Julian Scarfe wrote:

> The problem is repeatable after an init cluster.
>
> What happened please?

I am finding the same behaviour.

Could you perchance do the query:

select * from "_T1".sl_event where ev_type in ('ACCEPT_SET', 'MOVE_SET');

And do that on both nodes?  That would be helpful...

I'm seeing the situation where there is a NULL sequence number in the
4th data field for the ACCEPT_SET event, and that is causing the
ACCEPT_SET event to fall over relatively silently when the
remote_worker.c code tries to translate NULL into a sequence number.

I'm particularly suspicious of a bug in the creation of the ACCEPT_SET
event, which is a couple steps back from where you're seeing the problem.