diff --git a/src/Coordination/NuKeeperStorageDispatcher.cpp b/src/Coordination/NuKeeperStorageDispatcher.cpp
index 914985ee534..8ca5d3fff13 100644
--- a/src/Coordination/NuKeeperStorageDispatcher.cpp
+++ b/src/Coordination/NuKeeperStorageDispatcher.cpp
@@ -274,7 +274,13 @@ void NuKeeperStorageDispatcher::sessionCleanerTask()
                     LOG_INFO(log, "Found dead session {}, will try to close it", dead_session);
                     Coordination::ZooKeeperRequestPtr request = Coordination::ZooKeeperRequestFactory::instance().get(Coordination::OpNum::Close);
                     request->xid = Coordination::CLOSE_XID;
-                    putRequest(request, dead_session);
+                    NuKeeperStorage::RequestForSession request_info;
+                    request_info.request = request;
+                    request_info.session_id = dead_session;
+                    {
+                        std::lock_guard lock(push_request_mutex);
+                        requests_queue.push(std::move(request_info));
+                    }
                     finishSession(dead_session);
                 }
             }
diff --git a/tests/integration/test_testkeeper_multinode/test.py b/tests/integration/test_testkeeper_multinode/test.py
index 05879613ba6..51f60df7719 100644
--- a/tests/integration/test_testkeeper_multinode/test.py
+++ b/tests/integration/test_testkeeper_multinode/test.py
@@ -27,23 +27,8 @@ def started_cluster():
 def smaller_exception(ex):
     return '\n'.join(str(ex).split('\n')[0:2])
 
-def test_simple_replicated_table(started_cluster):
-
-    for i, node in enumerate([node1, node2, node3]):
-        node.query("CREATE TABLE t (value UInt64) ENGINE = ReplicatedMergeTree('/clickhouse/t', '{}') ORDER BY tuple()".format(i + 1))
-
-    node2.query("INSERT INTO t SELECT number FROM numbers(10)")
-
-    node1.query("SYSTEM SYNC REPLICA t", timeout=10)
-    node3.query("SYSTEM SYNC REPLICA t", timeout=10)
-
-    assert node1.query("SELECT COUNT() FROM t") == "10\n"
-    assert node2.query("SELECT COUNT() FROM t") == "10\n"
-    assert node3.query("SELECT COUNT() FROM t") == "10\n"
-
-
-def get_fake_zk(nodename):
-    _fake_zk_instance = KazooClient(hosts=cluster.get_instance_ip(nodename) + ":9181", timeout=30.0)
+def get_fake_zk(nodename, timeout=30.0):
+    _fake_zk_instance = KazooClient(hosts=cluster.get_instance_ip(nodename) + ":9181", timeout=timeout)
     def reset_last_zxid_listener(state):
         print("Fake zk callback called for state", state)
         _fake_zk_instance.last_zxid = 0
@@ -146,6 +131,56 @@ def test_watch_on_follower(started_cluster):
             pass
 
 
+def test_session_expiration(started_cluster):
+    try:
+        node1_zk = get_fake_zk("node1")
+        node2_zk = get_fake_zk("node2")
+        node3_zk = get_fake_zk("node3", timeout=3.0)
+
+        node3_zk.create("/test_ephemeral_node", b"world", ephemeral=True)
+
+        with PartitionManager() as pm:
+            pm.partition_instances(node3, node2)
+            pm.partition_instances(node3, node1)
+            node3_zk.stop()
+            node3_zk.close()
+            time.sleep(5)
+
+        assert node1_zk.exists("/test_ephemeral_node") is None
+        assert node2_zk.exists("/test_ephemeral_node") is None
+
+    finally:
+        try:
+            for zk_conn in [node1_zk, node2_zk, node3_zk]:
+                try:
+                    zk_conn.stop()
+                    zk_conn.close()
+                except:
+                    pass
+        except:
+            pass
+
+def test_simple_replicated_table(started_cluster):
+    # something may be wrong after partition in other tests
+    # so create with retry
+    for i, node in enumerate([node1, node2, node3]):
+        for i in range(100):
+            try:
+                node.query("CREATE TABLE t (value UInt64) ENGINE = ReplicatedMergeTree('/clickhouse/t', '{}') ORDER BY tuple()".format(i + 1))
+                break
+            except:
+                time.sleep(0.1)
+
+    node2.query("INSERT INTO t SELECT number FROM numbers(10)")
+
+    node1.query("SYSTEM SYNC REPLICA t", timeout=10)
+    node3.query("SYSTEM SYNC REPLICA t", timeout=10)
+
+    assert node1.query("SELECT COUNT() FROM t") == "10\n"
+    assert node2.query("SELECT COUNT() FROM t") == "10\n"
+    assert node3.query("SELECT COUNT() FROM t") == "10\n"
+
+
 # in extremely rare case it can take more than 5 minutes in debug build with sanitizer
 @pytest.mark.timeout(600)
 def test_blocade_leader(started_cluster):