Reproduction steps:
Turn off systemd watchdog with override

master -> maxscale (blr) -> replica

Create large dataset (suggest >1TB, 100000s tables and 1000s schema)

Snapshot

Slap it for 5 GiB of binlogs with maxscale binlog keeping logs

create replica using the snapshot and change master to mxs, and start it.

You know you're on to a winner when maxctrl times out.

Then run systemctl stop maxscale.

There's a 1/3 chance it'll leave rpl_state blank. Or sometimes all blr service rpl_states blank, even those without replicas attached.

systemctl stop maxscale

-rw-r--r-- 1 maxscale maxscale 0 Sep 24 12:33 rpl_state