diff --git a/src/tests/simulation/tests/integration_tests/recovery/background_ddl.rs b/src/tests/simulation/tests/integration_tests/recovery/background_ddl.rs
index 91aea55a2dc31..97c08d098f6c9 100644
--- a/src/tests/simulation/tests/integration_tests/recovery/background_ddl.rs
+++ b/src/tests/simulation/tests/integration_tests/recovery/background_ddl.rs
@@ -248,10 +248,13 @@ async fn test_ddl_cancel() -> Result<()> {
     Ok(())
 }
 
-#[tokio::test]
-async fn test_high_barrier_latency_cancel() -> Result<()> {
+/// When cancelling a stream job under high latency,
+/// the cancel should take a long time to take effect.
+/// If we trigger a recovery however, the cancel should take effect immediately,
+/// since cancel will immediately drop the table fragment.
+async fn test_high_barrier_latency_cancel(config: Configuration) -> Result<()> {
     init_logger();
-    let mut cluster = Cluster::start(Configuration::for_scale()).await?;
+    let mut cluster = Cluster::start(config).await?;
     let mut session = cluster.start_session();
 
     // Join 2 fact tables together to create a high barrier latency scenario.
@@ -280,7 +283,7 @@ async fn test_high_barrier_latency_cancel() -> Result<()> {
         cluster
             .kill_nodes_and_restart(["compute-1", "compute-2", "compute-3"], 2)
             .await;
-        sleep(Duration::from_secs(3)).await;
+        sleep(Duration::from_secs(2)).await;
 
         tracing::debug!("killed cn, waiting recovery");
 
@@ -305,9 +308,26 @@ async fn test_high_barrier_latency_cancel() -> Result<()> {
 
     tracing::info!("restarted cn: trigger stream job recovery");
 
+    // Make sure there's some progress first.
+    loop {
+        // Wait until at least 10% of records are ingested.
+        let progress = session
+            .run("select progress from rw_catalog.rw_ddl_progress;")
+            .await
+            .unwrap();
+        tracing::info!(progress, "get progress before cancel stream job");
+        let progress = progress.replace('%', "");
+        let progress = progress.parse::<f64>().unwrap();
+        if progress > 0.01 {
+            break;
+        } else {
+            sleep(Duration::from_micros(1)).await;
+        }
+    }
     // Loop in case the cancel gets dropped after
     // cn kill, before it drops the table fragment.
-    for _ in 0..5 {
+    for iteration in 0..5 {
+        tracing::info!(iteration, "cancelling stream job");
         let mut session2 = cluster.start_session();
         let handle = tokio::spawn(async move {
             let result = cancel_stream_jobs(&mut session2).await;
@@ -343,6 +363,16 @@ async fn test_high_barrier_latency_cancel() -> Result<()> {
     Ok(())
 }
 
+#[tokio::test]
+async fn test_high_barrier_latency_cancel_for_arrangement_backfill() -> Result<()> {
+    test_high_barrier_latency_cancel(Configuration::for_arrangement_backfill()).await
+}
+
+#[tokio::test]
+async fn test_high_barrier_latency_cancel_for_no_shuffle() -> Result<()> {
+    test_high_barrier_latency_cancel(Configuration::for_scale_no_shuffle()).await
+}
+
 // When cluster stop, foreground ddl job must be cancelled.
 #[tokio::test]
 async fn test_foreground_ddl_no_recovery() -> Result<()> {