Hi,

In our production env, Galera transaction fragment is used for running batch job. In some incidents (tmp directory full , VM reboot during hardware memory issue) , we encountered below 3 problems.

Problem #1: SST triggered to recover failed node but IST is expected
Problem #2: in some test, failed node encounters crash with signal 11 repeatedly until node 1 commit
Problem #3: local node state of donor node changed to "Donor/Desynced" unexpectedly after failed recovered

Workaround is manual restart node. But Galera should resume automatically on its own when hardware issue and running IST in most cases.

Repeatable testcase (galera-donor-desync.txt) is attached