CodexAuth>)
pub struct ConnectorRuntimeSnapshot<T> {
tools: Vec<T>,
refreshed_at: SystemTime,
generation: u64,
}
impl<T> ConnectorRuntimeSnapshot<T> {
@@ -166,24 +172,47 @@ impl<T: ConnectorRuntimePayload> ConnectorRuntimeManager<T> {
.entry(identity.clone())
.or_insert_with(|| Arc::new(ConnectorRuntimeEntry::new(identity, self.disk_cache)))
.clone();
ConnectorRuntimeContext { entry }
ConnectorRuntimeContext {
entry,
live_catalog: None,
}
}
}
/// Handle to one shared account/workspace connector runtime.
pub struct ConnectorRuntimeContext<T: ConnectorRuntimePayload> {
entry: Arc<ConnectorRuntimeEntry<T>>,
live_catalog: Option<LiveCatalog<T>>,
}
impl<T: ConnectorRuntimePayload> Clone for ConnectorRuntimeContext<T> {
fn clone(&self) -> Self {
Self {
entry: Arc::clone(&self.entry),
live_catalog: self.live_catalog.clone(),
}
}
}
impl<T: ConnectorRuntimePayload> ConnectorRuntimeContext<T> {
/// Groups executable catalogs by equivalent discovery inputs within this account/home.
/// The caller must include the endpoint, auth configuration and listing protocol in `scope`.
/// Account-wide discovery reads are unaffected.
pub fn with_live_scope(mut self, scope: String) -> Self {
self.live_catalog = Some(
lock_unpoisoned(&self.entry.live_catalogs)
.entry(scope)
.or_insert_with(|| watch::channel(None).0)
.clone(),
);
self
}
/// Subscribes to accepted live tools without refetching or waiting for other clients.
pub fn subscribe(&self) -> Option<watch::Receiver<Option<Arc<ConnectorRuntimeSnapshot<T>>>>> {
self.live_catalog.as_ref().map(watch::Sender::subscribe)
}
pub fn current_snapshot(&self) -> Option<Arc<ConnectorRuntimeSnapshot<T>>> {
self.entry.current_snapshot.load_full()
}
@@ -254,6 +283,23 @@ impl<T: ConnectorRuntimePayload> ConnectorRuntimeContext<T> {
) -> Arc<ConnectorRuntimeSnapshot<T>> {
let publish_start = Instant::now();
let mut last_accepted_generation = lock_unpoisoned(&self.entry.last_accepted_generation);
let snapshot = Arc::new(ConnectorRuntimeSnapshot {
tools,
refreshed_at: SystemTime::now(),
generation: ticket.generation,
});
if let Some(live_catalog) = &self.live_catalog {
live_catalog.send_if_modified(|current| {
if current
.as_ref()
.is_some_and(|current| current.generation >= ticket.generation)
{
return false;
}
*current = Some(Arc::clone(&snapshot));
true
});
}
if ticket.generation <= *last_accepted_generation
&& let Some(snapshot) = self.current_snapshot()
{
@@ -266,11 +312,6 @@ impl<T: ConnectorRuntimePayload> ConnectorRuntimeContext<T> {
return snapshot;
}
let snapshot = Arc::new(ConnectorRuntimeSnapshot {
tools,
refreshed_at: SystemTime::now(),
});
*last_accepted_generation = ticket.generation;
self.entry
.current_snapshot
@@ -326,6 +367,7 @@ struct ConnectorRuntimeEntry<T: ConnectorRuntimePayload> {
current_snapshot: ArcSwapOption<ConnectorRuntimeSnapshot<T>>,
next_fetch_generation: AtomicU64,
last_accepted_generation: Mutex<u64>,
live_catalogs: Mutex<HashMap<String, LiveCatalog<T>>>,
}
impl<T: ConnectorRuntimePayload> ConnectorRuntimeEntry<T> {
@@ -342,6 +384,7 @@ impl<T: ConnectorRuntimePayload> ConnectorRuntimeEntry<T> {
current_snapshot: ArcSwapOption::from(current_snapshot),
next_fetch_generation: AtomicU64::new(0),
last_accepted_generation: Mutex::new(0),
live_catalogs: Mutex::new(HashMap::new()),
}
}
}

View File

@@ -64,6 +64,7 @@ pub(crate) fn load_cached_connector_runtime_for_identity<T: ConnectorRuntimePayl
ConnectorRuntimeSnapshot {
tools: cache.tools,
refreshed_at: modified_at,
generation: 0,
},
)
}
@@ -233,6 +234,7 @@ pub(crate) fn write_cached_codex_apps_tools_for_test<T>(
let snapshot = ConnectorRuntimeSnapshot {
tools: tools.to_vec(),
refreshed_at: SystemTime::now(),
generation: 0,
};
cache_context
.entry
@@ -263,6 +265,7 @@ where
let snapshot = ConnectorRuntimeSnapshot {
tools: tools.to_vec(),
refreshed_at: SystemTime::now(),
generation: 0,
};
write_cached_connector_runtime(cache_context, &snapshot)
}

View File

@@ -492,6 +492,47 @@ fn codex_apps_tools_cache_scopes_non_utf8_home_disk_paths() {
assert_ne!(cache_paths[0], cache_paths[1]);
}
#[test]
fn live_catalogs_isolate_scopes_and_reject_older_refreshes() {
let codex_home = tempdir().expect("tempdir");
let manager = ConnectorRuntimeManager::<TestTool>::new_without_cache();
let context = manager.context(
codex_home.path().to_path_buf(),
ConnectorRuntimeContextKey::personal(
Some("account".to_string()),
/*chatgpt_user_id*/ None,
),
);
let scope_a = context.clone().with_live_scope("endpoint-a".to_string());
let scope_b = context.with_live_scope("endpoint-b".to_string());
let updates_a = scope_a.subscribe().expect("scope A subscription");
let updates_b = scope_b.subscribe().expect("scope B subscription");
let server_info = create_test_server_info("Codex Apps");
let older_a = scope_a.begin_fetch(ConnectorRuntimeFetchSource::Startup);
let newer_a = scope_a.begin_fetch(ConnectorRuntimeFetchSource::HardRefresh);
let newest_b = scope_b.begin_fetch(ConnectorRuntimeFetchSource::HardRefresh);
let tools_a = vec![create_test_tool(CODEX_APPS_MCP_SERVER_NAME, "newer-a")];
let tools_b = vec![create_test_tool(CODEX_APPS_MCP_SERVER_NAME, "newest-b")];
scope_b.publish_if_newest_accepted(newest_b, &server_info, tools_b.clone());
assert!(updates_a.borrow().is_none());
let discovery = scope_a.publish_if_newest_accepted(newer_a, &server_info, tools_a.clone());
assert_eq!(discovery, tools_b);
scope_a.publish_if_newest_accepted(
older_a,
&server_info,
vec![create_test_tool(CODEX_APPS_MCP_SERVER_NAME, "stale-a")],
);
assert_eq!(
updates_a.borrow().as_ref().expect("scope A tools").tools(),
&tools_a
);
assert_eq!(
updates_b.borrow().as_ref().expect("scope B tools").tools(),
&tools_b
);
}
#[test]
fn contexts_for_different_identities_keep_isolated_snapshots() {
let codex_home = tempdir().expect("tempdir");
@@ -504,6 +545,8 @@ fn contexts_for_different_identities_keep_isolated_snapshots() {
is_workspace_account: false,
},
);
let context_a = context_a.with_live_scope("same-endpoint".to_string());
let updates_a = context_a.subscribe().expect("account A subscription");
let tools_a = vec![create_test_tool(CODEX_APPS_MCP_SERVER_NAME, "tool-a")];
let snapshot_a = context_a.publish_runtime_if_newest_accepted(
context_a.begin_fetch(ConnectorRuntimeFetchSource::HardRefresh),
@@ -519,6 +562,8 @@ fn contexts_for_different_identities_keep_isolated_snapshots() {
is_workspace_account: false,
},
);
let context_b = context_b.with_live_scope("same-endpoint".to_string());
let updates_b = context_b.subscribe().expect("account B subscription");
let same_context_a = manager.context(
codex_home.path().to_path_buf(),
ConnectorRuntimeContextKey {
@@ -527,6 +572,7 @@ fn contexts_for_different_identities_keep_isolated_snapshots() {
is_workspace_account: false,
},
);
let same_context_a = same_context_a.with_live_scope("same-endpoint".to_string());
assert!(Arc::ptr_eq(
&snapshot_a,
@@ -566,6 +612,22 @@ fn contexts_for_different_identities_keep_isolated_snapshots() {
&snapshot_b,
&context_b.current_snapshot().expect("context B snapshot")
));
assert_eq!(
updates_a
.borrow()
.as_ref()
.expect("account A tools")
.tools(),
&newer_tools_a
);
assert_eq!(
updates_b
.borrow()
.as_ref()
.expect("account B tools")
.tools(),
&tools_b
);
}
#[test]