Swap usage won't decrease
We have a 5 DB nodes Galera Cluster in Production (MariaDB 10.0).
About a week ago a DB import was performed over an existing DB (Its size 30G) and during this process, all DB nodes started complaining about swap usage and haven't stopped complaining since then (e.g: One of them would be using 814 MB out of 4095 MB).
I have read a good few articles about MySQL and swapping issues (this is a good one: https://blog.jcole.us/2010/09/28/mysql-swap-insanity-and-the-numa-architecture/) and also about swapping in MariaDB (https://mariadb.com/kb/en/mariadb/configuring-swappiness/), but still can't find out a way of solving this problem without restarting MySQL.
We have NUMA architecture with 2 nodes and 96G of RAM. Our buffer pool size is 70G and it is set to use 12 innodb_buffer_pool_instances.
1) Is there any fix to this situation besides restarting MySQL?
2) Is there a way of tuning our cluster to avoid swapping in the future?
Answer Answered by Alexandra Arreaza in this comment.
It isn't rarely swapping but constantly since then, even though the system load has decreased.
It seems like adding a few parameters can improve this behaviour (I will have to test it first):
- innodb_buffer_pool_populate (under [mysqld] section) - flush_caches (under [mysqld_safe] section) - numa_interleave (under [mysqld_safe] section)
I don't think adding more RAM represents a workaround if the memory utilisation isn't balanced within a NUMA architecture.