BF abort for MDL conflict stuck when Thread Pool used

A whole cluster gets effectively blocked for writes when a DDL query fails to successfully trigger brute force abort. And this is happening when

thread_handling = pool-of-threads

is used.
So, as a result, we can observe long waiting for MDL lock, which should never happen in PXC/Galera due to priority nature of DDL handling, like:

           Id: 9
         User: msandbox
         Host: localhost
           db: test
      Command: Query
         Time: 332
        State: Waiting for table metadata lock
         Info: truncate table t1

An example debug level log from affected node in attachment.

How to reproduce

Use PXC node member with

thread_handling = pool-of-threads

.
In session one start a transaction on a simple table:

node1 [localhost:26529] {msandbox} (test) > begin; select * from t1;
Query OK, 0 rows affected (0.00 sec)

Empty set (0.00 sec)

In session 2 on the same node, try a DDL on the same table, like:

node1 [localhost:26529] {msandbox} (test) > truncate table t1;

Confirmed on PXC 5.7.26, .27 and .28.