My situation looks very similar to this Percona issue, resolved 2019: https://jira.percona.com/browse/PXC-2127 .
I have a 3-node cluster, stop a node gracefully:
- mysqld and mysqld_safe do not die.
- mysql.pid remains
- mysql-error.log shows shutdown is in progress, lots of Applier threads are killed.
- grastate.dat looks OK, bootstrap: 0, reasonable seqno.
- login attempts fail
- kill -TERM <mysql_pid> doesn't have any effect
- kill -9 <mysql_pid> does !