Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

test(agent): harden async child failure test against flakiness

Steven Enamakelfa0aa92722fb
diff --git a/crates/openhuman-core/src/agent/orchestration/tools/tools_e2e_tests.rs b/crates/openhuman-core/src/agent/orchestration/tools/tools_e2e_tests.rs
index 5a6b664cd4..a09de2fe33 100644
--- a/crates/openhuman-core/src/agent/orchestration/tools/tools_e2e_tests.rs
+++ b/crates/openhuman-core/src/agent/orchestration/tools/tools_e2e_tests.rs
@@ -1,4 +1,4 @@
-use super::SpawnSubagentTool;
+use super::*;
use crate::agent::harness::ParentExecutionContext;
use crate::agent_registry::host_definition::AgentDefinitionRegistry;
use crate::integrations::ConnectedIntegration;
@@ -249,11 +249,19 @@ async fn spawn_subagent_reports_a_stopped_child_as_incomplete_progress() {
}
#[tokio::test]
+#[tokio::test(start_paused = true)]
async fn failed_async_child_is_reported_to_the_parent_without_a_success_result() {
let _ = AgentDefinitionRegistry::init_global_builtins();
let workspace = tempfile::TempDir::new().expect("workspace");
let (progress_tx, mut progress_rx) = tokio::sync::mpsc::channel(32);
- let mut parent = parent_context(workspace.path(), Arc::new(FailingModel), vec![]);
+ let invocations = Arc::new(std::sync::atomic::AtomicUsize::new(0));
+ let mut parent = parent_context(
+ workspace.path(),
+ Arc::new(FailingModel {
+ invocations: Arc::clone(&invocations),
+ }),
+ vec![],
+ );
parent.on_progress = Some(progress_tx);
let live = super::live_parent_run_context("failed-async-child", parent);
let accepted = SpawnSubagentTool::new()
@@ -275,7 +283,7 @@ async fn failed_async_child_is_reported_to_the_parent_without_a_success_result()
"child failure is delivered asynchronously"
);
- let failure = tokio::time::timeout(std::time::Duration::from_secs(5), async {
+ let failure_result = tokio::time::timeout(std::time::Duration::from_secs(120), async {
loop {
match progress_rx.recv().await {
Some(crate::progress::AgentProgress::SubagentFailed {
@@ -286,8 +294,12 @@ async fn failed_async_child_is_reported_to_the_parent_without_a_success_result()
}
}
})
- .await
- .expect("asynchronous child failure reaches the parent");
+ .await;
+ let invocation_count = invocations.load(std::sync::atomic::Ordering::SeqCst);
+ assert!(invocation_count > 0, "the injected provider must be called");
+ let failure = failure_result.unwrap_or_else(|elapsed| {
+ panic!("asynchronous child failure reaches the parent after {elapsed}; provider calls={invocation_count}")
+ });
assert!(
failure.starts_with("spawn_subagent failed and did not complete."),
"failure copy remains recognizable to the parent: {failure}"
@@ -354,7 +366,9 @@ async fn unexecuted_tool_call_stub_is_projected_as_incomplete_not_success() {
);
}
-struct FailingModel;
+struct FailingModel {
+ invocations: Arc<std::sync::atomic::AtomicUsize>,
+}
#[async_trait]
impl ChatModel<()> for FailingModel {
@@ -372,6 +386,8 @@ impl ChatModel<()> for FailingModel {
_state: &(),
_request: ModelRequest,
) -> tinyinference_llm::Result<ModelResponse> {
+ self.invocations
+ .fetch_add(1, std::sync::atomic::Ordering::SeqCst);
Err(tinyinference_llm::Error::Model(
"scripted provider outage".into(),
))