Problem with pc.recovery

MySQL & MariaDB Percona XtraDB Cluster 8.x
1

I have a mysql8.4.10 cluster of 3 nodes A/B/C with default config pc.wait_prim = true; pc.wait_prim_timeout = PT30S; pc.wait_restored_prim_timeout = PT0S; pc.recovery=true.

When all nodes power off at almost the same time, 3 nodes will auto choose a bootstrap when they start up later and everything will be ok.

But if the down time be not the same, it will not process properly.

When node A power down, it will have 3 members in gvwstate.dat

After a few seconds, gvwstate.dat in node B/C will have only 2 members。

Then node B/C also power down.

Then start 3 nodes at the same time without a bootstrap node.

All nodes will become non-primary and mysql port will not be up and mysqld will never exit.

All mysqld will not process any signal. Script mysql.server can not stop/restart the running mysqld.

I can not set one node to bootstrap since the mysql listen port is not up.

I have to kill one mysqld and start it with bootstrap, then the cluster will be ok.

How can I fix this problem? The power down time will be random, I can not expect all gvwstate.dat to be the same. My cluster size is fixed , when all 3 nodes are up, the bootstrap node can be decided. Why the difference in gvwstate.dat will stop the cluster from auto choosing a bootstrap?

Can I change some params to fix this problem?

2

Could you clarify what you mean by “power off”? Are you shutting down the operating system cleanly, or are you abruptly powering off the VM or physical server without allowing the OS to shut down?

Also, could you share the contents of the grastate.dat file from all three nodes after they have been shut down? This will help us understand the recovered state and determine whether any of the nodes are marked as safe to bootstrap.