From c8f42b8dca0ef62d3c58301347d66fc4fe08a0f5 Mon Sep 17 00:00:00 2001
From: Chris <anzhengchao@gmail.com>
Date: Tue, 29 Sep 2026 07:16:17 +0800
Subject: [PATCH] fix(ci): isolate scanner deadline and expose test failure
 details (#8210)

* fix(ci): isolate scanner deadline fixture and expose readiness errors

* test(ecstore): report unexpected capacity reservation errors
---
 crates/e2e_test/src/common.rs       | 29 +++++++++++++++++++++++++++--
 crates/ecstore/src/core/pools.rs    |  5 ++++-
 crates/scanner/src/scanner/tests.rs | 17 +++++++++++++++--
 3 files changed, 46 insertions(+), 5 deletions(-)

diff --git a/crates/e2e_test/src/common.rs b/crates/e2e_test/src/common.rs
index ac27da8e0..698df2f76 100644
--- a/crates/e2e_test/src/common.rs
+++ b/crates/e2e_test/src/common.rs
@@ -22,6 +22,7 @@
 //! - Common test constants and utilities
 
 use aws_sdk_s3::config::{Credentials, Region};
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::{Client, Config};
 use aws_smithy_http_client::Builder as SmithyHttpClientBuilder;
 use http::header::{CONTENT_TYPE, HOST};
@@ -1577,6 +1578,7 @@ impl RustFSTestClusterEnvironment {
     /// retries up to 120 times with a 1-second interval between attempts.
     async fn wait_for_node_service_ready(&self, node_idx: usize) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         let client = self.create_s3_client(node_idx)?;
+        let mut last_error = None;
 
         for attempt in 0..120 {
             match client.list_buckets().send().await {
@@ -1584,13 +1586,36 @@ impl RustFSTestClusterEnvironment {
                     info!("Cluster node {} service ready after {} attempts", node_idx, attempt + 1);
                     return Ok(());
                 }
-                Err(_) => {
+                Err(err) => {
+                    last_error = Some(err);
                     sleep(Duration::from_secs(1)).await;
                 }
             }
         }
 
-        Err(format!("Cluster node {} service failed to become ready", node_idx).into())
+        let last_error = last_error.as_ref().map(|err| {
+            // SDK Display reports only the category. Do not expose raw response bodies or headers.
+            let service_code = err.as_service_error().and_then(|error| error.code()).filter(|code| {
+                matches!(
+                    *code,
+                    "ServiceUnavailable"
+                        | "ServerNotInitialized"
+                        | "InternalError"
+                        | "AccessDenied"
+                        | "InvalidAccessKeyId"
+                        | "SignatureDoesNotMatch"
+                )
+            });
+            format!(
+                "{err}; http_status={:?}; service_code={service_code:?}",
+                err.raw_response().map(|response| response.status().as_u16())
+            )
+        });
+        Err(format!(
+            "Cluster node {node_idx} service failed to become ready; last ListBuckets error={last_error:?}; capture_log_path={:?}",
+            self.node_capture_log_paths[node_idx]
+        )
+        .into())
     }
 
     /// Create an S3 client configured to communicate with a specific cluster node.
diff --git a/crates/ecstore/src/core/pools.rs b/crates/ecstore/src/core/pools.rs
index 70a6369ce..b40b520ff 100644
--- a/crates/ecstore/src/core/pools.rs
+++ b/crates/ecstore/src/core/pools.rs
@@ -19542,7 +19542,10 @@ mod tests {
             .await
             .expect("second-node activation should not panic")
             .expect_err("the second reservation must observe and reject the committed first reservation");
-        assert!(err.to_string().contains("requires 60 bytes, but 40 bytes are available"));
+        assert!(
+            err.to_string().contains("requires 60 bytes, but 40 bytes are available"),
+            "the second reservation must report the committed capacity rejection, got {err:?}"
+        );
 
         let mut persisted = PoolMeta::default();
         persisted
diff --git a/crates/scanner/src/scanner/tests.rs b/crates/scanner/src/scanner/tests.rs
index 62d987f32..255eebede 100644
--- a/crates/scanner/src/scanner/tests.rs
+++ b/crates/scanner/src/scanner/tests.rs
@@ -636,8 +636,8 @@ async fn cycle_budget_fence_accepts_bootstrap_pending_usage_marker() {
 
 #[tokio::test]
 async fn cycle_budget_deadline_handler_fences_and_releases_guard() {
-    let (_temp_dir, store) = setup_scanner_cycle_store().await;
-    let lock = store
+    let (_temp_dir, lock_store) = setup_scanner_cycle_store().await;
+    let lock = lock_store
         .new_ns_lock(RUSTFS_META_BUCKET, "leader.lock")
         .await
         .expect("scanner leader lock should be created");
@@ -646,6 +646,17 @@ async fn cycle_budget_deadline_handler_fences_and_releases_guard() {
         .await
         .expect("scanner leader lock should be acquired");
 
+    // Keep the real guard, but isolate the fencing deadline from filesystem I/O.
+    let store = Arc::new(MemoryConfigStore::default());
+    save_config(
+        store.clone(),
+        DATA_USAGE_OBJ_NAME_PATH.as_str(),
+        serde_json::to_vec(&complete_usage_with_bucket_count(Some(std::time::SystemTime::UNIX_EPOCH), 0))
+            .expect("scanner cycle usage baseline should encode"),
+    )
+    .await
+    .expect("scanner cycle usage baseline should persist");
+
     let ctx = CancellationToken::new();
     let mut cycle_info = CurrentCycle {
         current: 12,
@@ -679,6 +690,8 @@ async fn cycle_budget_deadline_handler_fences_and_releases_guard() {
     .await;
 
     assert!(guard.is_released());
+    assert_eq!(leader_epoch, 2, "deadline handler should claim the next epoch");
+    assert!(matches!(cycle_revision, DataUsageCacheRevision::Etag(_)));
     let persisted = read_config(store, &DATA_USAGE_BLOOM_NAME_PATH)
         .await
         .expect("deadline handler should persist a fenced cursor");