IT notes

galera SOS

Check status of cluster: Find process list not sleeping: SELECT user, time, state, info FROM information_schema.processlist WHERE command != 'Sleep' AND time >= 2 ORDER BY time DESC, id; Get locked tables: show open tables where in_use > 0; Finding locks: show engine innodb status; At the end: ---TRANSACTION 1135701157, ACTIVE 6768 sec MySQL thread id 5208136, OS thread handle 0x7f2982e91700, query id 882213399 xxxIPxxx my_user cleaning up kill the transaction thread:


To periodically check status of the cluster, create a script (/tmp/xx): #!/bin/sh mysql --defaults-file=/path/to/.my.cnf -s -e \ "SHOW GLOBAL STATUS WHERE Variable_name IN ('wsrep_ready', 'wsrep_cluster_size', 'wsrep_cluster_status', 'wsrep_connected', 'wsrep_local_state', 'wsrep_local_index');" And run it with: watch sh /tmp/xx If single node alive (ERROR 1047 WSREP has not yet prepared node for application use): SET GLOBAL wsrep_provider_options='pc.bootstrap=YES'; This node can be used now has the new master so others nodes can recover from it.

galera alter

To prevent bloking the cluster while doing ALTERS (SCHEMA UPGRADES), do this per node: SET wsrep_OSU_method='RSU'; Then Run the ALTER statement once done reset the Schema Upgrade method back to Total Order Isolation. SET wsrep_OSU_method='TOI';