MariaDB (RDS) going temporarily offline (with memory drop)

amazon-rdsconnectivitymariadbmariadb-10.4memory

We have an issue with a database of ours on RDS running MariaDB 10.4.8 on a t3-large instance type.

For the past week or so, it's been sporadically inaccessible with no errors I can find. During the minute or two that it's down I can't run 'show process list' it just hangs. It fixes itself and is back working within a few minutes. When it comes back it looks like it has a lot of connections and queries to catch up on.

Changed params from default:

ft_min_word_len 1
innodb_buffer_pool_size {DBInstanceClassMemory*17/20}
innodb_ft_enable_stopword   1   
innodb_ft_min_token_size    1   
log_bin_trust_function_creators 1
max_statement_time  30
sql_mode    NO_ZERO_IN_DATE,NO_ZERO_DATE,ERROR_FOR_DIVISION_BY_ZERO,NO_AUTO_CREATE_USER,NO_ENGINE_SUBSTITUTION

Freeable Memory:

enter image description here

Swap space:

enter image description here

Performance Insights:

enter image description here

If there any way to find out what's causing it. Is it not enough free memory? What's a normal threshold to be running to?

Best Answer

Enable the slow query log to see what is going on: https://mariadb.com/kb/en/slow-query-log/

Running a fulltext index with a minimum token length of 1 also slows things down, if not necessary: https://mariadb.com/kb/en/innodb-system-variables/#innodb_ft_min_token_size