Valerii Kravchuk added a comment - 2019-01-19 14:17 Why "Dix Version/s" say 10.3.12? For me XA RECOVER etc work on 10.3.7 as well: MariaDB [test]> select * from trocks; +----+------+ | id | c1 | +----+------+ | 1 | 1 | +----+------+ 1 row in set (0.008 sec)   MariaDB [test]> show create table trocks\G *************************** 1. row *************************** Table: trocks Create Table: CREATE TABLE `trocks` ( `id` int(11) NOT NULL, `c1` int(11) DEFAULT NULL, PRIMARY KEY (`id`) ) ENGINE=ROCKSDB DEFAULT CHARSET=latin1 1 row in set (0.001 sec)   MariaDB [test]> xa start 'xa1'; Query OK, 0 rows affected (0.001 sec)   MariaDB [test]> insert into trocks values (2,2); Query OK, 1 row affected (0.002 sec)   MariaDB [test]> xa end 'xa1'; Query OK, 0 rows affected (0.001 sec)   MariaDB [test]> xa prepare 'xa1'; Query OK, 0 rows affected (0.033 sec)   MariaDB [test]> select * from trocks; ERROR 2006 (HY000): MySQL server has gone away No connection. Trying to reconnect... ERROR 2002 (HY000): Can't connect to MySQL server on 'localhost' (10061) ERROR: Can't connect to the server I killed the server, now upon restart we see in the log: ... 2019-01-19 16:08:48 0 [Note] InnoDB: 10.3.7 started; log sequence number 6117865 3; transaction id 39687 2019-01-19 16:08:48 0 [Note] InnoDB: Loading buffer pool(s) from C:\Program File s\MariaDB 10.3\data\ib_buffer_pool 2019-01-19 16:08:48 0 [Note] Plugin 'FEEDBACK' is disabled. 2019-01-19 16:08:48 0 [Note] Recovering after a crash using pc-PC-bin 2019-01-19 16:08:48 0 [Note] Starting crash recovery... 2019-01-19 16:08:48 0 [Note] Found 1 prepared transaction(s) in ROCKSDB 2019-01-19 16:08:48 0 [Warning] Found 1 prepared XA transactions 2019-01-19 16:08:48 0 [Note] Crash recovery finished. 2019-01-19 16:08:48 0 [Note] Found 1 prepared transaction(s) in ROCKSDB 2019-01-19 16:08:49 0 [Warning] Found 1 prepared XA transactions 2019-01-19 16:08:49 0 [Note] Server socket created on IP: '::'. 2019-01-19 16:08:49 0 [Note] Reading of all Master_info entries succeded 2019-01-19 16:08:49 0 [Note] Added new Master_info '' to hash table 2019-01-19 16:08:49 0 [Note] mysqld: ready for connections. Version: '10.3.7-MariaDB-log' socket: '' port: 3316 mariadb.org binary distribution 2019-01-19 16:08:53 0 [Note] InnoDB: Buffer pool(s) load completed at 190119 16:08:53 Transaction is there in prepared state: MariaDB [test]> \r Connection id: 10 Current database: test   MariaDB [test]> select * from trocks; +----+------+ | id | c1 | +----+------+ | 1 | 1 | +----+------+ 1 row in set (0.003 sec)   MariaDB [test]> xa recover; +----------+--------------+--------------+------+ | formatID | gtrid_length | bqual_length | data | +----------+--------------+--------------+------+ | 1 | 3 | 0 | xa1 | +----------+--------------+--------------+------+ 1 row in set (0.001 sec)   MariaDB [test]> insert into trocks values (2,2); ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction MariaDB [test]> xa rollback 'xa1'; Query OK, 0 rows affected (0.045 sec)   MariaDB [test]> xa recover; Empty set (0.001 sec)   MariaDB [test]> select * from trocks; +----+------+ | id | c1 | +----+------+ | 1 | 1 | +----+------+ 1 row in set (0.003 sec)

Sergei Petrunia added a comment - 2019-01-20 19:07 I have set fixVersion to the version that I have pushed the testcase into. I'm not sure which specific revision fixed it. valerii do you think it is worth the time to find that out? (btw, is there any particular reason you are using different testcase from the one in the bug report? I suppose either should be fine for checking if XA is working, or there's some difference?)

Valerii Kravchuk added a comment - 2019-01-20 19:32 Current setting may be interpreted as the problem was fixed only in 10.3.12 and in all previous versions your test case fails. I think checking it with all 10.3 GA versions at least (and maybe recent 10.2.x) makes sense. My test is from my work on other cases (people stated there is a difference how MyRocks works vs InnoDB when mysqld crashes), then I was looking for known XA bugs related to MyRocks, found this one and decided to add that it must already work in 10.3.7 based on what I've seen).