Fixing MySQL/MariaDB Replication

Fixing MySQL/MariaDB Replication#

Introduction#

Multi-node Squirro deployments come with a Squirro Cluster service which takes care of electing a master node and making sure that the elected master node has the latest state of persisted data from the previous master in the event of the loss of the master node.

If the Squirro Cluster service fails to do so for mysql during a failover, please follow the operations guide below to fix the broken replication manually.

Replication Repair Steps#

The guide assumes that you have SSH access to the Squirro servers and the password to access MariaDB.

  1. Identify the current master

  2. For each broken replica, stop the Squirro cluster service:

    $ sudo systemctl stop sqclusterd

  3. Ensure you have a running system with one cluster node.

  4. At the master:

    mysql> RESET MASTER;
Query OK, 0 rows affected (0.14 sec)

mysql> FLUSH TABLES WITH READ LOCK;
Query OK, 0 rows affected (0.00 sec)

mysql> SHOW MASTER STATUS;
+---------------------+-------------+-----------------+---------------------+
| File                | Position    | Binlog_Do_DB    | Binlog_Ignore_DB    |
+---------------------+-------------+-----------------+---------------------+
| mysql-bin.000001    |     12268    |                 |                     |
+---------------------+-------------+-----------------+---------------------+
1 row in set (0.00 sec)
mysql> exit

$ mkdir /var/lib/squirro/cluster/mysql/<date>
$ mysqldump --host=127.0.0.1 --port=3306 --user=cluster --password=$(PASSWORD) --all-databases --master-data --single-transaction --result-file /var/lib/squirro/cluster/mysql/<date>/dump.db

$ mysql -u root -p
mysql> UNLOCK TABLES;
mysql> exit

$ cd /var/lib/squirro/cluster/mysql/<date>/
$ gzip dump.db
$ scp dump.db.gz $(SSH_USER)@$(IP_SLAVE1):/tmp/
$ scp dump.db.gz $(SSH_USER)@$(IP_SLAVE2):/tmp/

  5. On each replica:

    $ mkdir /var/lib/squirro/cluster/mysql/restore-<date>
$ mv /tmp/dump.db.gz /var/lib/squirro/cluster/mysql/restore-<date>
$ cd /var/lib/squirro/cluster/mysql/restore-<date>
$ gunzip dump.db.gz

$ mysql -u root -p
mysql> stop slave;
mysql> exit

$ mysql --host=127.0.0.1 --port=3306 --user=cluster --password=$(PASSWORD) -e "source dump.db;"

$ mysql -u root -p
mysql> RESET SLAVE;
mysql> change master to master_user='repl', master_password='PASSWORD';
mysql> show slave status \G    --> if master_host does not point to current master, e.g.: mysql> change master to master_host = "$(IP_MASTER)";
mysql> CHANGE MASTER TO MASTER_LOG_FILE='mysql-bin.000001', MASTER_LOG_POS=12268;            --> ensure that the position matches the value from SHOW MASTER STATUS; above
mysql> START SLAVE;
mysql> show slave status \G      --> ensure that Slave_IO_Running: Yes  and   Slave_SQL_Running: Yes

$ sudo systemctl start sqclusterd
$ squirro_status
Contents