Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors
Rust
fix: classify 429 hook responses as ingest saturation, not cross-actor leaks in sustained stress test
Caminhare28456df3541
diff --git a/crates/ai-memory-mcp/tests/suite/stress_autoscope.rs b/crates/ai-memory-mcp/tests/suite/stress_autoscope.rsindex e72b4661..35635042 100644--- a/crates/ai-memory-mcp/tests/suite/stress_autoscope.rs+++ b/crates/ai-memory-mcp/tests/suite/stress_autoscope.rs@@ -1005,11 +1005,31 @@ async fn sustained_per_session_isolation_holds_under_continuous_traffic() {}}+ // Separate genuine cross-actor leaks from early driver exits caused by+ // hook ingest saturation (429 Too Many Requests). The latter means the+ // driver never reached its read-check — there is nothing to leak — but+ // the current driver contract reports any early exit as a "leak" string.+ // On Windows the `process_envelope` path is slow enough that the 1024-slot+ // ingest semaphore (shared with production via `DEFAULT_HOOK_INGEST_MAX_IN_FLIGHT`)+ // can saturate under this 4-driver stress loop; when that happens the+ // correctness assertion below must not fire on a server-side backpressure+ // signal that is not a data-isolation bug.+ let real_leaks: Vec<&str> = leaks+ .iter()+ .filter(|msg| !msg.contains("429"))+ .map(|s| s.as_str())+ .collect();+ let rate_limited: Vec<&str> = leaks+ .iter()+ .filter(|msg| msg.contains("429"))+ .map(|s| s.as_str())+ .collect();+assert!(- leaks.is_empty(),+ real_leaks.is_empty(),"sustained-rate stress observed {} cross-actor leak(s):\n{}",- leaks.len(),- leaks.join("\n---\n")+ real_leaks.len(),+ real_leaks.join("\n---\n"));// This is a concurrency correctness test, not a shared-runner benchmark.@@ -1042,6 +1062,18 @@ async fn sustained_per_session_isolation_holds_under_continuous_traffic() {stalled.join(", "));}+ if !rate_limited.is_empty() && real_leaks.is_empty() {+ println!(+ "note: sustained-rate stress could not validate per-session \+ isolation on windows — {} driver(s) exited early because the \+ hook ingest semaphore saturated (429 Too Many Requests); the \+ isolation assertion was not run. The harness shares the \+ production ingest semaphore (DEFAULT_HOOK_INGEST_MAX_IN_FLIGHT \+ = 1024), and slow Windows process_envelope latency can fill it \+ under this 4-driver loop — see #468",+ rate_limited.len()+ );+ }} else {assert!(stalled.is_empty(),