Dedicated Zabbix Sql Server restarts

[bep6vaku]

I have a dedicated Zabbix SQl running MariaDB ver.15.1 on an Ubuntu 24.04 and when the memory utilization approached 78 or better precent, the SQL service restart. Not sure what is causing this or what settings need to ne adjusted. This is a physical server with 380GB of ram. I looked at the InnoDB settings but not sure if any need to be adjusted: Current version of Zabbix is 7.4.8

innodb_buffer_pool_size = 285G
innodb_log_file_size = 16G
innodb_log_buffer_size = 256M
innodb-file-per-table = 1
innodb_buffer_pool_instances = 16
innodb_old_blocks_time = 1000
innodb_stats_on_metadata = off
innodb-flush-method = O_DIRECT
innodb-flush-log-at-trx-commit = 2
innodb_read_io_threads = 64
innodb_write_io_threads = 64
innodb_io_capacity = 2000
innodb_io_capacity_max = 15000
connect_timeout = 10
table_open_cache_instances = 32
table_open_cache = 16000

[tim.mooney]

Is there anything useful in the mariadb error log (probably in /var/log/mariadb )?

At the time when the unexpected crash happened, is there anything in any of the other system logs? Does 'journalctl' for the time in question have anything useful?

When a database service unexpectedly restarts, it shouldn't be "quiet". There should be traces in multiple logs that point to an underlying problem.

I would have to research a few of your settings to know more about what they do, but nothing jumps out as obviously problematic. Besides, a poor tuning setting shouldn't cause the database to crash. It should just perform poorly (potentially very poorly).

Are "innodb_buffer_pool_instances" still a thing with your version of MariaDB? Dividing the overall InnoDB buffer pool into chunks was very important at early 10.x versions of MariaDB, but I thought that the instances were deprecated at like 10.5 or 10.6, and shouldn't be set any more?