Sequence object next_not_cached_value not replicated in Galera cluster.

We have 3 nodes Galera clusters with 2 db nodes and a arbitrator such that

arbitrator: version: 4-26.4.22
DB node 1: version 10.6.22
DB node 2: upgraded to 10.11.15

We are performaning upgrade the version from 10.6.22 to 10.11.15 using rolling upgrade in the intermediate stage so that node 2 is upgraded to 1011.15 and node 1 is still 10.6.22.

We have the follwoing siutuation :

• A sequence object is created with
  CREATE SEQUENCE test_seq CACHE 10;
• we have a table test_tbl with primary key id values using the test_seq:
  CREATE TABLE test_tbl (
  id BIGINT NOT NULL PRIMARY KEY,
  node_name varchar(10),
  timestamp datetime
  );
• we insert table test_tbl in both nodes currently by:
  INSERT INTO test_tbl (id, node_name, timestamp)
  VALUES (NEXT VALUE FOR test_seq, <node1/node2>, now());

• It has table result like:
  id node_name timestamp
  1 node1 10:00:01
  2 node1 10:00:02
  11 node2 10:00:03
  12 node2 10:00:04
  21 node1 10:00:05
  22 node1 10:00:06
  31 node2 10:00:07
  32 node2 10:00:08
  41 node1 10:00:09
  42 node1 10:00:10
  33 node2 10:00:11
  34 node2 10:00:12
  43 node1 10:00:13
  44 node1 10:00:14
  35 node2 10:00:15
  36 node2 10:00:16
  37 node2 10:00:17
  38 node2 10:00:18
  45 node1 10:00:19
  39 node2 10:00:20
  40 node2 10:00:21
  <At 10:00:22, node 2 insert with id=41 which collided with the record at 10:00:09> (Duplicate key error)

Please note that from 10:00:09 onwards, node 1 kept generating the id 4X but node 2 still kept generate the id 3X (not 5X). Finally at 10:00:22, node2 tried to insert record with id=41 which collided the key at 10:00:09.

Read from the binary logs of both DB nodes, it is found that at 10:00:09 node1 updated the "next_not_cached_value " of test_seq but it wasn't replicated to node 2. However before 10:00:09, the update of next_not_cached_value was replicated to the other node successfully.

Also, during 10:00:09 to 10:00:21, we encountered deadlock error when inserting the records.