- Tune MySQL config settings to be more clouds-specific (reads much
more common, reads:writes rate is 10(100):1)
* Limit innodb_buffer_pool_size to 10G and set is as 30% of RAM
* Update innodb_log_file_size as:
* If mysql ib_logfiles do not match 25% of innodb_buffer_pool_size (max 2G),
delete them and update innodb_log_file_size in the config appropriately;
* Otherwise, use existing ib_logfile0 size to re-evaluate the
innodb_buffer_pool_size config value to match its real size.
* Reduce innodb_write_io_threads to 2/2 from 2/8
* Use innodb_flush_log_at_trx_commit=2 for HA (was 1)
* Raise FD imits to 102400/102400 from 22058/22058,
* Raise max connections to 3000/4096 from 2048/2048,
* Raise open tables limits 10000/10000 from 10000/400.
* Lower wait_timeout to 3600/3600 (1h) from 28800/28800 (8h)
* Adjust myisam_sort_buffer_size to 64M from 8388608 bytes
* Adjust key_buffer_size to 64M from 16777216 bytes
- Sync Puppet galera module params for MySQL module as well
Note: doubled values should be treated as simple/HA case
CI retest
Partial blueprint: optimize-db-performance-at-scale
Change-Id: Ie341da55711e70e38116389a54bda3b4357dafe7
Signed-off-by: Bogdan Dobrelya <bdobrelia@mirantis.com>
24c87cd7bb