From 051d5d989420747b0145b83392538f8b4e2d1d74 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=BC=A0=E4=B8=87=E4=B9=89?= <arvin.zhang@htx-inc.com>
Date: Wed, 19 Aug 2026 16:23:47 +0800
Subject: [PATCH] [server] Fix race condition in table/partition deletion
 ordering

Swap the submission order so that metadata deletion (ZK) is submitted
to the ioExecutor before remote directory cleanup. Since the ioExecutor
is a single-threaded pool, slow remote filesystem operations can block
the faster metadata deletion, causing flaky test timeouts in
CoordinatorEventProcessorTest.testCreateAndDropTable.

This closes #1026.
---
 .../org/apache/fluss/server/coordinator/TableManager.java     | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fluss-server/src/main/java/org/apache/fluss/server/coordinator/TableManager.java b/fluss-server/src/main/java/org/apache/fluss/server/coordinator/TableManager.java
index 2381447bcb8..c085cc42b06 100644
--- a/fluss-server/src/main/java/org/apache/fluss/server/coordinator/TableManager.java
+++ b/fluss-server/src/main/java/org/apache/fluss/server/coordinator/TableManager.java
@@ -272,8 +272,8 @@ private void resumePartitionDeletions() {
     private void completeDeleteTable(long tableId) {
         Set<TableBucketReplica> replicas = coordinatorContext.getAllReplicasForTable(tableId);
         replicaStateMachine.handleStateChanges(replicas, ReplicaState.NonExistentReplica);
-        asyncDeleteRemoteDirectory(tableId);
         asyncDeleteTableMetadata(tableId);
+        asyncDeleteRemoteDirectory(tableId);
         coordinatorContext.removeTable(tableId);
         // Release the manager's in-flight tracking so the next batch can be submitted.
         lifecycleThrottler.onTableDropCompleted(tableId);
@@ -284,8 +284,8 @@ private void completeDeletePartition(TablePartition tablePartition) {
                 coordinatorContext.getAllReplicasForPartition(
                         tablePartition.getTableId(), tablePartition.getPartitionId());
         replicaStateMachine.handleStateChanges(replicas, ReplicaState.NonExistentReplica);
-        asyncDeleteRemoteDirectory(tablePartition);
         asyncDeletePartitionMetadata(tablePartition.getPartitionId());
+        asyncDeleteRemoteDirectory(tablePartition);
         coordinatorContext.removePartition(tablePartition);
         lifecycleThrottler.onPartitionDropCompleted(tablePartition);
     }