MDEV-31239: Second-level slave hangs when master crashes between START and COMMIT ALTER
Test case only that demonstrates (one instance of) the problem.
During ALTER TABLE with binlog_alter_two_phase=ON, if the master crashes in
the middle, the binlog will contain a START ALTER event without a matching
COMMIT ALTER event. A third-level slave will thus wait forever in the SA
thread in the START ALTER, waiting for a signal from COMMIT ALTER that never arrives.
Signed-off-by: Kristian Nielsen <email address hidden>
MDEV-31207: Memory leak when START ALTER fails on slave
Fix the code so that the rgi->sa_info object is initially owned by the rgi,
and ownership later explicitly transferred to the start_alter_list. This way
the memory will be freed both when START ALTER is successful and when it
fails.
Signed-off-by: Kristian Nielsen <email address hidden>
MDEV-31140: FLUSH BINARY LOGS DELETE_DOMAIN_ID=(D) can errorneously delete active domains
Fix the code in rpl_binlog_state::drop_domain(), so that _all_ entries for
the domain in the binlog state must match an entry in the initial GTID_LIST,
not just one entry match.
Signed-off-by: Kristian Nielsen <email address hidden>
dh_missing: warning: usr/share/mysql/mysql-test/mysql-test-run exists in debian/tmp but is not installed to anywhere
dh_missing: warning: usr/share/mysql/mysql-test/mtr exists in debian/tmp but is not installed to anywhere
dh_missing: warning: usr/share/mysql/mysql-test/mariadb-test-run exists in debian/tmp but is not installed to anywhere
dh_missing: warning: usr/share/mysql/mysql-test/mysql-test-run.pl exists in debian/tmp but is not installed to anywhere
Add all to mariadb-test.install and remove mariadb-test.links
