From d2175d1e1e33ab7fb9bdf6b49e78dfa27ee353c7 Mon Sep 17 00:00:00 2001
From: Chris <anzhengchao@gmail.com>
Date: Tue, 29 Sep 2026 14:07:30 +0800
Subject: [PATCH] fix(test): keep bucket disk faults across reconnects (#8224)

---
 crates/ecstore/src/store/bucket.rs | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)

diff --git a/crates/ecstore/src/store/bucket.rs b/crates/ecstore/src/store/bucket.rs
index b4b961f9b..9debdafbc 100644
--- a/crates/ecstore/src/store/bucket.rs
+++ b/crates/ecstore/src/store/bucket.rs
@@ -1569,12 +1569,27 @@ mod tests {
         disk_indexes: &[usize],
     ) -> Vec<(usize, crate::disk::DiskStore)> {
         let offline = {
-            let mut disks = set.disks.write().await;
+            let disks = set.disks.read().await;
             disk_indexes
                 .iter()
-                .map(|index| (*index, disks[*index].take().expect("fault-injection disk should start online")))
+                .map(|index| (*index, disks[*index].clone().expect("fault-injection disk should start online")))
                 .collect::<Vec<_>>()
         };
+        for (_, disk) in &offline {
+            disk.close().await.expect("fault injection should stop per-disk monitoring");
+            disk.force_offline_for_test();
+        }
+        // Empty slots can be renewed from the still-present disk paths by the endpoint monitor.
+        // Keep the original handles and prove reconnect cannot clear their injected IO failure.
+        set.connect_disks().await;
+        {
+            let disks = set.disks.read().await;
+            for (index, disk) in &offline {
+                let current = disks[*index].as_ref().expect("fault-injection slot should remain populated");
+                assert!(Arc::ptr_eq(current, disk), "reconnect must preserve the fault-injection handle");
+                assert_eq!(disk.runtime_state(), crate::disk::health_state::RuntimeDriveHealthState::Offline);
+            }
+        }
         let local_disk_map = ecstore.ctx.local_disk_map();
         let mut local_disks = local_disk_map.write().await;
         for (_, disk) in &offline {
@@ -1595,9 +1610,12 @@ mod tests {
                 local_disks.insert(disk.endpoint().to_string(), Some(Arc::clone(disk)));
             }
         }
-        let mut disks = set.disks.write().await;
+        let disks = set.disks.read().await;
         for (index, disk) in offline {
-            assert!(disks[index].replace(disk).is_none(), "fault-injection disk slot should remain empty");
+            let current = disks[index].as_ref().expect("fault-injection slot should remain populated");
+            assert!(Arc::ptr_eq(current, &disk), "fault-injection disk handle should remain unchanged");
+            assert_eq!(disk.runtime_state(), crate::disk::health_state::RuntimeDriveHealthState::Offline);
+            disk.reset_health_for_store_init_retry();
         }
     }